Asset Analytics

Performance and Safety Management

Series Editors: Ajit Kumar Verma · P. K. Kapur · Uday Kumar

B. Vipin
C. Rajendran
Ganesh Janakiraman
Deepu Philip Editors

Emerging
Frontiers in
Operations and
Supply Chain
Management
Theory and Applications
Asset Analytics

Performance and Safety Management

Series Editors
Ajit Kumar Verma, Western Norway University of Applied Sciences, Haugesund,
Rogaland Fylke, Norway
P. K. Kapur, Centre for Interdisciplinary Research, Amity University, Noida, India
Uday Kumar, Division of Operation and Maintenance Engineering, Luleå
University of Technology, Luleå, Sweden
The main aim of this book series is to provide a floor for researchers, industries, asset
managers, government policy makers and infrastructure operators to cooperate and
collaborate among themselves to improve the performance and safety of the assets
with maximum return on assets and improved utilization for the benefit of society
and the environment.
Assets can be defined as any resource that will create value to the business. Assets
include physical (railway, road, buildings, industrial etc.), human, and intangible
assets (software, data etc.). The scope of the book series will be but not limited to:
• Optimization, modelling and analysis of assets
• Application of RAMS to the system of systems
• Interdisciplinary and multidisciplinary research to deal with sustainability issues
• Application of advanced analytics for improvement of systems
• Application of computational intelligence, IT and software systems for decisions
• Interdisciplinary approach to performance management
• Integrated approach to system efficiency and effectiveness
• Life cycle management of the assets
• Integrated risk, hazard, vulnerability analysis and assurance management
• Adaptability of the systems to the usage and environment
• Integration of data-information-knowledge for decision support
• Production rate enhancement with best practices
• Optimization of renewable and non-renewable energy resources

More information about this series at http://www.springer.com/series/15776

B. Vipin · C. Rajendran · Ganesh Janakiraman ·
Deepu Philip
Editors

Emerging Frontiers
in Operations and Supply
Chain Management
Theory and Applications
Editors
B. Vipin C. Rajendran
Department of Industrial Management Department of Management Studies
and Engineering Indian Institute of Technology Madras
Indian Institute of Technology Kanpur Chennai, Tamil Nadu, India
Kanpur, Uttar Pradesh, India
Deepu Philip
Ganesh Janakiraman Department of Industrial Management
Naveen Jindal School of Management and Engineering
The University of Texas at Dallas Indian Institute of Technology Kanpur
Richardson, TX, USA Kanpur, Uttar Pradesh, India

ISSN 2522-5162 ISSN 2522-5170 (electronic)

Asset Analytics
ISBN 978-981-16-2773-6 ISBN 978-981-16-2774-3 (eBook)
https://doi.org/10.1007/978-981-16-2774-3

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore
Pte Ltd. 2021
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface

In the era of advanced technological capabilities and sophisticated tools, managing

the transformation process in the manufacturing and service industries is increasingly
challenging due to the presence of uncertainty. Operations and supply chain manage-
ment play a critical role in achieving a better match of supply with demand in volatile
and uncertain conditions. With high volumes of data generated, data analytics, oper-
ations research, game theory, and other advanced optimization tools and techniques,
the opportunity to improve the performance of operations and supply chain manage-
ment is huge. A challenging issue in managing a supply chain in the present world is
to manage the uncertainty associated with globalization. For example, the disruptions
of low likelihood and high impact such as COVID-19 reinforce the need to advance
the body of knowledge in the field of operations and supply chain management to
achieve resilience.
This edited book covers various quantitative models on operations and supply
chain management such as inventory optimization, machine learning-operations
research integrated model for in health care operations, game-theoretic analysis of
review strategies in truthful information sharing, design of contracts in supply chains,
supply chain optimization, and shop floor scheduling. In addition to the quantitative
models, several innovative heuristics are proposed for different problems. This book
explores qualitative models on improving the performance of small and medium
enterprises and petroleum industries and a simulation model for staff allocation
in the information technology industry. Finally, this book provides comprehensive
review articles on vaccine supply chains and behavioral operations management.
The contributions of the chapters for this book are the submissions from the Society
of Operations Management Conference (SOM XXIII) held at the Indian Institute
of Technology Kanpur during December 19–21, 2019, and the invited research arti-
cles. The chapters are contributed by authors from India, United States, Germany,
Singapore, and United Arab Emirates.
This book has thirteen chapters allocated to three broad domains. In the first part
of the book, quantitative models on various operations and supply chain management
are presented. Chapter “A Comparative Study on Classical and New Hybrid Contin-
uous-Review Inventory Ordering Policies in a Supply Chain Using Mathematical
Models” presents new hybrid inventory policies by combining the traditional periodic
v
vi Preface

and continuous review policies under stochastic demand and compares the perfor-
mance of the developed policies with that of the traditional inventory policies. Chapter
“Truthful Information Sharing in a Multiretailer Supply Chain: Role of Review
Strategies” presents review strategies using repeated games to achieve truthful infor-
mation sharing in a multi-retailer supply chain setting. In chapter “Mathematical
Models and Heuristics for an Inventory Routing Problem Without Split Deliveries”,
mathematical models and heuristics are proposed to address the inventory routing
problem without split deliveries. Modeling of the supply chain under a price-sensitive
demand subjected to service level constraints in the presence of discrete transporta-
tion lead time is given in chapter “Modelling a Supply Chain with Price-Dependent
Stochastic Demand and Discrete Transportation Lead Time”. In chapter “Optimal
and Heuristic Profit Sharing Using Sales Rebate Contract in a Multi-level Supply
Chain”, the sales rebate contract is explored in a multi-level supply chain incorpo-
rating the fairness concern using heuristics based on genetic algorithm and simulated
annealing. Chapter “A Deterministic Heuristic Algorithm to Minimize the Length
of a Manufacturing Line in Transformation of Jobshops into Flowshops” focuses on
the development of a deterministic heuristic algorithm for solving the problem of
transforming jobshop into flowshop to minimize a manufacturing line length.
The second part of the book explores data analytics, qualitative, and simula-
tion approaches to address different operations and supply chain issues. Chapter
“A Classification Algorithm Based on Linear Regression and Linear Programming
for Predicting the Breast Cancer” presents a classification algorithm based on linear
regression and linear programming to predict breast cancer using the data from
the Wisconsin Diagnostic Breast Cancer data set. Evaluation of sustainable supply
chain using artificial neural network is discussed in chapter “Predicting Sustain-
able Supply Chain Performance Based on GRI Metrics and Multilayer Perceptron
Neural Networks”. In chapter “Green Supply Chain Management in the Indian
Petroleum Industry Using AHP-VIKOR Approaches”, a multi-criteria decision-
making approach is used to prioritize the green practices in the petroleum industry.
A simulation-based approach is employed to investigate the staff allocation problem
in the information technology industry in chapter “Staff Allocation for Projects in IT
Service Industries: A Simulation-Based Approach”.
The third part of the book is on the review of the literature and conceptual
framework development on issues related to operations and supply chain manage-
ment. Chapter “Supply Chain Management Practice and Supply Chain Performance:
A Conceptual Systematization of Terminology and Proposed Framework for SCMP
Implementation in Indian Manufacturing SMEs” proposes a conceptual frame-
work for supply chain management practice implementation in the Indian manufac-
turing industry. Chapter “Behavioural Operations Management: Trends and Insights”
presents a review of literature on behavioral operations management, analyzes the
trends, and provides insights from the literature. Chapter “Investigating the Vaccine
Supply Chain: A Review” investigates the literature on vaccine supply chain manage-
ment with a special focus on the Indian context primarily from the supply chain
coordination and supply chain network design perspectives.
Preface vii

We gratefully acknowledge the merit of all submissions to the book which could
not successfully go through the review process. We sincerely thank the reviewers
G. Anand (IIM Kozhikode), Deepak Eldho Babu (MACE Kothamangalam), Mukesh
Kumar Barua (IIT Roorkee), S. G. Deshmukh (IIT Delhi), M. S. Gajanand (IIM
Tiruchirappalli), P. Kalpana (IIITDM Kancheepuram), Arshinder Kaur (IIT Madras),
T. V. Krishna Mohan (IIM Bodh Gaya), C. Rajendran (IIT Madras), R. Rajesh
(IIIT Gwalior), A. Ramesh (IIT Roorkee), Sakthivel Madankumar (Trimble Infor-
mation Technologies, Chennai), Ramakrishnan Ramanathan (University of Bedford-
shire), Usha Ramanathan (Nottingham Trent University), R. Sridharan (NIT Calicut),
B. Srirangacharyulu (IIM Visakhapatnam), Godwin Tennyson (IIM Tiruchirappalli),
G. Thangamani (IIM Kozhikode), S. Venkataramanaiah (IIM Lucknow), B. Vipin
(IIT Kanpur), and S. Yamini (NIT Tiruchirappalli) who extended their support in
upholding the quality of this book by providing insightful review comments. We
thank Raghu Nandan Sengupta, IIT Kanpur, and Nupoor Singh, Springer for the
initiative, help, and support extended in materializing this edited book.
We hope that this edited book will advance the body of knowledge in the field
of operations and supply chain management and serve as a valuable resource for
academicians and practitioners. We believe that the book will be a good refer-
ence for academicians, practitioners, and students who are keen to learn the recent
advancements in operations and supply chain management.

Kanpur, India B. Vipin

Chennai, India C. Rajendran
Richardson, USA Ganesh Janakiraman
Kanpur, India Deepu Philip
Contents

Quantitative Models on Operations and Supply Chain Management

A Comparative Study on Classical and New Hybrid
Continuous-Review Inventory Ordering Policies in a Supply Chain
Using Mathematical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Bharathan Santhanam, P. V. R. Sethupathi,
Chandrasekharan Rajendran, and Hans Ziegler
Truthful Information Sharing in a Multiretailer Supply Chain:
Role of Review Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
C. Kavitha, R. K. Amit, and B. Vipin
Mathematical Models and Heuristics for an Inventory Routing
Problem Without Split Deliveries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
K. R. Preethi, Chandrasekharan Rajendran, and S. Viswanathan
Modeling a Supply Chain with Price-Dependent Stochastic
Demand and Discrete Transportation Lead Time . . . . . . . . . . . . . . . . . . . . . 65
Susheel Yadav, Anil K. Agrawal, and Manu K. Vora
Optimal and Heuristic Profit Sharing Using Sales Rebate Contract
in a Multi-level Supply Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
I. Thomson, Raichel Elsa Thomas, K. Susithra, S. Shobana,
Arshinder Kaur, and Chandrasekharan Rajendran
A Deterministic Heuristic Algorithm to Minimize the Length
of a Manufacturing Line in Transformation of Jobshops
into Flowshops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
J. Krishnaraj

ix
x Contents

Data Analytics, Qualitative, and Simulation Approaches to

Operations and Supply Chain Issues
A Classification Algorithm Based on Linear Regression and Linear
Programming for Predicting the Breast Cancer . . . . . . . . . . . . . . . . . . . . . . 139
Sakthivel Madankumar
Predicting Sustainable Supply Chain Performance Based on GRI
Metrics and Multilayer Perceptron Neural Networks . . . . . . . . . . . . . . . . . 159
Devendra Singh, Krishnanand Lanka, and P. R. C. Gopal
Green Supply Chain Management in the Indian Petroleum
Industry Using AHP-VIKOR Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
Sourabh Kumar and Mukesh Kumar Barua
Staff Allocation for Projects in IT Service Industries:
A Simulation-Based Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
V. V. Rajarajan and M. S. Gajanand

Literature Review and Conceptual Framework on Operations and

Supply Chain Management Aspects
Supply Chain Management Practice and Supply Chain
Performance: A Conceptual Systematization of Terminology
and Proposed Framework for SCMP Implementation in Indian
Manufacturing SMEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
Rohit Kumar and Manish Gupta
Behavioural Operations Management: Trends and Insights . . . . . . . . . . . . 233
S. Yamini
Investigating the Vaccine Supply Chain: A Review . . . . . . . . . . . . . . . . . . . . 251
Dheeraj Chandra and B. Vipin

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
Editors and Contributors

About the Editors

B. Vipin is an Assistant Professor at the Department of Industrial and Manage-

ment Engineering, Indian Institute of Technology Kanpur. He obtained his PhD
from Indian Institute of Management Madras, Chennai, India. He has published his
research findings in leading journals and he is a regular participant in leading oper-
ations management conferences. His areas of research interest are decision theory,
behavioral operations management, sustainable operations, and healthcare operations
management.

C. Rajendran is a Professor and RAGS Family Foundation Institute Chair at

the Department of Management Studies, Indian Institute of Technology Madras,
Chennai, India. He obtained his PhD in industrial engineering and management
from the same institute. He has been a recipient of Alexander von Humboldt
Research Fellowship more than 10 times since 1996 and has published in top jour-
nals of operations management. He has been awarded Dr. rer. pol. h. c. (Honorary
Doctorate) from the University of Passau in 2017, and elected Fellow of Indian
National Academy of Engineering (FNAE) in 2018. His areas of research interest
include scheduling in manufacturing and service systems, logistics management,
inventory optimization in deterministic systems, economic lot sizing and capacitated
lot sizing problems, heuristics & metaheuristics, and their applications in operations
and supply chain management, total quality management and service quality.

Ganesh Janakiraman is Ashbel Smith Professor of Operations Management at the

Naveen Jindal School of Management, The University of Texas at Dallas, USA. He
obtained his PhD from Cornell University, Ithaca, USA. He has numerous articles in
high ranking journals of operations management and has also attended several confer-
ences and seminars as an invited speaker. His areas of interest include stochastic
inventory theory, and, more broadly, decision making under uncertainty.

xi
xii Editors and Contributors

Deepu Philip is a Professor at the Department of Industrial and Management Engi-

neering, Indian Institute of Technology Kanpur. He obtained his PhD from MSU
Bozeman, USA. He is currently working on several research projects funded by
national and international agencies and has papers published in well-established
research journals. His areas of research interest include production and operations
management, systems engineering and simulation, and local search and optimization.

Contributors

Anil K. Agrawal Department of Mechanical Engineering, Indian Institute of Tech-

nology (BHU), Varanasi, India
R. K. Amit Department of Management Studies, Indian Institute of Technology
Madras, Chennai, India
Mukesh Kumar Barua Indian Institute of Technology Roorkee, Haridwar, India
Dheeraj Chandra Department of Industrial and Management Engineering, Indian
Institute of Technology Kanpur, Kanpur, Uttar Pradesh, India
M. S. Gajanand Faculty of Operations Management and Decision Sciences, Indian
Institute of Management, Tiruchirappalli, Tamil Nadu, India
P. R. C. Gopal Department of School of Management, National Institute of Tech-
nology Warangal, Warangal, Telangana, India
Manish Gupta Department of Mechanical Engineering, Motilal Nehru National,
Institute of Technology Allahabad, Prayagraj, India
Arshinder Kaur Department of Management Studies, Indian Institute of Tech-
nology Madras, Chennai, India
C. Kavitha Department of Management Studies, Indian Institute of Technology
Madras, Chennai, India
J. Krishnaraj Department of Mechanical Engineering, MLR Institute of Tech-
nology, Hyderabad, Telangana, India
Rohit Kumar Department of Mechanical Engineering, Motilal Nehru National,
Institute of Technology Allahabad, Prayagraj, India
Sourabh Kumar Indian Institute of Technology Roorkee, Haridwar, India
Krishnanand Lanka Mechanical Engineering Department, National Institute of
Technology Warangal, Warangal, Telangana, India
Sakthivel Madankumar Trimble Information Technologies India Private Limited,
Chennai, India
K. R. Preethi Indian Institute of Technology, Madras, India
Editors and Contributors xiii

V. V. Rajarajan Manager - Business Analytics, ZoomRx Healthcare Solutions Pvt

Ltd, Chennai, Tamil Nadu, India
Chandrasekharan Rajendran Department of Management Studies, Indian Insti-
tute of Technology Madras, Chennai, India
Bharathan Santhanam Department of Management Studies, Indian Institute of
Technology Madras, Chennai, India
P. V. R. Sethupathi Department of Management Studies, Indian Institute of Tech-
nology Madras, Chennai, India
S. Shobana Department of Industrial Engineering, College of Engineering Guindy,
Anna University, Chennai, India
Devendra Singh Mechanical Engineering Department, National Institute of Tech-
nology Warangal, Warangal, Telangana, India
K. Susithra Department of Industrial Engineering, College of Engineering Guindy,
Anna University, Chennai, India
Raichel Elsa Thomas Department of Mechanical Engineering, Indian Institute of
Technology BHU, Varanasi, India
I. Thomson Reckitt Benckiser Arabia, Dubai, UAE
B. Vipin Department of Industrial and Management Engineering, Indian Institute
of Technology Kanpur, Kanpur, Uttar Pradesh, India
S. Viswanathan Nanyang Technological University, Nanyang Ave, Singapore
Manu K. Vora Business Excellence Inc, Naperville, IL, USA
Susheel Yadav Department of Mechanical Engineering, Indian Institute of Tech-
nology (BHU), Varanasi, India
S. Yamini Faculty of Operations and Analytics, National Institute of Technology
Tiruchirappalli, Tiruchirappalli, Tamil Nadu, India
Hans Ziegler School of Business, Economics and Information Systems, University
of Passau, Passau, Germany
Quantitative Models on Operations
and Supply Chain Management
A Comparative Study on Classical
and New Hybrid Continuous-Review
Inventory Ordering Policies in a Supply
Chain Using Mathematical Models

Bharathan Santhanam, P. V. R. Sethupathi, Chandrasekharan Rajendran,

and Hans Ziegler

Abstract One of the most challenging tasks in managing a supply chain is to select
an efficient inventory ordering policy, i.e., deciding on ‘when to order’ and ‘how
much to order’ while minimizing the total cost and maximizing the service level.
Classical inventory policies based on periodic and continuous-review are generally
implemented in practice. In our work, we attempt to combine the characteristics
of both periodic-review order-up-to (R, S) policy and continuous-review (s, Q)/(s,
S) policies to propose two new hybrid ordering policies, namely continuous-review
(s, Q*) and continuous-review (s, OQ∗ ) hybrid policies. We further develop mixed
integer linear programming (MILP) models to obtain the optimal policy parameters
by considering a single-stage and two-stage supply chain with discrete, determin-
istic demand over a finite planning horizon. The proposed policies are benchmarked
against the existing order policies, namely (R, S), (s, Q), (s, S) and hybrid (R, S, Qmin )
policies. From results, we observe that the performance of the proposed hybrid poli-
cies outperforms the existing classical and hybrid policies in terms of total supply
chain cost.

Keywords Supply chain · Continuous-review · Hybrid inventory ordering policy ·

Finite time horizon · Mathematical models · Comparative study

B. Santhanam (B) · P. V. R. Sethupathi · C. Rajendran

Department of Management Studies, Indian Institute of Technology Madras, Chennai 600036,
India
C. Rajendran
e-mail: [email protected]
H. Ziegler
School of Business, Economics and Information Systems, University of Passau, 94032 Passau,
Germany
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 3
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_1
4 B. Santhanam et al.

1 Introduction

The worldwide out-of-stock rate is 7–8%. About 45% of the customers, who experi-
ence a stock-out would buy a different product, 31% of the customers buy elsewhere,
9% do not buy the product at all, and only 15% the customers would wait for the
product (Gruen et al., 2002). The comprised inventory can cost 20–40% of the product
price (Manatkar et al., 2016). Efficient management of inventory in a supply chain
improves the service offered to the consumers (Lee and Billington, 1992). Deciding
on ‘when to order’ and ‘how much to order’ is critical for minimizing the total cost
and maximizing the service level. One of the problem areas in operational decision-
making is inventory management (Ganeshan et al., 1999), and it depends on the
nature of demand (deterministic or stochastic), kind of decision control (decentral-
ized system or centralized) and inventory control policy. Inventory control policies
are classified as time and event-based. In time-based models, the inventory position
is reviewed in a fixed interval of time R, and the inventory position is raised to S. This
policy is referred to as (R, S) policy. In event-based models, the inventory position
is continuously reviewed, and whenever the inventory position equals or falls below
a predetermined reorder level, s either a fixed batch of Q units is ordered or the
inventory position is raised to S. The former is referred to as (s, Q) policy and the
latter as (s, S) policy.
The existing research on inventory systems is quite extensive. Hence, we discuss
only the most closely relevant papers related to the use of mathematical models for
inventory order policies and the hybrid order policies. Daniel and Rajendran (2006)
considered a supply chain in series with backorders allowed at each member and
proposed genetic algorithms (GA) to find the base-stock for each member. Later,
Sethupathi and Rajendran (2010) extended the work to determine the optimal review
period and base-stock for a supply chain controlled by periodic-review base-stock
policy. Further, Sethupathi et al. (2014) proposed mathematical models to evaluate
the performance of periodic-review order-up-to (R,S) policy and (s, S) policy. They
found (s, S) policy to perform superior when compared to (R, S) policy in most
settings for a supply chain in series with backorders. Movahed and Zhang (2015)
proposed a MILP model with uncertain lead time and demand to determine the
optimal order parameters for (s, S) policy. The (R, s, S) policy is one of first hybrid
policy proposed in literature. In this policy, the inventory position is reviewed in a
fixed interval of time R, and the inventory position is raised to S if the inventory
position equals or falls below a predetermined reorder level s. Popp (1965) extended
the (R, s, S) policy in which, if the demand is small, then the retailer orders using
(s, S) policy and if demand is large with a lower probability, then demand is directly
delivered from the manufacturer. This kind of hybrid ordering is ideal for scenarios
when holding costs are very high when compared to ordering cost and lead time is
zero. Zhou et al. (2007) proposed a new policy called (R, s, t, Qmin ) policy, in which
the inventory position is reviewed every R periods, and if it is lower than or equal
to the reorder level s, an order is placed to raise the inventory position to s + Qmin,
and when the inventory position is above s but lower than threshold t, then Qmin is
A Comparative Study on Classical and New Hybrid … 5

Table 1 Classification of
Author Parameters Classification
inventory ordering policy
papers from literature Popp (1965) (R, s, S) Time
Zhou et al. (2007) (R, s, t, Qmin ) Time
Sethupathi and Rajendran (2010) (R, S) Time
Kiesmüller et al. (2011) (R, S, Qmin ) Time
Sethupathi et al. (2014) (s, S) Event

ordered. Kiesmüller et al. (2011) proposed (R, S, Qmin ) policy, in which the inventory
position is reviewed every R period, and if it is less than S, an order is placed to raise
the inventory to S. However, if this order is smaller than Qmin , the order quantity is set
to Qmin . The policies proposed in literature can be summarized as given in Table 1.
From literature review, we observe that the hybrid policies proposed in the liter-
ature are time-based, i.e., the inventory position is reviewed in a fixed interval of
time. There seems to be no event-based hybrid order policy. Also, to the best of our
knowledge, no study has been undertaken on the relative assessment and comparison
of various time and event-based inventory order policies. The salient features of our
work are as follows.
• New hybrid inventory order policies: Two new hybrid inventory ordering policies
that combine the benefits from continuous-review policies and periodic-review
policies are proposed. When the inventory is continuously reviewed, it is likely to
lead to less shortage cost and ordering a fixed order quantity. Q possibly minimizes
the long-term holding and ordering costs when demand is uniform. The benefit
from periodic-review order-up-to (R, S) policy or continuous-review (s, S) policy
is to raise the inventory position to S when there is an order placement, resulting
in lower shortage costs, especially when there is high fluctuation in demand.
• Development of mathematical models: Mathematical models are developed for
the proposed and the existing policies to determine the optimal policy parameters
by considering a single-stage and two-stage supply chain operating with discrete,
deterministic demand over a finite planning horizon.
• Supply chain with order costs: Optimal control of inventory in supply chain is diffi-
cult when order costs are applicable, and the conventional relationship between
the order policies in the case of a single installation might not be applicable for
the case of a supply chain (refer to Wang, 2011, for details). In addition, Wang
found that the performance of stock-based (s, Q) policy is superior than the time-
based (R, S) policy for a single installation and expected this dominance to carry
over to supply chain. This observation was more of a hypothesis, and there is no
experimental study to prove it. This is experimentally investigated in this work.
• Comparative evaluation of inventory order policies: For the first time in litera-
ture, a relative comparison of six inventory control policies, namely the proposed
continuous-review (s, Q* ) hybrid policy and continuous-review (s,OQ* ) hybrid
policy, and the existing policies, namely (R, S), (s, Q), (s, S) and hybrid (R, S, Qmin )
6 B. Santhanam et al.

policies, are presented by considering a two-stage supply chain operating with

backorders and discrete, deterministic demand. Kiesmüller et al. (2011) observed
that the (R, S, Qmin ) policy performs better than hybrid (R, s, S) policy, and hence,
it is not considered in this study.
The remainder of the paper is organized as follows. Section 2 describes the
assumptions, notations and the sequence of events with an example. Section 3
discusses the proposed mathematical models. Section 4 describes the data sets and
compares the performance of proposed versus existing inventory policies. Section 5
presents limitations and future work. Finally, Sect. 6 discusses the summary of the
work.

2 Assumptions, Notations and Sequence of Events

This work considers a two-stage supply chain with a distributor (DC) and a retailer
(R). Both members operate with their respective local/installation holding cost per
unit per day, backorder cost per unit per day and an order placement cost per order.
The cost of obsolescence and damage are considered in local/installation holding cost
per unit per day, and costs of loading/unloading are considered as part of ordering
cost per order. The processing lead time is integrated with transportation lead time to
arrive at a replenishment lead time for each member. The retailer faces the discrete
customer demand which is known a priori. The excess demand is assumed to be
backordered. The unit of time is one day and is assumed to be discrete. The notations
used in the mathematical model are as below:
Notation Description
TSC total supply chain cost
T planning horizon in days over which TSC is optimized
N number of members in the supply chain
i index for members in the supply chain
T current day
ROLi reorder level or reorder point for member i
/*referred to as ‘s’ in continuous-review (s, Q), continuous-review
(s, Q*) and continuous-review (s, OQ∗ ) policies. However, for
better clarity, s is represented as ROL and order-up-to-level as
S*/
SUMDEMi,t total demand received by member i up to t
SUMORDi,t−1 sum of orders triggered at member i up to t−1
Si order-up-to-level at member i
OQi notional fixed order quantity for member i
hi cost of holding a product per day for member i
Oi ordering cost per order for memberi
bi cost incurred due to product shortage for member i
LTi replenishment lead time for member i
A Comparative Study on Classical and New Hybrid … 7

OIi,t on-order inventory for member i on day t

Ii,t on-hand inventory for member i on day t
Bi,t backorder cost per product per day at member i on day t
Di,i+1,t demand by member i to upstream member (i + 1) at t
∗
Q i,t variable order quantity which is defined as the maximum of
 
(SUMDEMi,t − SUMORDi,t−1 ); OQi for member i on day t
∗
OQi,t variable order quantity which is defined as maximum of order-
up-to level minus the inventory position and notional fixed order
quantity for member i on day t
θi,t a binary variable to decide on the order quantity if there is an order
triggered at member i on day t; else it takes the value 0
∗
θi,t a binary variable with a value 1, if there is an order triggered at
member i on day t; else it takes the value 0
QSi,i−1,t quantity shipped from member i to (i−1) on day t
QS N +1,N ,t quantity of products that is shipped to member N on day t
M a large positive integer value
The order of events on each day for each member is as follows:
(1) The quantity shipped (QSi+1,i,t−L T i ) from the immediate upstream member is
received if the quantity is to arrive by considering the replenishment lead time
for that member.
(2) The distributor receives demand (Di−1,i,t ) from the retailer, and the retailer
receives the customer demand. Demand is fulfilled based on the available
on-hand inventory (Ii,t ) and the order quantity received from the upstream
member (QSi+1,i,t−L T i ). Any excess demand is backordered (Bi,t ). The on-
hand inventory (Ii,t ) or backorder (Bi,t ) information is updated.
(3) The inventory position of the member is defined as the sum of pipeline inventory
and the inventory level. Pipeline inventory is the difference between previous
day’s on-order inventory (OIi,t−1 ) and quantity shipped (QSi+1,i,t−L T i ) from
upstream member. Inventory level is calculated based on the difference between
the on-hand inventory (Ii,t ) and the backorder (Bi,t ). If the inventory position
is equal to or less than the reorder level (ROLi ) at the member, an order is
triggered to the upstream member. No order placement takes place otherwise.
The order quantity depends on the policy under study.
(4) Finally, the quantity shipped (QSi,i−1,t ) between members and the on-order
information (OIi,t ) are updated.
The behavior of different continuous-review policies, namely (s, Q), (s, S), (s, Q ∗ )
and (s, OQ∗ ) is explained with an example. In all policies, the inventory position is
continuously reviewed, and when it equals or falls below the reorder level, an order
is placed. The order quantity depends on the policy. Here, we consider a single-stage
system with randomly distributed demand between 20 and 60 and assume the fixed
order quantity and order-up-to level as 40 and reorder level as 20. The lead time is
fixed as 2 days. Holding, backorder and ordering costs are set as 8 per unit per day,
16 per unit per day and 640 per order, respectively.
8 B. Santhanam et al.

From Table 2, we observe that whenever there is unexpected rise or dip in demand,
the order quantity is based on the cumulative demand since the last order or the
order-up-to level minus the inventory position in the continuous-review (s, Q ∗ ) and
(s, OQ∗ ) hybrid policies, respectively. If there is dip in demand or when the demand
is uniform, then order quantity is based on the fixed order quantity. On comparing the
holding cost, it is the least for continuous-review (s, Q) policy. This is not unexpected
considering the fixed order quantity. The ordering cost is the least for continuous-
review (s, S) policy while shortage cost is the least for the proposed continuous-review
(s, OQ∗ ) policy. In terms of the total cost, (s, OQ∗ ) performs the best followed by
(s, Q ∗ ), (s, S) and (s, Q) policies.

3 Mathematical Models

Mathematical models for the proposed continuous-review (s, Q ∗ ) hybrid policy,

continuous-review (s, O Q ∗ ) hybrid policy and existing continuous-review (s, Q)
policy are given in Sects. 3.1, 3.2 and 3.3, respectively. The mathematical models
are generic and can be run for a supply chain with any number of members.

3.1 Mathematical model for the Proposed

Continuous-Review (s, Q ∗ ) Hybrid Policy

In the proposed continuous-review (s, Q ∗ ) hybrid policy, when the inventory position
for a member is equal to or less than the reorder level s, a replenishment order equal
to the maximum of the notional fixed order quantity and the difference between total
demand received by the member up to current day t and total order quantity up to
day (t-1) are placed to the next upstream member. The order quantity is referred as
the notional order quantity because it is fixed (time invariant) and the actual order
placement reckons with this order quantity as well as with the total demand received
by the member up to the current day t and the total order quantity up to day t-1. When
there is an abrupt rise or dip in demand, the order quantity will be the total demand
received by the member up to the current day t minus the total order quantity up
to day t-1. However, when there is uniform demand with less variations, the order
quantity will be equal to the notional fixed order quantity. This flexibility in order
quantity aims to reduce the shortage cost when there is variation in demand, and
the holding cost when there is uniform demand or dip in demand. Also, the notional
fixed order quantity can be used to define the minimum order quantity (if any) that
exists with the upstream member.
Table 2 Order policy behavior (in terms of order quantity) and associated costs for (s, Q), (s,S), (s, Q*) and (s, OQ*) policy for a sample demand stream
Day Demand Order quantity Holding cost Shortage cost Ordering cost
(s, Q) (s, S) (s, Q*) (s,OQ*) (s, Q) (s, S) (s, Q*) (s,OQ*) (s, Q) (s, S) (s, Q*) (s, OQ*) (s, Q) (s, S) (s, (s, OQ*)
Q*)
1 10 0 0 0 0 240 240 240 240 0 0 0 0 0 0 0 0
2 15 40 25 40 40 120 120 120 120 0 0 0 0 640 640 640 640
3 20 0 20 0 0 0 0 0 0 80 80 80 80 0 640 0 0
4 30 40 30 40 40 40 0 40 40 0 160 0 0 640 640 640 640
5 15 40 40 40 50 0 0 0 0 160 80 160 160 640 640 640 640
6 10 0 0 0 0 160 120 160 160 0 0 0 0 0 0 0 0
7 15 40 30 40 40 360 320 360 440 0 0 0 0 640 640 640 640
8 5 0 0 0 0 320 280 320 400 0 0 0 0 0 0 0 0
9 15 0 0 0 0 520 400 520 600 0 0 0 0 0 0 0 0
10 10 40 30 40 0 440 320 440 520 0 0 0 0 640 640 640 0
A Comparative Study on Classical and New Hybrid …

11 50 0 20 0 40 40 0 40 120 0 160 0 0 0 640 0 640

12 45 40 45 40 40 0 0 0 0 0 400 0 480 640 640 640 640
13 55 40 85 55 45 0 0 0 0 880 960 880 720 640 640 640 640
14 50 40 40 50 90 0 0 0 0 1040 1040 1040 880 640 640 640 640
15 45 40 70 45 45 0 0 0 0 1120 400 880 880 640 640 640 640
16 55 40 95 55 60 0 0 0 0 1360 640 960 320 640 640 640 640
17 25 40 0 40 70 0 40 0 0 1120 0 640 0 640 0 640 640
18 25 40 35 40 0 0 600 0 280 880 0 160 0 640 640 640 0
19 25 40 50 40 40 0 400 40 640 640 0 0 0 640 640 640 640
Total 2240 2840 2280 3560 7280 3920 4800 3520 8320 8960 8320 7680
Total cost (holding + shortage + ordering costs) 17,840 15,720 15,400 14,760
9
10 B. Santhanam et al.

Objective function:
The objective is to minimize the overall cost that includes the holding, backo-
rder/shortage and ordering costs for all members over T days.


T 
N
 
∗
Minimize TSC = h i Ii,t + bi Bi,t + Oi θi,t (1)
t=1 i=1

Constraints:
for t =1, 2,…T
[
for i =1, 2,…N
{
The on-hand inventory or backorder for the current day is calculated using Eq. (2)
based on the available on-hand inventory at member i on previous day, the member’s
backorder as on the previous day, the demand received from the downstream member
(i-1) and the order quantity received at t considering the replenishment lead time of
member i. Retailer receives the customer demand.

Ii,t − Bi,t = Ii,t−1 + QSi+1,i,t−LTi − Bi,t−1 − Di−1,i,t (2)

If adequate on-hand inventory is not available with member i to meet the

accumulated demand, then the excess demand is backordered. The assumption is
bi + h i ≥ h i−1 , i ≥ 2, and if that does not hold, then Ii,t and Bi,t will coexist. To
prevent this, Expression (3) needs to be added as an additional constraint.
∗∗
 ∗∗

Ii,t ≤ M × θi,t and Bi,t ≤ M 1 − θi,t ,
∗∗
with θi,t ∈ {0, 1}, for ∀i ≤ N and ∀t ≤ T (3)

Expression (4) updates the sum of demand received at member i till day t.

SUMDEMi,t = SUMDEMi,t−1 + Di−1,i,t (4)

Constraints (5) and (6) monitor the inventory position at member i and trigger an
order if the inventory position is equal to or less than the reorder level at i. No order
placement takes place otherwise.
 ∗

Ii,t − Bi,t + OIi,t−1 − QSi+1,i,t−LTi ≤ ROLi + M 1 − θi,t (5)

∗
Ii,t − Bi,t + OIi,t−1 − QSi+1,i,t−LTi ≥ ROLi + 1 − Mθi,t (6)
A Comparative Study on Classical and New Hybrid … 11

Constraints (7)–(13) set the order quantity exactly equal to the maximum of
notional order quantity OQi and the total demand minus the cumulative total order
quantity SUMDEMi,t − SUMORDi,t−1 at the member i if there is an order place-
∗
ment at i. An order placement is indicated by θi,t when the inventory position is equal
to or less than the ROLi . When the inventory position is greater than or equal to the
∗
ROLi , θi,t becomes 0 indicating no order placement, and Di,i+1,t becomes equal to
zero.
∗
Q i,t ≥ SUMDEMi,t − SUMORDi,t−1 (7)

∗
Q i,t ≥ OQi (8)

∗
Q i,t ≤ M × θi,t + SUMDEMi,t − SUMORDi,t−1 (9)

∗
 
Q i,t ≤ M 1 − θi,t + OQi (10)

∗
 ∗

Di,i+1,t ≤ Q i,t + M 1 − θi,t (11)

∗
 ∗

Di,i+1,t ≥ Q i,t − M 1 − θi,t (12)

∗
Di,i+1,t ≤ M × θi,t (13)

Equation (14) calculates the quantity shipped between members based on the
on-hand inventory and quantity received from the upstream member.

QSi,i−1,t = Ii,t−1 − Ii,t + QSi+1,i,t−LTi (14)

Equation (15) updates the on-order inventory at member i on day t.

OIi,t = OIi,t−1 − QSi+1,i,t−LTi + Di,i+1,t (15)

Equation (16) updates the sum of orders at i at the close of day t.

SUMORDi,t = SUMORDi,t−1 + Di,i+1,t (16)

}
Equation (17) shows the quantity of finished products shipped immediately to the
distributor.

QS N +1,N ,t = D N ,N +1,t (17)

12 B. Santhanam et al.

]
The sequence of events repeats on each day for all members in the supply chain.
Initial conditions:
For all members, the initial inventory is initialized to be notional order quantity, the
reorder level should be less than the notional order quantity, and all other variables
related to inventory position are initialized to zero.

Ii,0 = OQi , ROLi ≤ OQi , Bi,0 = 0, OIi,0 = 0, SUMDEMi,0

= 0, SUMORDi,0 = 0 (18)

QSi+1,i,t−LTi = 0, ∀t ≤ LTi and ∀i ≤ N (19)

Binary variables and non-negativity constraints:

∗
θi,t , θi,t ∈ {0, 1}, all other variables ≥ 0, ∀i ≤ N and ∀t ≤ T. (20)

3.2 Mathematical Model for the Proposed

Continuous-Review (s, O Q * ) Policy

In the proposed continuous-review (s, OQ∗ ) policy, whenever the inventory position
equals or falls below the reorder level s, an order equal to the maximum of the
notional order quantity (adapted from conventional (s, Q) policy) and the order-up-
to-level minus the inventory position (adapted from (R, S) and (s, S) policies) is
placed to the upstream member. By making the order quantity dynamic based on
the inventory position, we aim to handle variation in demand. For example, if there
is uniform demand with less variation, then the order quantity will be based on the
notional fixed order quantity, but when there is abrupt rise or dip in demand, the
order quantity will be based on the difference between the order-up-to-level and the
inventory position. This flexibility in order quantity reduces the shortage cost when
there is more variation in demand and the holding cost when there is less variation in
demand. The notations and the objective functions are the same as used in Sect. 3.1.
Constraints:
for t =1, 2,…T
[
for i =1, 2,…N
{
A Comparative Study on Classical and New Hybrid … 13

In addition to the Constraints (2)–(6) and (13)–(17), and omitting others,

mentioned in Sect. 3.1, the following constraints are added.
Constraints (21)–(26) set the order quantity exactly equal to the maximum
of {notional
  fixed order quantity (OQi ); order-up-to-level
 minus inventory posi-
tion Si − Ii,t − Bi,t + OIi,t−1 − QSi+1,i,t−LTi if there is an order placement at
∗
member i, (an order placement is indicated by θi,t set to one when the inventory posi-
tion at the current day t is equal to or less than ROLi ). Note that when the inventory
position at the current day t is greater than ROLi , no order takes place, and hence
Di,i+1,t becomes equal to zero.
∗
 
OQi,t ≥ Si − Ii,t − Bi,t + OIi,t−1 − QSi+1,i,t−LTi (21)

∗
OQi,t ≥ OQi (22)

∗
 
OQi,t ≤ M × θi,t + Si − Ii,t − Bi,t + OIi,t−1 − QSi+1,i,t−LTi (23)

∗
 
OQi,t ≤ M 1 − θi,t + OQi (24)

∗ ∗
Di,i+1,t ≤ OQi,t + M(1 − θi,t ) (25)

∗ ∗
Di,i+1,t ≥ OQi,t − M(1 − θi,t ) (26)

}
]
The sequence of events repeats on each day for all members in the supply chain.
Initial conditions:
For all members, the initial inventory is initialized to be the maximum of the notional
fixed order quantity and the order-up-to-level, the reorder level should be less than
the notional fixed order quantity, and all other variables related to inventory position
are initialized to zero.

Ii,0 ≥ OQi , Ii,0 ≥ Si , ROLi ≤ OQi , Bi,0 = 0∀i ≤ N ; (27)

 
Ii,0 ≤ Si + M × θi,0 , Ii,0 ≤ OQi + M 1 − θi,0 , ∀i ≤ N ; (28)

QSi+1,i,t−L Ti = 0, ∀t ≤ LTi and ∀i ≤ N . (29)

Binary variables and non-negativity constraints:

14 B. Santhanam et al.

∗
θi,t , θi,t ∈ {0, 1}, all other variables ≥ 0, ∀i ≤ N , all ∀t ≤ T. (30)

3.3 Mathematical Model for Continuous-Review (s,Q) Policy

The same notations and objective function are the same as used in Sect. 3.1.
Constraints:
for t =1, 2,…T
[
for i =1, 2,…N
{
In addition to the Constraints (2)–(6), (13)–(17), and omitting others, mentioned
in Sect. 3.1, the following constraints are added:
Constraints (31) and (32) force the order quantity to the upstream member to be
equal to the fixed order quantity if there is an order placement or else it is 0.
 ∗

Di,i+1,t ≤ OQi + M 1 − θi,t (31)

 ∗

Di,i+1,t ≥ OQi − M 1 − θi,t (32)

}
]
The sequence of events repeats on each day for all members in the supply chain.
Initial conditions:
For all members, the initial inventory on-hand is initialized to be the fixed order
quantity, the reorder level should be less than the fixed order quantity, and all other
variables related to inventory position are initialized to zero.

Ii,0 = OQi , ROLi ≤ OQi , Bi,0 = 0, ∀i ≤ N ; (33)

QSi+1,i,t−LTi = 0, ∀t ≤ LTi and ∀i ≤ N . (34)

Binary variables and non-negativity constraints:

∗
θi,t ∈ {0, 1}, all other variables ≥ 0, ∀i ≤ N and ∀t ≤ T. (35)
A Comparative Study on Classical and New Hybrid … 15

4 Experimental Evaluation

Section 4.1 discusses the data sets and the experimental settings considered in this
work. In Sect. 4.2, the performance evaluation of proposed and existing policies is
presented.

4.1 Data Sets and Experimental Settings

The proposed as well as the existing inventory policies are evaluated by considering
640 supply chain test problem instances (2 lead time settings × 2 scenarios × 16
cost settings × 10 demand streams). Two lead time settings, namely LT1 and LT2,
with replenishment lead time of one and two days, respectively, for each member,
are considered. In the scenario 1, a single-stage supply chain is considered where
the retailer has unlimited availability of finished products from its upstream member.
In the scenario 2, a two-stage supply chain is considered where the distributor has
unlimited availability of finished products from its upstream member and follows
the same ordering policy with cost of holding, shortage and ordering. The different
cost settings considered in this work are reported in Table 3.

Table 3 Supply chain setting with respect to supply chain members

Supply chain Member holding cost Member shortage cost Member ordering cost
settings DC R DC R DC R
CS1 4 8 8 16 320 640
CS2 4 8 16 32 320 640
CS3 4 8 32 64 320 640
CS4 4 8 64 128 320 640
CS5 4 8 8 16 640 1280
CS6 4 8 16 32 640 1280
CS7 4 8 32 64 640 1280
CS8 4 8 64 128 640 1280
CS9 4 8 8 16 1280 2560
CS10 4 8 16 32 1280 2560
CS11 4 8 32 64 1280 2560
CS12 4 8 64 128 1280 2560
CS13 4 8 8 16 2560 5120
CS14 4 8 16 32 2560 5120
CS15 4 8 32 64 2560 5120
CS16 4 8 64 128 2560 5120
16 B. Santhanam et al.

Ten uniformly distributed customer demand streams are sampled between 20 and
60. The different cost settings are assumed as follows: shortage cost = C × holding
cost where C = 2, 4, 8, 16 and ordering cost =K × E(D)× holding cost where E(D)
is the expected retail demand with respect to the demand distribution and K = 2,
4, 8, 16. The rationale for this is that we relate the holding cost as the base for the
ordering cost because we assume a supply chain with a very high service level (see
Silver et al., 1998).

4.2 Comparison on the Performance of Proposed Hybrid

Policies versus Existing Policies

The objective of this study is to make a comparative evaluation of proposed

continuous-review hybrid policies against continuous-review (s, Q) policy,
continuous-review (s, S) policy, periodic-review order-up-to (R, S) policy, (R, s,
S) policy and (R, S, Qmin ) policy. The (R, S) order policy is considered for compar-
ison because it is generally used in real life, especially when products are ordered
from the same supplier so that coordination of replenishments of related products is
achieved (Silver et al., 1998). We adapt the mathematical model from Sethupathi and
Rajendran, (2010) for the (R, S) policy and Sethupathi et al. (2014) for (s, S) policy
and studied the same for a two-stage supply chain. For the (R, S, Qmin ) policy, the
(R, S) model, with necessary modifications in equations related to order quantity is
run with different review periods varying from 1 to 5 and the best optimal solution is
considered. The mathematical models are run for 40 days for single- and two-stage
supply chain using ILOG CPLEX solver for different supply chain settings.
The relative percentage deviation for an order policy is calculated as described in
Eq. (36):
 
TCpolicy − TCbest
Relative percentage deviation = × 100, (36)
TCbest

where TCpolicy is the optimal total cost obtained by the respective policies for a
particular lead time and cost settings, and TCbest is the best total cost obtained across
all policies for a particular lead time and cost settings. The average relative percentage
deviation obtained for each policy in terms of total supply chain cost is given in Table
4 for a single-stage supply chain and Table 5 for a two-stage supply chain. In Table
6, the total supply chain cost for a sample demand stream is given for a single-stage
supply chain. From results, we see that the (s, OQ∗ ) performs the best in terms of
total supply chain cost, followed by (s, Q ∗ ), (s, Q), (s, S), (R, S, Qmin ) and (R, S)
policies in same order. (R, S, Qmin ) policy performs better than (R, S) policy and
continuous-review policies outperform the existing hybrid (R, S, Qmin ) policy where
the inventory control is time-based. (s, S) policy performs better than (R, S). From
the results, we clearly see the superior performance of continuous-review policies
Table 4 Performance comparison of inventory order policies in respect of average relative percentage deviation for total supply chain cost for single-stage
supply chain
Supply chain setting LT1 LT2
(R, S) (s, S) (s, Q) (R, S, Qmin ) (s, Q∗) (s, O Q∗) (R, S) (s, S) (s, Q) (R, S, Qmin ) (s, Q ∗ ) (s, O Q ∗ )
CS1 9.98 1.48 1.51 2.59 1.32 0.00 10.02 4.7 2.21 2.64 1.46 0.88
CS2 12.47 2.30 1.56 1.93 1.18 0.00 9.08 4.86 1.42 3.01 1.04 0.51
CS3 11.45 1.67 1.83 2.36 1.59 0.00 10.39 4.01 2.01 2.38 1.31 0.29
CS4 14.35 1.62 1.13 2.64 0.98 0.62 10.67 3.87 2.06 1.96 1.16 0.82
CS5 9.52 4.86 2.01 1.86 1.96 0.00 12.14 3.61 1.33 1.57 1.39 0.02
CS6 12.24 4.24 1.90 1.93 1.54 0.00 12 2.95 1.52 1.85 1.13 0.21
CS7 13.73 5.02 2.16 1.77 0.79 0.00 12.1 2.79 2.33 1.61 1.13 0.60
CS8 14.87 6.89 1.16 0.33 0.91 0.13 11.8 4.69 1.78 1.8 0.94 0.50
A Comparative Study on Classical and New Hybrid …

CS9 12.72 2.55 1.11 1.9 2.43 0.14 10.82 3.0 2.17 2.30 1.22 0.04
CS10 10.88 2.69 1.22 1.97 0.76 0.00 12.31 4.41 2.31 2.03 1.40 0.90
CS11 16.31 3.34 1.20 1.79 1.05 0.00 12.51 3.01 2.41 2.6 0.93 0.86
CS12 14.10 3.83 0.76 1.38 1.28 0.43 10.01 3.93 2.04 2.84 1.45 0.1
CS13 13.05 3.97 1.81 2.11 1.97 0.07 11.94 2.96 1.72 2.38 1.05 0.14
CS14 11.90 2.00 0.88 2.72 1.00 0.25 11.25 3.42 2.5 1.73 1.03 0.38
CS15 13.74 1.57 0.68 2.18 0.59 0.00 11.22 3.92 1.39 1.5 1.14 0.39
CS16 10.21 2.11 1.02 2.04 0.81 0.00 11.74 4.09 2.6 2.06 1.33 0.73
Avg 12.60 3.13 1.37 1.97 1.26 0.10 11.30 3.76 1.96 2.11 1.21 0.46
17
 18

Table 5 Performance comparison of inventory control policies in respect of average relative percentage deviation for total supply chain cost for two-stage
supply chain
Supply chain setting LT1 LT2
(R, S) (s, S) (s, Q) (R, S, Qmin ) (s, Q∗) (s, OQ∗ ) (R, S) (s, S) (s, Q) (R, S, Qmin ) (s, Q ∗ ) (s, OQ∗ )
CS1 9.47 3.75 2.67 3.25 1.79 0.39 10.08 4.45 2.06 3.40 1.35 0.33
CS2 10.73 4.89 1.82 3.81 1.29 0.03 10.47 4.49 2.57 3.62 2.08 0.29
CS3 9.38 3.85 2.39 2.64 1.68 0.28 9.84 4.88 1.75 3.73 2.13 0.42
CS4 9.22 4.38 3.29 3.84 1.49 0.05 10.41 3.87 2.09 3.30 1.77 0.18
CS5 9.06 3.56 2.38 3.13 2.47 0.44 9.76 3.65 1.95 3.21 1.49 0.34
CS6 10.66 3.78 2.73 3.07 1.10 0.20 10.25 4.07 1.72 2.91 1.62 0.33
CS7 10.73 3.65 2.13 3.13 2.27 0.24 9.84 4.99 3.02 3.32 1.30 0.01
CS8 9.21 4.88 2.12 3.73 1.11 0.41 10.29 3.68 1.64 2.69 1.96 0.15
CS9 10.62 5.14 1.66 3.87 2.30 0.30 9.76 3.73 2.21 3.58 2.00 0.14
CS10 10.91 4.74 3.11 2.64 2.16 0.04 9.78 3.98 2.96 3.04 1.70 0.04
CS11 10.72 4.41 1.53 3.34 2.48 0.15 10.05 3.60 2.77 3.25 1.45 0.10
CS12 10.61 4.16 2.56 3.91 1.58 0.41 10.42 4.57 1.68 2.72 2.37 0.09
CS13 10.98 4.34 2.77 3.28 1.21 0.42 10.66 3.96 2.97 3.26 1.56 0.06
CS14 9.07 4.44 2.45 3.87 2.27 0.43 9.90 3.69 2.97 2.73 1.35 0.39
CS15 10.42 5.11 3.16 3.47 1.13 0.34 10.22 3.59 2.07 3.20 2.49 0.31
CS16 10.90 4.99 2.33 3.84 1.52 0.19 10.23 5.17 3.13 2.81 1.53 0.12
Avg 10.17 4.38 2.44 3.43 1.74 0.27 10.12 4.15 2.35 3.17 1.76 0.21
B. Santhanam et al.
A Comparative Study on Classical and New Hybrid … 19

Table 6 Performance comparison of inventory control policies in respect of total supply chain cost
for a sample demand stream in a single-stage supply chain
Supply chain setting (R, S) (s, S) (s, Q) (R, S, Qmin ) (s, Q ∗ ) (s, OQ∗ )
CS1 22,124 21,520 20,512 20,884 20,505 20,496
CS2 24,296 23,784 22,416 22,564 22,416 22,416
CS3 26,656 25,552 24,688 24,908 24,161 23,376
CS4 28,856 26,928 25,632 26,734 24,753 24,136
CS5 30,716 29,472 28,896 30,188 28,800 28,824
CS6 34,620 32,240 32,288 33,822 32,174 31,504
CS7 37,080 35,152 33,728 35,870 33,712 33,520
CS8 40,276 36,624 34,744 35,501 34,657 34,376
CS9 41,964 39,368 39,536 39,664 39,368 39,216
CS10 47,284 44,352 44,384 46,064 44,149 44,096
CS11 49,880 46,960 46,992 48,894 46,586 46,168
CS12 53,076 48,184 47,888 51,402 47,460 47,040
CS13 55,552 57,288 54,804 55,335 57,288 54,804
CS14 62,184 62,032 60,880 61,591 61,869 60,704
CS15 68,064 64,880 65,144 66,860 65,128 64,880
CS16 72,616 66,104 66,160 69,099 65,848 65,720

over the periodic-review policies in terms of total supply chain cost. On comparing
the (s, S) and (s, Q) policies, (s, Q) policy seemed to be superior to the (s, S) policy in
terms of holding and ordering costs. On comparing the (s, Q) and hybrid policies, we
see that the (s, Q) policy had relatively lower holding cost. However, in the proposed
hybrid policies, since order quantities are made dynamic, the system adapts quickly
to dynamic demand patterns, thereby balancing both holding and backorder costs
across the members. The dynamically varying actual order quantity helps a member
to adjust to any surge or dip in demand since the last order, thereby resulting in less
backorder costs.
Overall, the proposed hybrid policies outperform the existing (R, S), (s, S), (s,
Q), and (R, S, Qmin ) policies in terms of total cost. We find that the performance
of proposed policies is quite robust when evaluated across different cost settings
(e.g., high ratio of shortage to holding cost as well as in presence of relative low and
high ordering costs). It is evident that, on the whole, the (s, OQ∗ ) policy with the
continuous-review system emerges to be the best, followed by the (s, Q ∗ ) policy, (s,
Q) policy, (s, S) policy, (R, S, Qmin ) and then (R, S) policy. The (s, OQ∗ ) policy takes
advantage of the tight control of (s, Q) policy, and it is also able to exploit the dynamic
order quantities similar to the (R, S) policy by setting the actual order quantity equal
to the maximum of notional fixed order quantity and the order-up-to-level minus the
inventory position, and thereby this hybrid policy yields the least total supply chain
cost in almost all the settings. The (s, Q ∗ ) policy capitalizes on the tight control of (s,
Q) policy and is able to adapt to dynamic changes in demand as the order quantity is
20 B. Santhanam et al.

set to maximum of the notional fixed order quantity and the difference between total
demand received by the member up to current day t and total order quantity up to
day (t-1). The performance of the continuous-review (s, Q) policy and continuous-
review (s, S) policy is superior to the periodic-review (R, S) policy, and this can be
attributed to the continuous-review system followed in (s, Q) and (s, S) policy. This
experimentally confirms the hypothesis made by Wang (2011) that the performance
of stock-based (s, Q) policy is superior than the time-based (R, S) policy for a single
installation and that dominance may carry over to supply chains.

5 Limitations and Future Work

The main purpose of this paper is to propose new hybrid inventory ordering policies,
develop mathematical models and compare its performance against existing order
policies in literature. To evaluate the polices, we considered a two-stage system with
discrete deterministic demand and treated any unfulfilled demand as backordered
with no service level constraints. However, in current business environment with
availability of alternate products and impatient behavior of customers, we may need to
consider unfulfilled demand as lost demand. As future work, we wish to comprehend
the performance of proposed policies when unfulfilled demand is treated as lost sales,
when there are more than two stages in the supply chain, when there is lead time
variability and understand the behavior of the policies when service level constraints
exist.

6 Summary

In this work, two new hybrid inventory ordering policies that combine the benefits
and characteristics of periodic-review (R, S) policy, continuous-review (s, Q) policy
and continuous-review (s, S) policy are proposed and studied in the context of single
and two-stage supply chain operating under backorders. Mathematical models are
proposed to determine the optimal inventory order policy parameters with the consid-
eration of discrete, deterministic demand over a finite planning horizon. A relative
evaluation of the proposed two hybrid policies (continuous-review (s, Q ∗ ) policy and
continuous-review (s, OQ∗ ) policy) with the existing policies such as periodic-review
order-up-to-S (R, S) policy, continuous-review (s, Q) policy, continuous-review (s,
S) policy and hybrid (R, S, Qmin ) policy is presented. Based on the computational
results, the proposed continuous-review (s, OQ∗ ) policy performs the best because
it attempts to exploit the characteristics of both (R, S) and (s, Q) policies, followed
by continuous-review (s, Q ∗ ) because it is able to adapt to dynamic changes in
consumer demand, followed by the existing policies. To the best of our knowledge,
it is perhaps for the first time a comparative study on the performance of the hybrid
A Comparative Study on Classical and New Hybrid … 21

and classical order policies are undertaken by considering deterministic demand over
a finite planning horizon.

Acknowledgements The support from Alexander von Humboldt Foundation and the University
of Passau is gratefully acknowledged. The authors are grateful to the reviewers and editors for their
valuable feedback that have helped us improve the earlier version of the work.

References

Daniel, J. S. R., & Rajendran, C. (2006). Heuristic approaches to determine base-stock levels in
a serial supply chain with a single objective and with multiple objectives. European Journal of
Operational Research, 175, 566–592.
Ganeshan, R., Jack, E., Magazine, M. J., & Stephens P. (1999). A taxonomic review of supply
chain management research. In Quantitative models for supply chain management (pp. 839–879).
Kluwer Academic Publishers
Gruen, T. W., Corsten, D. S., & Bharadwaj, S. (2002). Retail out-of-stocks: A worldwide
examination of extent, causes and consumer responses. Grocery Manufacturers of America.
Kiesmüller, G. P., De Kok, A. G., & Dabia, S. (2011). Single item inventory control under periodic
review and a minimum order quantity. International Journal of Production Economics, 133(1),
280–285.
Lee, H. L., & Billington, C. (1992). Managing supply chain inventory: Pitfalls and opportunities.
Sloan Management Review, 33, 65–72.
Manatkar, R. P., Karthik, K., Kumar, S. K., & Tiwari, M. K. (2016). An integrated inventory
optimization model for facility location-allocation problem. International Journal of Production
Research, 54(12), 3640–3658.
Movahed, K. K., & Zhang, Z. H. (2015). Robust design of (s, S) inventory policy parameters in
supply chains with demand and lead time uncertainties. International Journal of Systems Science,
46(12), 2258–2268.
Popp, W. (1965). Simple and combined inventory policies, production to stock or to order?
Management Science, 11(9), 868–873.
Radhakrishnan, P., Prasad, V. M., & Gopalan, M. R. (2009). Inventory optimization in supply chain
management using genetic algorithm. International Journal of Computer Science and Network
Security, 9(1), 33–40.
Sethupathi, P. R., & Rajendran, C. (2010). Optimal and heuristic base-stock levels and review
periods in a serial supply chain. International Journal of Logistics Systems and Management,
7(2), 133–164.
Sethupathi, P. R., Rajendran, C., & Ziegler, H. (2014). A comparative study of periodic-review
order-up-to (T,S) policy and continuous-review (s, S) policy in a serial supply chain over a
finite planning horizon. In U. Ramanathan & R. Ramanathan (Eds.), Supply chain strategies
(pp. 113–152). Issues and Models Springer.
Shang, K. H., & Zhou, S. X. (2010). Optimal and heuristic echelon (r, nQ, T ) policies in serial
inventory systems with fixed costs. Operations Research, 58(2), 414–427.
Silver, E., Pyke, D. F., & Peterson, R. (1998). Inventory management and production planning and
scheduling (3rd ed.). John Wiley and Sons.
Wang, Q. (2011). Control policies for multi-echelon inventory systems with stochastic demand.
Supply chain coordination under uncertainty (pp. 83–108). Springer-Verlag.
Zhou, B., Zhao, Y., & Katehakis, M. N. (2007). Effective control policies for stochastic inventory
systems with a minimum order quantity and linear costs. International Journal of Production
Economics, 106(2), 523–531.
Truthful Information Sharing
in a Multiretailer Supply Chain:
Role of Review Strategies

C. Kavitha, R. K. Amit, and B. Vipin

Abstract Eliciting truthful forecast information from the supply chain partners is
one of the major challenges in a supply chain. In this study, we consider a demand
forecast sharing situation in a supply chain between many retailers and a supplier.
The retailers wish to procure capacity from the supplier before the demand is real-
ized and therefore share forecast with the supplier. Retailers may give an excessively
optimistic demand forecast as they do not pay to establish the capacity at the sup-
plier. We construct review strategies for both the supplier and the retailers to ensure
credible information transmission. We study the repeated game scenario in such a
forecast information sharing. Credibility tests are constructed based on multivariate
statistics, to infer from the history of information available with the supplier. We
prove that, in the repeated game, there exist bounds for the duration of review phase,
credibility index, and discount rate above which truth telling results in a perfect
public equilibrium which is Pareto optimal for all the parties.

1 Introduction

Forecasting plays a crucial role in managing the supply chain efficiently. Forecast
communications are generally costless, non-binding, and non-verifiable. In many
supply chains, upstream members often do not forecast their own demand and they
rely on downstream members to furnish the forecasts. This may be due to unavailabil-
ity of data or lack of accuracy. To ensure abundant supply, the downstream member
often has an incentive to inflate her forecast information. Dissemination of credible

C. Kavitha (B) · R. K. Amit

Department of Management Studies, Indian Institute of Technology Madras,
Chennai 600036, India
e-mail: [email protected]
B. Vipin
Department of Industrial and Management Engineering, Indian Institute of Technology Kanpur,
Kanpur 208016, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 23
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_2
24 C. Kavitha et al.

information is often more crucial in settings wherein the sender of the information
has an incentive to misreport the same. Forecast manipulation in the form of reporting
overoptimistic forecasts is pervasive across industries from electronics and semicon-
ductors to medical equipment and commercial aircraft (Lee et al., 1997; Cohen et al.,
2003). Cisco, a major networking equipment supplier, had to write off $2.25 billion
in excess inventory in 2001 because of inflated customer forecasts (CNET, 2002).
In order to ensure credible information sharing, numerous contracts exist. Ren
et al. (2006) suggest that repeated interactions lead to truthful information sharing.
The authors also state that the repeated sourcing decisions will form implicit contracts
between the supply chain members which is termed as relational contracts. This can
be viewed as the feasible substitute for complex pricing contracts that ensure coordi-
nation in supply chain. Relational contracts are vastly studied in sociology, law, and
economics but are relatively new to supply chain management. According to Helper
and Henderson (2014), General Motors, a firm which was once recognized as the
most successful, suffered a huge market loss with their US market share falling from
62.6% in 1980 to 19.8% in 2009, mainly due to the failure to develop and implement
relational contracts with their suppliers. This is largely attributed to General Motors’
“arms’ length” approach to suppliers, whereas Toyota’s success is attributed to its
“relational procurement.” Thus, relational contracts become mandatory to modern
manufacturing and to survive in the competitive market.
There is an emerging trend among global organizations to shift toward platform
based business models to run their operations. Since majority of their supply chains
are multiagents ones, it becomes crucial to study the information sharing among
the supply chain agents. Owing to the importance relational contracts have gained
in modern operations management, it becomes essential that we study problems
associated with such contracts and ways to overcome them. We address the issue
imperfect monitoring in a build-to-order supply chain with multiple retailers and
supplier. Since the lead time of the product is very high, the supplier has to plan
his capacity well in advance so that he could fulfill the retailers’ orders. To help
supplier plan his capacity, the retailers forecast the demand and share them with the
supplier. Anticipating their demand to be high, there is incentive for each retailer to
over-forecast their corresponding demand as there is no cost for overcasting. Once
the supplier allocates the capacity for the retailers, the final order is placed by the
retailers. The suppliers produce minimum of the retailer’s allocated capacity and
the final orders. This is considered to be a stage game. This game is repeated for
indefinite duration.
Consider a two-tier supply chain where there is a sole supplier of a long lead time
product and multiple retailers who buy from this supplier. We study repeated game
scenario (where a stage game is iterated or repeated number of times). The following
events form a stage game: A stage game starts with retailers privately observing the
value of their demand. The retailers share the forecast to the supplier which is a
rough estimate of demand being high or low. The supplier tries to allocate capacity
that maximizes his expected profit given the forecast. The true demands are then
realized, and the final orders are placed with the supplier. The supplier satisfies the
orders accordingly and generates revenue for each unit of order satisfied. The demand
Truthful Information Sharing in a Multiretailer … 25

distributions of each retailer are known, and it is a common knowledge between the
retailer and the supplier. The demand for each retailer is correlated to the demand of
other retailers and the exact nature of this correlation is unknown. Each retailer is
non-strategic with other participating retailers.
The time is subdivided into a series of review phases and punishment phases.
Credibility tests are constructed based on multivariate statistics, to infer from the
history of information available with the supplier. Throughout the review phase,
each retailer is maintained a credibility index which is either incremented or kept
unchanged based on the results in the credibility tests. If the credibility index exceeds
the predefined threshold value, the retailer passes the review, and a new review phase
is initiated. Otherwise, the game enters the punishment phase in which the supplier
disregards the forecasts shared. After a particular punishment period, a new review
phase is started.
In this research work, we prove that the retailers share the forecast information
truthfully with the supplier notably in the initial period of the review phase and
the supplier allots the optimal capacity for each retailer assuming the information
received from retailer is credible. We show that there exist bounds for the duration of
review phase, credibility index and discount rate above which truth telling results in
a perfect public equilibrium which is Pareto optimal for all the parties. We examine
the sensitivity of the review phase with respect to the changes in the parameters of
the games and provide the justification of the observed behavior.

2 Literature Review

We review the literature on relational contracts and repeated games. Any economic
activity will often span through various organizations and individuals. Agents fol-
low different policies and procedures to achieve their objective. Self-interest and
rationality do not allow them to share common objective (Amit and Mehta, 2010).
Inefficiencies arise in the system due to these conflicting objectives. To overcome
this inefficiency, numerous coordination mechanisms such as contracts, information
technology, information sharing and joint decision making has been discussed in lit-
erature. Myerson (1991) suggests that coordination theory is built from game theory,
where payoffs for participating agents not only depend on their actions alone, but
also on other agents actions.
Contracts can be viewed as one of the governance structure where it prescribes
the rights and duties of each agent for each and every future state of the world. It
should also suggest the actions in case of uncertainty, whereas it is impractical to
state every possible future state of world due to limited time or bounded rationality of
individuals as stated by Hohn (2010). Moreover, in case of indefinite contract dura-
tion, to overcome the dilemma between offering a means of flexibility and providing
a means of commitment to the participating agents (Baird, 1990) suggests to create
long-term contractual relationship. These long-term relationships are often called
relational contracts and are informal quid pro quos between the contracting parties.
26 C. Kavitha et al.

Since the actions performed by agents are not verifiable by non contracting parties,
they cannot be enforced by the third party or the court and has to be self enforcing.
A key feature in many strategic situations like buyer–supplier interaction is that
players interact repeatedly over time. Radner (1986) states that if an organization has
short-lived players, then the Nash equilibrium of the game is Pareto inefficient, i.e.,
there exist payoffs which are greater than the Nash equilibrium payoffs. He further
states that at least two approaches are possible to achieve Pareto optimal payoffs.
The first approach is suggested by Aumann (1967), and it states that agents form
binding agreements to achieve efficient payoffs. The overview and classification of
supply chain contracts can be found in Govindan et al. (2013). Another approach to
achieve efficient outcomes that is very common in literature is that players repeat this
game for indefinite times. By doing so, the Nash equilibrium of the supergame (i.e.,
the repeated game) is efficient. This is put forth as Folk theorem which postulates
that whenever the players are sufficiently patient, any Pareto superior payoff vector
to one period Nash equilibrium payoff can be sustained as the Nash equilibrium of
the repeated game. The assurance of reward and the threat of punishment are the key
drivers that sustain the relationship in a repeated game and ensure desirable behavior
in the present situation.
Information sharing is essential to any supply chain which aims to integrate the
activities of its constituent agents to achieve efficient performance. According to
Chopra and Meindl (2007), information is one of the six drivers of supply chain
management. Information had been identified as a chief efficiency in competitive
markets by Hayek (1945). Terwiesch et al. (2005) report that big retail players such
as Walmart and Best Buy along with their suppliers, have all benefited from collabo-
rative planning, forecasting, and replenishment policies. Among all the information
that is shared and received, forecast information of retailers is of utmost impor-
tance to the supplier. It is with this information that the supplier will plan and build
his capacity to maximize his profit. Several authors, including (Cachon and Fisher,
2000; Lee et al., 2000; Chen, 2003; Iyer and Villas-Boas, 2003), explore the value
of information shared in a supply chain. Gavirneni et al. (1999) discuss the value
of information shared in a capacitated system. Cachon and Lariviere (2001) have
explained how contracts can be used for credible information sharing.
Lack of information has been a roadblock in reaching agreements. Crawford and
Sobel (1982) in their classic paper asserts that sharing of information will lead to
better agreements, but at the same time, it also leads to suspicion that revealing all
the information will result in less advantageous position. There exists a chance that
the other player may misuse the information shared to his advantage leaving the
sender of the information worse off. The authors also state that even a completely
self-interested agent will reveal some information in order to attain benefits. Demand
forecast manipulation has been reported in Lee et al. (1997) and Cohen et al. (2003).
According to BusinessWeek (2001), a similar situation was faced by Flextronics
while receiving forecast information from Ericsson and insists that “forecasts from
telecom and electronics companies are often inflated.” The financial impact of wrong
forecast sharing by American Airlines costed Boeing around 2.7 million dollars as
the supplier in 2004 (WallStreetJournal, 2004). In order to overcome these problems
Truthful Information Sharing in a Multiretailer … 27

with forecast information sharing, there are various contracts that are suggested to
ensure truthful exchange of information.
There are certain situations where an agent takes decisions on behalf of other
agent. In these cases, the actions of one contracting party are not verifiable by the
other contracting agents. This gives rise to the situation of imperfect monitoring
where each player receives an imperfect signal of the strategies and actions of rest of
the players. Hence to coordinate the supply chain, the better informed sender should
send some information to the uninformed receiver. Imperfect monitoring arises due
to asymmetric information, where one party misuses this informational advantage to
exploit the other party (the sender of information manipulates the information to his
advantage). This class of problems is discussed by Abreu et al. (1990) and Fudenberg
et al. (1994). The above scenario is described as cheap talk, as it does not directly
affect the payoffs of the individual players. The term “cheap talk” refers to direct
and costless communication among players. Incentive schemes and monitoring are
used to reduce this type of agency problem. Thus, credibility and credulity of the
information being shared are the key factors in information exchange. Testing the
credibility of the information received is often more ambiguous. The examples on
forecast manipulation fall under this category of cheap talk games.
According to Roberts (2004), for a successful partnership, the threat of punishment
should be such that the affected party must be able to retaliate (i.e., by ending the
relationship). Cooperation is sustained through two means: the larger the gains, the
stronger the ties and the worser the punishment, the better the relationship. The parties
should be more dependent on each other so that the net gains from outside options
is almost negligible. The author states that at any point in time the present value of
cooperation should be greater than present value of defection for a relationship to
continue.
The pioneering work on strategic communication model on information sharing
between agents is described by Crawford and Sobel (1982). They considered a sce-
nario where there are two agents, one of them (Sender) has private information about
a random variable. The sender introduces noise into the signal, i.e., he sends a noisy
estimate of the random variable to the uninformed receiver. The receiver then takes
action based on the information sent by the sender, which affects the payoffs of both
the agents. Therefore, sender’s signal is expected utility maximizing given receiver’s
action. The receiver will update his prior based on sender’s signal using Bayes’ rule
given sender’s signaling strategy. The authors conclude that direct communication
is possible only if the agents’ interests coincide.
Several authors including (Radner, 1985; Stocken, 2000) have discussed simple
strategy pairs which ensure truthful information transmission between the principal
and the agent. These strategy pairs are called review strategies and are characterized
by duration of review phase, penalty phase, and criteria to pass the review. Ren et al.
(2010) devise credibility tests for a single supplier-single retailer forecast information
sharing game and proposed a multiperiod review strategy profile for both the players
to ensure truth telling by the retailer.
Radner (1985) describes review strategies in the context of principal–agent inter-
action with separate discount factors for the principal and the agent. The author finds
28 C. Kavitha et al.

that for all discount factors above some critical values, there are equilibria in review
strategies which yields expected utilities greater than one period expected utilities for
the principal and the agent, respectively. In addition, he asserts that if the principal
stops paying the reward as promised, it initiates the punishment phase. The agent is
shown to optimize myopically in this phase until the principal reverts to the constant
reward function. Hence, the author finds optimal review strategy profiles for both the
principal and agent in the supergame.
Stocken (2000) describes a situation where a corporate manager observes imper-
fect information on the nature of environment as favorable or unfavorable for invest-
ment. He then reveals it to the investor, who has to decide on whether to fund the
project or not. This decision is based on the following information: the information
given by the corporate manger and the credibility of the corporate manager. After this
stage of the game, the accounting report is released and is publicly observable. The
investor then observes the accounting report and updates his belief on the corporate
manager and thereby evaluates his credibility. To motivate the manager to voluntar-
ily disclose the information, the manager’s payoff is assumed to be proportional to
the expected contribution on the project. The author found that in equilibrium the
manager truthfully discloses the information on environment in the review phase,
whereas in punishment phase he does not do so.
Rubinstein (1979) studies optimal conviction policies for offenses that have been
committed not deliberately. He designed review strategies to ensure that taxpayers do
not conceal some part of their respective incomes to the tax authorities to evade taxes.
Penalties are imposed in such a way that tax authorities can distinguish between
deliberate tax evasion and offenses that have committed by accident. The author
suggested that there exists a pair of strategies for taxpayer and the tax authorities in
a repeated game which are jointly optimal.
It is now clear that the partners in a supply chain often rely on informal or unwritten
agreements, in addition to contracts. Typically, a supplier must invest in capacity far
in advance of the selling season. For example, semiconductor equipment industries
commit to prior production capacity before having final orders in hand. It is well
documented that the downstream buyer gives the forecast of the demands in terms
of “soft orders.” With this information, the supplier builds his capacity as reported
in Terwiesch et al. (2005).
The literature on relational contracts does not take into account a realistic case
where there is a supply chain with multiagents at a specific level. Motivation for this
works comes from the fact that most of the multinational companies suffer from huge
loss due to incorrect forecasts. With this work, it becomes increasingly helpful for
the top management of companies which primarily relies on relational procurement,
to guarantee credible information sharing by its downstream supply chain members.
In this study, we consider a demand forecast sharing situation in a supply chain with
multiple retailers and a single supplier. There is no direct communication between
the retailers. Due to information asymmetry between the supplier and the retailers,
the supplier does not have sufficient information to forecast the demand as accurately
as the retailers. Since the demand is a function of the random state of nature, once the
final order is received supplier may wish to find the reason behind demand mismatch
Truthful Information Sharing in a Multiretailer … 29

(of forecast and final order), whether it is due to nature of environment or intentional
lie on the part of the retailers. The retailer may give an excessively optimistic initial
demand forecast so as to induce him to build more capacity. The retailers do not pay
to establish the supplier capacity, and he prefers more capacity, anticipating demand
to be high. When the final order is low, it affects the supplier’s profits in the form
of excess inventory. We study the review strategies in such a setting under relational
contract for truthful information sharing.

3 Model Background

The practice of forecast information sharing by the customer to the supplier and
subsequent supply chain coordination using a game-theoretical model in the presence
of information asymmetry was studied by Ren et al. (2010). The retailer being close
to the market is able to forecast the demand more accurately. The customer wishes to
acquire a part of capacity from the supplier before the actual demand is realized by
him. The demand is assumed to be a scaled random variable θ.X , where X ∼ N (μ, σ )
which is positive. θ is assumed to be a random variable which take only two values
either high (θh ) or low (θl ). The customer then forecasts the demand to be high or
low based on market conditions. Let,

Probability(θh ) = α
Probability(θl ) = 1 − α, where α ∈ (0, 1)

θ is assumed independent of X . The distributions of θ and X are common knowledge

among agents. The expected utilities for the supplier and the buyer are as follows:

u(θi , K) = r min(K, Di ) − h(K − Di )+ − cK (1)

v(θi , K) = (p − r)min(K, Di ) − g(K − Di )+ (2)

where K is the supplier’s capacity, c is the supplier’s unit capacity cost, r is the
price the customer pays for unit capacity allocated and utilized, and p is the per unit
revenue for the customer. Overage cost for the supplier is h, whereas g is the lost
sales for the customer.
He then reveals it to the supplier (m = L or H ) who makes capacity allocation
accordingly. The supplier can honor the information given by the customer (where he
will allot the system optimal capacity to the customer) or betray the customer (where
supplier will allot optimal capacity given no truthful information is shared) with the
belief that information shared is untrue and hence omits it. If m = H (m = L) the
supplier believes the customer that demand is DH (DL ), then he will allocate KH (KL )
that maximizes his expected utility given by
30 C. Kavitha et al.

KH = argmaxE[u(θH , K)] (3)

K
KL = argmaxE[u(θL , K)] (4)
K

whereas if the supplier does not believe the customer, then he will allocate K0 as if
there was no meaningful information on demand.

K0 = argmax(1 − α)E[u(θH , K) + (α)[u(θH , K)]] (5)

K

The real demand is then realized, and final order is placed. At this point in time, the
supplier will come to know whether the forecast which the customer shared matches
with the final order placed by the customer. Subsequently, the customer will come
to know whether capacity allocated was optimal according to his forecast.
In a single stage version of the game, the customer will anticipate high demand
even when the forecast is low and tend to inflate demand forecast. Foreseeing the
customer will inflate the demand, supplier does not allocate system optimal capacity.
Thus, Pareto optimal capacities are not reached because of misaligned incentives in a
one-shot interaction. In case of repeated interaction between the supply chain agents,
the author describes two strategies:
• Trigger strategy: Credibility check is triggering, and punishment starts in subse-
quent phase.
• Review strategy: To ensure that supplier is not punishing a truth telling customer (as
the demand is a random variable and could not be accurately predicted), authors
have designed review strategies. In the first R review phases, the supplier will
maintain a scorecard It and a threshold for credibility γt . At the end of each time
period between t=0 and t=R, once the demand is realized and orders placed, the
supplier will evaluate the credibility of the customer through a series of tests and
updates the score It .
The credibility tests are explained as follows:
1. If the customer reports the demand to be high, i.e., mt = H at any time period
t ∈ [1, R], then the customer’s score is subject to two credibility tests.
a. Proportion of periods test: This test evaluates the proportion of time the cus-
tomer has reported mt = H . The probability of high demand happening is
1 − α in each period, so in long run the proportion of times high demand is
realized is approximately 1 − α. Let Nt be the proportion of periods customer
forecasted and declared high demand √ so far in the review phase. The sam-
pling distribution of α follows N (α, α(1 − α)/(t − Nt ). Therefore
√ for 95%
confidence level Nt should not be greater than (1 − α) + zα σ α(1 − α)/Nt .
This test is performed to discourage the customer to over-forecast and sub-
sequently report high demand.
b. Order quantities test: If the customer reports the demand to be high, i.e.,
mt = H at any time period t ∈ [1, R], then the customer’s score is subject to
Truthful Information Sharing in a Multiretailer … 31

another test of truthfulness. If the customer forecasts

√ truthfully, then sample
mean of demand realizations follows N (θH μ, θH σ ). Therefore, the lower
bound or threshold for the demand values at high demand periods is

d H ,t = θH μ − zH σ θH /Nt

If the customer passes both the above tests, then his score It = It−1 + 1,
otherwise the score remains unchanged It = It−1
2. If the customer reports the demand to be low, i.e., mt = L at any time period
t ∈ [1, R], then the customer’s score is subject to only one test of truthfulness. If
the customer
√ orders truthfully, then sample mean of demand realizations follows
N (θL μ, θL σ ). Therefore, the upper bound or threshold for the demand values
at lower demand periods is

d L,t = θL μ + zL σ θL /(t − Nt )

If the customer clears the above test, then his score It = It−1 + 1, otherwise
It = It−1
While in review phase, the customer is always trusted and supplier builds the
system optimal capacity. In case the agent passes the review, then the review phase
is restarted. Whereas, if the customer fails the review, then for M subsequent periods
the supplier will allocate optimal capacity given no truthful information is shared.
Hence after R+M phase, same process is repeated. Thus authors conclude by saying,
in repeated game more efficient equilibrium is achieved through truth telling, leaving
both supplier and customer better off compared to the one shot game.
As discussed earlier, most of the real-world supply chains will have multiple
retailers buying from a single supplier. When we try to model such case, the credibility
tests devised as above may not be relevant. There are multiple sources of information
from the retailers, and the supplier wishes to construe the actual forecast scenario
from these sources. Therefore, we resort to multivariate statistical models in place
of uni-variate models which are discussed in Ren et al. (2010). We try to implement
Hotelling’s T 2 methodology of hypothesis testing for formulating the credibility tests
which is discussed in the following section.

4 The Model

Consider a 2-level supply chain where there is a sole supplier of a long lead time
product and m retailers who buy from this supplier. The retailers owing to the prox-
imity to their customers will have initial forecast of the demand for the product
during certain time period. The supplier do not have access to the market informa-
tion, and he relies on the retailer for the forecast. Since the retailers want to acquire
some capacity from the supplier before the actual demand is realized, they share this
32 C. Kavitha et al.

forecast information to the supplier. The supplier then plans and builds the capacity
accordingly. The demand for the product for each retailer is in some way correlated
to the demand of the other retailer.
We use similar notations as in Ren et al. (2010). Let the demand for each retailer
be Di , where i = 1 to i = n refers to Retailer-1 to Retailer-n, respectively. Let the
demand be a random variable Di = θi,j . The retailer’s message to the supplier is
mi , and it is the demand size parameter, referring to demand being low and high,
respectively. It is assumed that probabilities of high and low demands for each retailer
i are known and the distribution of θ is common knowledge. Table 1 provides the
notations used in this work. In order to model a truthful forecast information sharing
situation involving many retailers, we resort to multivariate statistical models.
We model a situation with m retailer (m ≥ 2). Let us assume that we are on the
nth time period of the review phase. Taking into account the forecast history of m
players, we have n × m matrix of forecast data, where the rows represent the n time
periods and the columns denote m retailers. Therefore, the forecast matrix F if given
by: ⎡ ⎤
f11 f12 f13 . . . f1m
⎢f21 f22 f23 . . . f2m ⎥
⎢ ⎥
F =⎢ . . . . . ⎥, fjk ∈ IR (6)
⎣ .. .. .. . . .. ⎦
fn1 fn2 fn3 . . . fnm

In the above matrix, j represents the n-th time period and k represent the m-th retailer.
With this information furnished by the retailer, the supplier will try to allocate
the capacity to each retailer that maximizes his expected profit. After the supplier
allots specific capacity, the actual demand is realized by the retailer. In order to fulfill
the demand, the retailers place the final order with the supplier. This information is
represented as the order quantity matrix which is also n × m and is denoted by X:
⎡ ⎤
x11 x12 x13 ... x1m
⎢x21 x22 x23 ... x2m ⎥
⎢ ⎥
X =⎢ . .. .. .. .. ⎥ , xjk ∈ IR (7)
⎣ .. . . . . ⎦
xn1 xn2 xn3 . . . xnm

In order to find the defecting retailer across each period, we define a matrix
=(F − X ). This matrix will reveal those retailers who have over-forecasted or
under-forecasted.
⎡ ⎤
11 12 13 . . . 1m
⎢21 22 23 . . . 2m ⎥
⎢ ⎥
=⎢ . .. .. . . .. ⎥ , jk ∈ IR (8)
⎣ .. . . . . ⎦
n1 n2 n3 . . . nm

Let  denote the average of the differences across time periods for each retailer.
Truthful Information Sharing in a Multiretailer … 33

Table 1 Notations
Notations Explanations
ci Capacity cost per unit for the supplier
ri Price which retailer i pays to get one unit allocated and utilized
pi Unit revenue retailer i gets by selling the product
hi Unit overage cost for supplier for capacity allocated and not utilized
gi Unit loss of goodwill cost for retailer when capacity is not sufficient to satisfy
the realized demand
u Utility or profit function of the supplier
vi Utility or profit function of the retailer i
mi Forecast message sent by retailer i to the supplier
Ki,j Capacity allocated by the supplier for the retailer i for demand state j
vic One stage expected payoff for retailer i with system optimal capacity
vi0 One stage expected payoff for retailer i under non-cooperation, i.e., ignoring
retailers’ forecast
vi (I0i ) Normalized and discounted expected profit at the beginning of the game for
retailer i
vi (IRi i −ni ) Normalized expected profit when there are ni periods remaining in the review
phase Ri
vi (θil ) Expected one stage payoff when retailer i observes θil and defects
βi Conditional probability that retailer i is assessed to be truthful given he is
actually truthful
ηi Conditional probability that retailer i is assessed to be truthful given he defects
i vi (θil ) − vi (θil , Kilc )
τ and ρ Review size parameters

 = 1 2 3 . . . m , k ∈ IR (9)

Thus, we could test this matrix ij using Hotelling’s T 2 method. We have p retailers
which we take as variables and n time units. The following subsection explains the
procedure to conduct Hotelling’s T 2 test.

4.1 Hotelling’s T 2

The following are the steps involved in Hotelling’s T 2 .

• Let us define the null hypothesis that the mean vector is zero.

H0 : μ = 0
H1 : μ = 0
34 C. Kavitha et al.

T
• The test statistic which we need to calculate is T2 = n ×  × S−1 × 
• We reject the null hypothesis at 95% confidence if

(n − m − 1)
T2 > F(m−1),(n−m−1),95% × (10)
(n − 1)(m − 1)

4.2 Confidence Ellipsoids

The above test helps us to understand whether the given mean vector is equal to the
hypothesized mean vector. But our objective is to find which variable mean or means
in the vector has contributed to the rejection of null hypothesis. This would help the
supplier to find and punish the deviating retailer. In order to find that, confidence
ellipsoids are constructed around the mean. If the mean vector lies outside the con-
fidence ellipsoid, then we reject the null hypothesis. We could plug in appropriate
values, in order to find the deviating retailer. The equation for confidence ellipsoid
is
 p(n − 1)
(X − μ) S −1 (X − μ) ≤ Fp,n−p,95% (11)
n(n − p)

Therefore, generalizing for p retailers, the confidence ellipsoid can be written as

p p
p(n − 1)
a11 (X 1 ) + a22 (X 2 ) + ... + app (X p ) + 2
2 2 2
aij X i X j ≤ Fp,n−p,95%
i=1 j=1
n(n − p)
(12)
If the hypothesized mean does not satisfy this equation, then we reject the null
hypothesis. However, we wish to get simultaneous confidence intervals for p retailers,
which is given by

p(n − 1)  p(n − 1) 
Xi − Fp,n−p,95% Sii /n ≤ μi ≤ X i + Fp,n−p,95% Sii /n
n(n − p) n(n − p)
(13)
The above equation will give the confidence interval for each of the p retailers. If
the hypothesized difference lies in the confidence region, then the credibility index
is incremented by one. If it does not lie in the confidence range, then the score is not
incremented.
To summarize, the review phase works as follows:
• At any time period in the review phase t, the retailers owing to proximity to
customers will share the demand forecast with the supplier. This is done in order
to help the supplier to build his capacity.
• The supplier plans and builds his capacity accordingly.
• Once the final demand is realized, the retailers place the order with the supplier.
Truthful Information Sharing in a Multiretailer … 35

• The above-mentioned credibility check is performed at each stage, and the retail-
ers’ credibility index is updated.
• The updated index is compared with the moving threshold given by

γt = ((R.S − J )/R)t

where, J = τ Rρ , τ > 0 and 0.5 < ρ < 1.

• If the retailer’s index is less than the threshold, the review phase is terminated
and punishment phase is started for this retailer. The punishment phase extends to
M = XR where, X > 0. While in punishment phase, the forecast of the corre-
sponding retailer is ignored and the system optimal capacity given no truthful
forecast shared is only allocated for the retailer irrespective of the forecast. The
review phase is terminated for the rest of the retailers, and a new review phase is
started.
• If the retailer’s index exceeds the threshold, yet less than the overall threshold
given by
γ = R.S − J

t is incremented and the whole procedure is repeated. If the index of any one retailer
exceeds the overall threshold, then review phase is terminated for all the retailers
and a new review phase is started. This is done to discourage those retailers who
have acquired enough credibility index and can coast for rest of the time periods
in the current review phase.

5 Results

Theorem 1 For every retailer, ∃ a discount rate δ, a review phase of length Ri and
a credibility assessment value S such that for ∀δ > δ, Ri > Ri , S > S, > 0,
there exists a perfect public equilibrium of the repeated game in which vi (I0i ) > vic −
, provided the efficient capacity levels Kijc are Pareto improving for all the parties
compared to Kin . In the equilibrium, the retailers always share truthful forecast and
the supplier always allocate the system optimal capacity Kijc
Proof The outline of the main idea for proof of the theorem is as follows:
1. The players do not have strict incentives to deviate from their designated equi-
librium strategies.
2. In Lemma 2, it is shown that the retailer i will fail the review without doubt at
any time ti < Ri , when Itii < max(0, ti − (Ri − qi )).
3. In Lemma 3, it is proved that the probability the players fail the review when
they use prescribed review strategies is less than the likelihood of retailer failing
when he is truth telling and is evaluated at the end of review phase Ri only.
4. We also prove that the characterized equilibrium ensures almost complete reve-
lation of the private information of each retailer.
36 C. Kavitha et al.

Lemma 1 Assuming the retailer i is always truth telling, then

ni −1 ni
vi (IRi −n ) = (1 − δ)vic (Sδ)k + vi (I0i )(Sδ)ni + [(1 − δ Mi )vi0 + δ Mi vi (I0i )] (1 − S)δ k S k−1
i i
k=0 k=1
(14)
Proof When ni = 1, then

vi (IRi i −1 ) = (1 − δ)vic + δSvi (I0i ) + δ(1 − S)[(1 − δ Mi )vi0 + δ Mi vi (I0i )] (15)

By induction, for ni

vi (IRi i −ni ) = (1 − δ)vic + δSvi (IRi i −(ni −1) ) + δ(1 − S)[(1 − δ Mi )vi0 + δ Mi vi (I0i )]
(16)
By repeated substitution, we get

ni −1 ni
vi (IRi i −ni ) = (1 − δ)vic (Sδ)k + vi (I0i )(Sδ)ni + [(1 − δ Mi )vi0 + δ Mi vi (I0i )] (1 − S)δ k S k−1
k=0 k=1
(17)

Lemma 2 The retailer would invariably fail the review phase if Itii < max(ti − (Ri −
qi ), 0), specifically when Itii is such that the likelihood of the retailer failing the review
is 1.

Proof To prove the above Lemma, we have to show that there exist parameters of the
game where the retailer will truthfully reveal his private information when IRi i −ni =
qi − ni For any Ri , when IRi i −ni = qi − (ni − 1), then the retailer has incentive to
always tell the truth whenever

(1 − δ)vi (θil ) + δηi vi (IRi i −(ni −1) ) + (1 − ηi )[δ(1 − δ Mi )vi0 + δ Mi +1 vi (I0i )] ≤

Mi +1
(1 − δ)vi (θil , Kilc ) + δβi vi (IRi i −(ni −1) ) + (1 − βi )[δ(1 − δ Mi
)vi0 +δ vi (I0i )]

Rearranging,

(βi − ηi )
i ≤ δ[vi (IRi i −(ni −1) ) − (1 − δ Mi )vi0 + δ Mi +1 vi (I0i )] (18)
(1 − δ)

Substituting equation (17) in equation (18) yields

ni −2 ni −1

(βi − ηi )
i ≤ δ (1 − δ)vic (Sδ)k + vi (I0i )(Sδ)ni −1 + [(1 − δ Mi )vi0 + δ Mi +1 vi (I0i )]( ((1 − S)δ k S k−1 ) − 1)
(1 − δ)
k=0 k=1

Using L’Hopital rule as δ approaches 1,

Truthful Information Sharing in a Multiretailer … 37

ni −2
(βi − ηi ) 
i ≤ [(−1)vic (S)k + (ni − 1)vi (I0i )(S)ni −1 + vi (Ioi )S ni −1 +
(−1)
k=0
ni −1

(−Mi vi0 + Mi vi (I0i ) + vi (Ioi ))( ((1 − S)S k−1 ) − 1) +
k=1
ni −1
vi (I0i ) K(1 − S)(S)k−1 + vi (I0i )S ni −1 +
k=1
ni −1
( (1 − S)(S)k−1 − 1)vi (I0i )] (19)
k=1

We know that
ni −1  
1 − Sin − ni S ni −1 (1 − S)
K(S)k−1 = (20)
(1 − S)2
k=1

ni −1  
1 − (S)ni −1
(S)k−1 = (21)
(1 − S)
k=1

Using the above equations in Eq. (20) and simplifying,

  
1 − S ni −1 ni −1
i ≤ (βi − ηi ) (vic − vi (I0i )) + Mi S (vi (I0i ) − vi0 ) (22)
1−S

When n is very large, then

  
1
i ≤ (βi − ηi ) (vic − vi (I0i )) (23)
1−S
 
From the above equation, it is clear that for some δ ≤ δ ≤ 1, ∃ S and Ri such that
  
 1
for all S ≥ S and Ri ≥ Ri , (βi − ηi ) (vi − vi (I0 )) will exceed δi . The
c i
1−S
retailer’s incentive is violated when Iti < max(0, ti − (Ri − qi )).
i

Lemma 3 The probability the retailers fail the review phase when the agents use
the prescribed review strategies is less than or equal to the likelihood of failing the
review when the retailer reports truthful forecasts and is assessed only at the end
of the review phase. Therefore, the retailer will fail the review if IRi i ≤ qi and when
IRi i ≥ qi , the retailer will pass the review.

Proof We can show that there exist histories such that the retailer can pass the review
before the Ri , i.e., review period ends but fails when he is assessed only at the end
of review phase Ri .
38 C. Kavitha et al.

Proposition 1 There exists a specified equilibrium which ensures nearly complete

revelation of the each of retailer’s private information.

Proof To show that the specified equilibrium yields practically complete revelation
of private information of each retailer, we need to obtain sufficient conditions for
lower bound of retailer’s normalized expected profit. Since the review phase can be
terminated before Ri , assume that if the retailer fails the review, he fails at the earliest
possible time ti and suppose he passes the review, he passes at the earliest possible

time t . We know that vi (I0i ) ∈ [vi0 , vic ]. By the Markovian property of the repeated
game, we find the lower bound for the retailer’s normalized discounted expected
profit vi (I0i )

Ri −qi

Ri −qi Ri −qi +Mi
vi (I0i ) ≥ λi (1 − δ) δ k−1 vic +δ (1 − δ Mi
)vi0 +δ vi (I0i ) +
k=1
⎡ 
⎤
t

(1 − λi ) ⎣(1 − δ) δ k−1 vic + δ t vi (I0i )⎦ (24)
k=1

Simplifying the above equation, we get


λi vic (1 − δ Ri −qi ) + λi δ Ri −qi (1 − δ Mi )vi0 + (1 − λi )vic (1 − δ t )
vi (I0i ) ≥  (25)
1 − δ Ri −qi +Mi λi − (1 − λi )δ t

As δ → 1 , using L’Hopital rule

Mi vi0 (1 − λi ) c 
(vic + )λi − v (t )
Ri − qi Ri − qi i
 (26)
Mi λi (1 − λi )t
λi + +
Ri − qi Ri − qi

When we substitute Mi = χ Ri ,

M χ
lim =
Ri →∞ Ri − qi (1 − S).

We know that the conditional probability that the retailer fails the review at the
start of the game is λi . The upper bound for this probability is denoted by λi (IRi i <
qi ) = λi (IRi i < Ri S − Ji ). If each retailer reports the demand truthfully, then IRi is a
binomially distributed random variable. Therefore, using Chebyshev’s inequality,

Ri S(1 − S)
λi (IRi i < qi ) =
Ji2
Truthful Information Sharing in a Multiretailer … 39

ρ
Since Ji = τ Ri , when Ri → ∞ =⇒ λi → 0. Therefore, equation (11) simplifies
to vi (I0i ) ≥ vic . This can be rewritten as, vi (I0i ) > vic − . Hence, for any > 0, ∃ a
discount factor δ such that ∀ δ ≥ δ and ∃ a Ri such that ∀ Ri > Ri , vi (I0i ) > vic − ,
which is a lower bound for retailer’s expected profit.

6 Numerical Illustration

In order to illustrate the above model, we give numerical example for how the effective
review length changes with the parameters of the model such as τ , ρ and the maximum
specified review length. We also compare the review length of trigger strategy and
review strategy. The forecast and the demand matrix are obtained by random number
generation. The forecast and the demand are assumed to follow normal distribution
N (100, 75).
We have constructed graph for the effective review length for the retailers when
the maximum review period is changed (Fig. 1a). We find that the effective review
length increases when the maximum review length increases. For increase in the
maximum review period specified, by the threshold equation, the overall threshold
also increases. Hence, the retailers require more time periods to pass the review phase
and move on to the next review phase.
From the graph 1b, we study how the review length changes with τ . We find that
as τ increases, the review length decreases. This is in accordance with our intuition

(a) Maximum review period (b)

(c) (d) Number of retailers

Fig. 1 Impact of different parameters on review length

40 C. Kavitha et al.

that as τ increases the allowable margin increases, thereby decreasing the number of
time periods required to pass the review.
The variation in review length when ρ changes is studied as shown in Fig. 1c. It is
evident that as the ρ increases, then the effective review length decreases. According
to the threshold equation, as ρ increases, the number of time periods required to pass
the review phase and restart the next review decreases.
It is quite interesting to study how the review phase changes with increase in the
number of retailers (Fig. 1d). Our intuition is that as the number of retailers increases,
the supplier may get more information of the same demand scenario, so that he can
easily detect the defecting retailers. But, the graph is seen to increase with the number
of retailers. This can be attributed to the fact that, as the number of retailers increases,
there increased distortion to the forecast information. Hence, the supplier may not
be able to detect the truth telling as against the defecting retailer in lesser time frame.
Hence, the review phase increases with the increase in number of retailers.

7 Managerial Implications and Conclusions

Manipulation of demand forecasts has mostly been observed whenever there are
multiple agents in a supply chain. A relevant case is the cloud kitchen inventory
pooling promoted by delivery platforms Swiggy and Zomato. There is a single kitchen
which houses cooking facilities of competing restaurants. They agree to pool their
inventories but may not always be willing to share the actual demand information. The
problem of inflated forecasts is especially relevant in the current pandemic scenario
as the uncertainties in supply and demand create inconsistent orders in supply chain.
AmerisourceBergen, a pharmaceutical distributor in the USA, experiences an inflated
demand from the acute healthcare providers for increasing their inventory (Zenk,
2020; FiercePharma, 2020). Our review strategies can address the challenges in
information sharing in such scenarios.
This research work studies the demand forecast sharing scenario between a single
supplier of a long lead time product and multiple retailers who buy from the supplier.
We develop review strategies for both the supplier and the retailers to ensure credible
information transmission. Credibility tests for the retailers which guarantees truthful
forecast sharing are developed. We perform sensitivity analysis with respect to the
parameters in the repeated game. Though these strategies can achieve credible fore-
cast sharing in a multiretailer setting, increasing the number of retailers results in
increased review length—an undesirable feature, observed in our study. This result
warrants the use of relational contract as an alternative to the classical mechanism
design approach in a supply chain setting with high number of agents for truthful
information sharing.
This work aims to provide insights into the review strategies and significance
of them in real-life supply chain to ensure truthful forecast sharing. We consider
repeated interactions among the agents in the form of a relational contract in an
effort to ensure credible forecast sharing. A repeated and long-term relational contract
Truthful Information Sharing in a Multiretailer … 41

between supply chain parties gives each agent “opportunities to review the credibility
of the other party, reward truth telling, punish otherwise, and therefore provides the
right incentive for truthful information sharing” as asserted by Ren et al. (2010). This
repeated interactions resulting in relational contract between supply chain agents can
be considered as a feasible alternative to traditional approaches in ensuring truthful
information sharing.

Acknowledgements We thank the editors and two reviewers for their comments and suggestions
which helped us improve the quality of the manuscript.

References

Abreu, D., Pearce, D., & Stacchetti, E. (1990). Toward a theory of discounted repeated games with
imperfect monitoring. Econometrica, 58(5), 1041–1063.
Amit, R. K., & Mehta, P. (2010). Repeated contracts for supply chain coordination under asymmetric
information: A review and future directions. Technology, Operations and Management, 1(2), 31–
36.
Aumann, R. J. (1967). A survey of cooperative games without side payments. In Essays in Math-
ematical Economics in Honor of Oscar Morgenstern (pp. 3–27). Princeton, NJ: Princeton Univ
Press.
Baird, D. G. (1990). Self-interest and cooperation in long-term contracts. Journal of Legal Studies,
19, 583–596.
BusinessWeek. (2001). The inventory mess—faulty forecasts. https://www.bloomberg.com/news/
articles/2001-03-18/table-the-inventory-mess. Accessed: 18/11/2020.
Cachon, G. P., & Fisher, M. (2000). Supply chain inventory management and the value of shared
information. Management Science, 46, 1032–1048.
Cachon, G. P., & Lariviere, M. A. (2001). Contracting to assure supply: How to share demand
forecasts in a supply chain. Management Science, 47(5), 629–646.
Chen, F. (2003). Information sharing and supply chain coordination. In S. Graves, T. de Kok
(Eds.), Supply chain management: Design, coordination and operation. Handbooks in Oper-
ations Research and Management Science (Vol.11, pp. 341–421). North-Holland: Elsevier.
Chopra, S., & Meindl, P. (2007). Supply chain management. Strategy, planning & operation. Gabler.
CNET. (2002). Cisco’s 2.25 usd billion mea culpa. https://www.cnet.com/news/ciscos-2-25-
billion-mea-culpa/. Accessed 18/11/2020.
Cohen, M. A., Ho, T. H., Ren, Z. J., & Terwiesch, C. (2003). Measuring imputed cost in the
semiconductor equipment supply chain. Management Science, 49(12), 1653–1670.
Crawford, V., & Sobel, J. (1982). Strategic information transmission. Econometrica: Journal of the
Econometric Society, 1431–1451.
FiercePharma. (2020). How do allocations work for the pharma supply chain?. https://
www.fiercepharma.com/pharma/how-do-allocations-work-for-pharma-supply-chain. Accessed
17/11/2020.
Fudenberg, D., Levine, D. K., & Maskin, E. (1994). The folk theorem with imperfect public infor-
mation. Econometrica, 62(5), 997–1039.
Gavirneni, S., Kapuscinski, R., & Tayur, S. (1999). Value of information in capacitated supply
chains. Management Science, 45(1), 16–24.
Govindan, K., Popiuc, M. N., & Diabat, A. (2013). Overview of coordination contracts within
forward and reverse supply chains. Journal of Cleaner Production, 47, 319–334.
Hayek, F. (1945). The use of knowledge in society. American Economic Review, 35(September),
519–30.
42 C. Kavitha et al.

Helper, S., & Henderson, R. (2014). Management practices, relational contracts, and the decline
of general motors. National Bureau of Economic Research.
Hohn, M. I. (2010). Relational supply contracts. Lecture notes in economics and mathematical
systems, 629.
Iyer, G., & Villas-Boas, J. M. (2003). A bargaining theory of distribution channels. Journal of
Marketing Research, 40(1), 80–100.
Lee, H. L., Padmanabhan, V., & Whang, S. (1997). Information distortion in a supply chain: The
bullwhip effect. Management Science, 43(4), 546–558.
Lee, H. L., So, K., & Tang, C. (2000). The value of information sharing in a two-level supply chain.
Management Science, 46(5), 626–643.
Myerson, R. (1991). Game theory: Analysis of conflict. Harvard University Press.
Radner, R. (1985). Repeated principal-agent games with discounting. Econometrica, 53(5).
Radner, R. (1986). Repeated partnership games with imperfect monitoring and no discounting.
Review of Economics Studies, L111, 43–57.
Ren, Z. J., Cohen, M. A., Ho, T. H., & Terwiesch, C. (2006). Sharing forecast information in a
long-term supply chain relationship. Operations Research.
Ren, Z. J., ACM, Ho, T. H., & Terwiesch, C. (2010). Information sharing in a long-term supply
chain relationship: The role of customer review strategy. Operations Research, 58(1), 81–93.
Roberts, J. (2004). The modern firm: Organizational design for performance and growth. Oxford
University Press.
Rubinstein, A. (1979). An optimal conviction policy for offenses that may have been committed by
accident. In Applied game theory (pp. 406–413).
Stocken, P. C. (2000). Credibility of voluntary disclosure. RAND Journal of Economics, 31(2),
359–374.
Terwiesch, C., Ren, Z. J., Ho, T. H., & Cohen, M. A. (2005). An empirical analysis of forecast
sharing in the semiconductor equipment supply chain. Management Science, 51(2), 208–220.
WallStreetJournal. (2004). American airlines defers delivery of 54 boeing jets. https://www.wsj.
com/articles/SB110114437791280976?page=1. Accessed 18/11/2020.
Zenk, H. (2020). The pharma supply chain: Responding to Covid-19. https://www.pharmexec.com/
view/pharma-supply-chain-responding-covid-19. Accessed 17/11/2020.
Mathematical Models and Heuristics
for an Inventory Routing Problem
Without Split Deliveries

K. R. Preethi, Chandrasekharan Rajendran, and S. Viswanathan

Abstract We consider a finite horizon Inventory Routing Problem in which the

vendor monitors the inventory at the retailers and makes replenishment decisions for
them. A single vendor manages several retailers who face deterministic but dynamic
demands. The vendor decides the delivery schedule, delivery quantities, and vehicle
routes that minimize the total supply chain cost. For practical purposes, we assume
that split deliveries to a retailer are not allowed and that the backorders are allowed
at the retailers. Two mixed integer linear programming models are proposed to solve
the specified Inventory Routing Problem (IRP). Since IRP is NP-hard, two heuristics
are also developed for the same problem. Numerical experiments were conducted on
randomly generated data instances to evaluate the proposed models. The numerical
study reveals that the proposed models are computationally efficient. We observed
that the optimality gap of the solutions from the mathematical models was no more
than 3% in all the tested cases. We also show that the proposed heuristics provide
consistent near-optimal solutions. The proposed models and heuristics could serve
as efficient decision-making tools to help supply chain managers make quick tactical
level decisions.

Keywords Supply chain · Logistics · Vehicle routing · Inventory routing · No split

delivery · Integer program · Heuristics

1 Introduction

Vendor managed inventory (VMI) is a supply chain coordination mechanism in

which a vendor monitors and manages the inventory at the retailers. In a traditional

K. R. Preethi (B) · C. Rajendran

Indian Institute of Technology, Madras, India
C. Rajendran
e-mail: [email protected]
S. Viswanathan
Nanyang Technological University, Nanyang Ave, Singapore
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 43
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_3
44 K. R. Preethi et al.

order-based replenishment, the vendor fulfils the retailer orders that she receives. In
VMI, the retailers cooperate with the vendor. The vendor has access to the demand
and inventory information of the retailers using electronic data interchange (EDI).
The vendor then makes replenishment decisions for the retailers such that the total
supply chain cost is minimized.
In a VMI system, the vendor can consolidate and coordinate multiple deliveries
to the retailers, thus minimizing transportation costs. Here, the retailers also benefit
from not allocating resources for inventory management (Coelho et al., 2013). For
a detailed discussion on the benefits of VMI, see Waller et al. (1999). Proctor &
Gamble and Walmart successfully implemented VMI in their distribution networks
in the 1980s. Companies are increasingly adopting VMI due to advancements in data
analysis techniques and information system technologies.
The vendor faces three main decisions in a VMI setting (Campbell et al., 1998):
When to serve a customer? How much to deliver? Which delivery routes to follow?
The vendor must simultaneously determine the delivery schedule, the delivery
quantities, and the vehicle routes. This combined inventory and distribution problem
is termed as the Inventory Routing Problem (IRP). It achieves a trade-off between
frequent, small-sized deliveries (low inventory costs but high transportation costs),
and large shipments (low transportation but high inventory holding costs).
Since IRP is a modified vehicle routing problem (Campbell & Savelsbergh, 2004),
it is NP-hard. Industrial applications are driven by the ability to generate good quality
solutions rapidly. Therefore, solving real-life sized instances in a computationally
efficient manner is very important.
In our paper, we propose two efficient mixed integer linear programming (MILP)
models to find optimal solutions for the IRP with backorders and no split deliveries.
We also propose two heuristics that give near-optimal solutions to the problem. We
report additional numerical experiments to evaluate the performance of the MILP
models and the heuristics. We ran the MILP models for a maximum of 2 h (7200
secs) on randomly generated data instances. The numerical experiments show that
the proposed MILP models are computationally efficient. The optimality gap is no
more than 3% in all the tested cases. We also show that the proposed heuristics
provide near-optimal solutions. In Sect. 1.1, we discuss some of the literature on IRP
and how our paper is different from the existing papers.

1.1 Literature Review

Infinite horizon IRP determines inventory replenishment and vehicle routing deci-
sions over the infinite horizon for a continuous review system with a constant demand
rate. Replenishment strategies that can be repeated over time are developed to mini-
mize the long-run average costs (Refer to Anily & Federgruen, 1990; Viswanathan
& Mathur, 1997).
In the finite horizon IRP literature, a wide range of solution methodologies (exact
methods and heuristics) has been employed. Some of the papers are discussed below.
Mathematical Models and Heuristics … 45

Archetti et al. (2007) were the first to develop a branch and cut algorithm for a
single product, single-vehicle IRP with two different inventory policies at the retailer.
Coelho and Laporte (2013a) proposed a branch and cut algorithm for solving a multi-
product IRP. They also discussed exact methods for three different variants of IRP
(Coelho & Laporte, 2013b). Bertazzi et al. (2002) developed a two-phase heuristic
for a deterministic demand, finite horizon IRP with an order-up to-level inventory
policy at the retailer. Campbell and Savelsberg (2004) developed a decomposition
approach for solving a rolling horizon framework IRP with a constant demand rate.
We note that the papers assume that some kind of inventory replenishment policy
exists at the retailer. This assumption may not be true in practice. However, it reduces
the search space considerably. Also, a periodic review framework with deterministic
but dynamic demands is more relevant for the finite horizon IRP, especially in an
enterprise resource planning (ERP) system. Therefore, we propose a finite horizon
IRP with a periodic review inventory system and deterministic, dynamic retailer
demands. We do not assume any replenishment policy at the retailers.
The papers discussed above also assume that the backorders are not allowed at
the retailers. However, Abdelmaguid et al. (2009) argued that backordering is bene-
ficial when savings in transportation cost can be realized by delaying distribution.
Backorders might become inevitable especially when there are capacity constraints,
Preethi et al. (2018) proposed mathematical models for a finite horizon IRP with
backorders. They showed that their mathematical model performs better than the
model proposed by Abdelmaguid et al. (2009).
Both the above-mentioned papers assume that split deliveries are allowed in the
IRP. However, split deliveries will lead to problems in terms of inventory traceability.
In practice, a single vehicle is sent out for delivery from the supplier’s end keeping
the customer’s operational convenience in mind. To this end, conventional vehicle
routing problems also assume that split deliveries by multiple vehicles are not allowed
(see Laporte, 1992; Toth & Vigo, 2002). Therefore, we consider a finite horizon IRP
with backorders and no split deliveries in our study.
The paper is organized as follows. We discuss the problem specification and the
two proposed MILP model formulations in Sect. 2. The computational experiments
for the optimal MILP models are given in Sect. 3. In Sect. 4, we present the two
proposed heuristics, their computational performance, and solution quality compared
against MILP model 1 from Sect. 2. Section 5 summarizes the contribution of our
work and explores further possible extensions.

2 Problem Specification

We consider a finite horizon, two-echelon, Inventory Routing Problem with deter-

ministic but dynamic retailer demands. A single vendor serves N number of retailers
using V heterogeneous vehicles each with a capacity of Capv . The length of the
discrete planning horizon is T .
46 K. R. Preethi et al.

The sequence of events in any period t is as follows: The vehicles assigned for
delivery are dispatched from the vendor. Retailers that are scheduled for delivery
receive their orders. Then, a demand Dem i,t is realized at each retailer i for the
current period t. The retailer satisfies the demand using on-hand inventory Ii,t . Excess
demands Bi,t are backordered.
Retailer i incurs a holding cost of h i per unit per period and a backordering cost of
πi per unit per period. The vendor incurs a fixed transportation cost of Fv if vehicle
v is used in any period. Without loss of generality, we assume that the variable
transportation cost per unit distance is 1 for all the vehicles. The vendor determines
the routes, delivery quantities, and schedules which minimizes the total supply chain
cost. The supply chain cost is the sum of fixed and variable transportation costs,
inventory holding costs, and backordering costs.

2.1 Mixed Integer Linear Programming Model 1

The parameters of the finite horizon IRP are given in Table 1.

The proposed model consists of integer decision variables and continuous decision
variables. v,t are indicator variables which take a value 1 if vehicle v is used in
period t. We call them the vehicle assignment variables because they contain the
information on which vehicles are dispatched for delivery in each period.
The binary integer variables δi,v,t are equal to 1, if retailer i is served by vehicle v
in period t. They are called retailer allocation variables because they determine the
set of retailers allocated to each vehicle in every period. Also, these variables provide
us with delivery schedules for retailers.
X i, j,v,t are indicator variables associated with the arc between i and j. X i, j,v,t is
equal to 1 if vehicle v visits retailer j immediately after visiting retailer i in period
t. They help us determine the routes taken by the vehicles. Subscript B is used to
denote the vendor node and i = 1, 2, …, N is used to denote the retailer nodes.
We also use dummy nodes to indicate the vendor in this model. There is a total of
2V dummy nodes for the vendor, starting from N + 1 to N + V + V . Imagine that
at the beginning of every period, vehicle 1 is stationed at node N + 1. It leaves from
the node N + 1 when it leaves for distribution. After delivery, vehicle 1 returns to
N + V + 1th node at the end of the period. Similarly, any vehicle ν ∈ 1, 2, . . . V is
stationed at the node N + v at the beginning of a period. It leaves from node N + v
and returns to node N + V + v after the deliveries are made.
Now, we define binary routing variables from and to the vendor node using the
dummy nodes (vendor). x N +v,i,t = 1 means that the vehicle v travels from the vendor

to retailer i in period t. xi,N +V +v,t = 1 means that the vehicle v returns to the vendor
from retailer i in period t.
The dummy nodes break the vendor to vendor circuits/routes into paths. This
enables us to solve a Hamiltonian path problem for the vehicles instead of solving
a Hamiltonian cycle. Consider a delivery route B-1–3-5-B. This is a closed route
that takes more computational effort to solve. The presence of our dummy nodes
Mathematical Models and Heuristics … 47

Table 1 Parameters of the

Subscripts
IRP
i, j Retailers
v Vehicles
t Periods
B The vendor
N + 1 to N + 2V Dummy nodes for the vendor
Notations Definitions (Parameters)
N Total number of retailers
V Total number of vehicles available for delivery in
each period
T Number of discrete periods
Dem i,t Demand at retailer i in period t
Capv The capacity of vehicle v
d j,i Distance travelled to reach retailer i from retailer j
d B,i Distance travelled to reach retailer i from the
vendor
di,B Distance travelled to reach the vendor from
retailer i
hi Inventory holding cost per unit per period at
retailer i
πi Backordering cost per unit per period at retailer i
Ci Inventory storage capacity at retailer i
Fv Fixed cost of operating vehicle v

breaks this circuit into a path between the dummy nodes N + v and N + V + v
where v is the vehicle assigned for delivering in this route. This is illustrated in
Fig. 1. The dummy nodes also prevent the formation of any sub-tours containing the
vendor node. Therefore, formulating the IRP with dummy nodes saves considerable
computational time.
All the decision variables including the inventory-related variables are explained
in Table 2.

Fig. 1 Illustration of dummy nodes extending the routes into paths

48 K. R. Preethi et al.

Table 2 Decision variables used in MILP model 1

Notations Description

v,t 1, If vehicle v is dispatched in period t
0, otherwise

δi,v,t 1, if retailer i is served by vehicle v in period t
0, otherwise

X i, j,v,t 1, if vehicle v visits retailer j immediately after visiting retailer i in period t
0, otherwise
 
x N +v,i,t 1, if vehicle v travels from the vendor to retailer i in period t
0, otherwise
 
xi, N +V +v,t 1, if vehicle v travels from retailer i to the vendor in period t
0, otherwise
Q i,v,t Quantity delivered to retailer i using vehicle v in period t
Ii,t Inventory level at retailer i in period t
Bi,t Number of back ordered units at retailer i in period t
CQi,t The cumulative load carried by any vehicle when it leaves retailer i in period t

CQv,t Overall (cumulative) load carried by vehicle v while leaving the vendor in period t

Minimize Total Supply Chain Cost (Z 1 )


T 
V 
T 
N
 
= Fv v,t + h i Ii,t + πi Bi,t
t=1 v=1 t=1 i=1


T 
V 
N 
N 
T 
V 
N

+ X i, j,v,t di, j + x N +v,i,t d B,i
t=1 v=1 i=1 t=1 v=1 i=1
j =1
j = i

T 
V 
N

+ xi,N +V +v,t di,B (1)
t=1 v=1 i=1

Subject to


V
  V
 
X i, j,v,t + X j,i,v,t ≤ δi,v,t + δ j,v,t /2
v=1 v=1
∀i = 1, 2.., N − 1, j = i + 1, i + 2, .., N , ∀t (2)
Mathematical Models and Heuristics … 49


V
δi,v,t ≤ 1 ∀i, t (3)
v=1


N
δi,v.t ≤ N v,t ∀v, t (4)
i=1


N
X j,i,v,t + x N +v,i,t = δi,v,t ∀i, v, t (5)
j = 1,
j = i


N

X i, j,v,t + xi,N +V +v,t = δi,v,t ∀i, v, t (6)
j = 1,
j = i


N
x N +v,i,t = v,t ∀v, t (7)
i=1


N

xi,N +V +v,t = v,t ∀v, t (8)
i=1

Q i,v,t ≤ Mi δi,v,t ∀i, v, t (9)


N
Q i,v,t ≤ Capv v,t ∀v, t (10)
i=1


V 
T 
T
Q i,v,t ≤ Demi,t ∀i (11)
v=1 t=1 t=1


V
Ii,t − Bi,t = Ii,t−1 − Bi,t−1 − Demi,t + Q i,v,t ∀i, t (12)
v=1

Ii,t ≤ Ci ∀i, t (13)


N
CQv,t = Q i,v,t ∀v, t (14)
i=1


V
CQi,t ≤ maxcap δi,v,t ∀i, t (15)
v=1
50 K. R. Preethi et al.

 
CQi,t ≤ CQv,t − Q i,v,t + maxcap 1 − x N +v,i,t ∀i, v, t (16)

 

V 
V
CQi,t ≤ CQ j,t − Q i,v,t + maxcap 1 − X j,i,v,t ∀i, j, j = i, t (17)
v=1 v=1
 

V

CQi,t ≤ maxcap 1 − xi,N +V +v,t ∀i, t (18)
v=1

where maxcap and Mi are very large numbers (Big M).  They are tight-
ened using the following expressions: maxcap = max Capv , Mi =
 v
T
min t=1 Demi,t , maxcap .
The constraints are explained below. The total supply chain cost consists of fixed
transportation cost, inventory holding, and backordering cost, and variable trans-
portation cost. Equation (1) is the objective function that represents the total supply
chain cost. The routing variables between any two retailers can be active only if both
the retailers are scheduled for delivery in a period (Eq. 2). A retailer is visited by
no more than one vehicle in any period (Eq. 3). A vehicle could visit a maximum
of N retailers in a period (Eq. 4). When a retailer is scheduled for delivery by a
vehicle, one of the arcs that end at the retailer node is active (Eq. 5). At the same
time, one of the arcs starting from the retailer node is active in the solution (Eq. 6).
If a vehicle is scheduled for dispatch in a period, routing variables from and to the
corresponding dummy vendor nodes should be active in that period (Eqs. 7 and 8).
The retailer receives a positive quantity only when it is allocated to a vehicle in a
period (Eq. 9). The total load that a vehicle carries should be less than or equal to
its capacity (Eq. 10). The total quantity that a retailer receives over the entire time
horizon should be less than or equal to the retailer’s total demand in that horizon
(Eq. 11). A set of constraints (Eqs. 2 and 11) are used to tighten the model. Equa-
tion (12) represents the inventory balance constraint. The inventory level at a retailer
should be less than its inventory storage capacity (Eq. 13). Then, we have sub-tour
elimination constraints (Eqs. 14–18).
The sub-tour elimination in this model works in the following manner. The cumu-
lative load is highest when the vehicle leaves the vendor. It keeps decreasing as the
vehicle makes the deliveries and becomes zero when the vehicle returns to the vendor.
So, the cumulative load when the vehicle leaves any given node should be less than
or equal to the cumulative load when the vehicle leaves the preceding node. Consider
a sub-tour 2-4-5-2. According to constraints (14–18), for any such sub-tour to exist,
CQ2,t ≥ CQ4,t ≥ CQ5,t ≥ CQ2,t which is infeasible. Therefore, sub-tours will be
eliminated in the optimal solution.
Mathematical Models and Heuristics … 51

2.2 Mixed Integer Linear Programming Model 2

In this subsection, we propose an alternate MILP model for the Inventory Routing
Problem. We retain vehicle assignment and retailer allocation variables from MILP
model 1. However, we reduce the number of decision variables in model 2 by using
three-dimensional routing variables X i, j,t , unlike the previous model which uses
four-dimensional X i, j,v,t . Therefore, the number of such routing variables becomes
N 2 T instead of N 2 TV. However, we need additional constraints that link the routing
variables X i, j,t with retailer allocation variables δi,v,t (Eqs. 20 and 21), to capture
the complete information on each vehicle’s routes. We have other network flow
constraints (Eqs. 22, 23, 26–29), vehicle capacity constraints (Eq. 31), inventory
balance constraints (Eq. 33), inventory storage capacity constraints (Eq. 34), sub-
tour elimination constraints (Eqs. 35–37), and no split deliveries constraint (Eq. 24)
in this model.
We use cumulative distance (C Di,t ) variables to eliminate sub-tours. The cumu-
lative distance travelled by a vehicle when it reaches a node should be greater
than or equal to the cumulative distance at the preceding node. Therefore, a sub-
tour 2-4-5-2 will be rendered infeasible as the constraints would specify that
CD2,t ≤ CD4,t ≤ CD5,t ≤ CD2,t . The new decision variables in MILP model 2
are given in Table 3.

Minimize Total Supply Chain Cost (Z 2 )

 v
T  
T 
N 
T 
N 
N
= Fv v,t + (h i Iit + πi Bit )+ X i, j,t di, j
t=1 v=1 t=1 i=1 t=1 i=1 j=1


T 
V 
N 
T 
V 
N
+ x N +v,i,t d B,i + 
xi,N +V +v,t d B,i (19)
t=1 v=1 i=1 t=1 v=1 i=1

Subject to

V 
δi,v,t + δ j,v,t
X i, j,t + X j,i,t ≤ ∀ i = 1, 2, .., N − 1,
v=1
2
j = i + 1, i + 2, .., N , ∀t (20)

Table 3 Decision variables in MILP model 2

Notations Definitions (decision variables)

X j,i,t 1, if any vehicle moves from retailer j to retailer iin period t
0, otherwise
CDi,t The cumulative distance travelled by any vehicle when it reaches retailer i in period t
52 K. R. Preethi et al.

δ j,v,t − δi,v,t ≤ 1 − X j,i,t ∀v, t, i, j and j = i (21)

x N +v,i,t ≤ δi,v,t ∀i, v, t (22)


xi,N +V +v,t ≤ δi,v,t ∀i, v, t (23)


V
δi,v,t ≤ 1 ∀i, t (24)
v=1


N
δi,v.t ≤ N v,t ∀ v, t (25)
i=1


N 
V 
V
X j,i,t + x N +v,i,t = δi,v,t ∀i, t (26)
v=1 v=1
j = 1,
j = i


N 
V 
V

X i, j,t + xi,N +V +v,t = δi,v,t ∀i, t (27)
v=1 v=1
j =1
j = i


N
x N +v,i,t = v,t ∀v, t (28)
i=1


N

xi,N +V +v,t = v,t ∀v, t (29)
i=1

Q i,v,t ≤ Mi δi,v,t ∀i, v, t (30)


N
Q i,v,t ≤ Capv v,t ∀v, t (31)
i=1


V 
T 
T
Q i,v,t ≤ Demi,t ∀i (32)
v=1 t=1 t=1


V
Ii,t − Bi,t = Ii,t−1 − Bi,t−1 − Demi,t + Q i,v,t , ∀i, t (33)
v=1

Ii,t ≤ Ci ∀i, t (34)

Mathematical Models and Heuristics … 53


V
CDi,t ≤ MV δi,v,t ∀i, t (35)
v=1


V
 
CDi,t ≥ d B,i x N +v,i,t ∀ i, t (36)
v=1
 
CDi,t ≥ CD j,t + d j,i − MV 1 − X j,i,t ∀i, t, j and j = i (37)

where maxcap, M j and Mv are very large numbers (big M). maxcap and Mi s are
N  
defined as before and MV = i=1 d B,i + max j, j =i d j,i .

3 Computational Experiments

The computational performance of the MILP models is tested on randomly generated

data instances. The retailer node locations are generated randomly on a 20 × 20
square. The vendor is assumed to be located at the centre of the square. The Euclidean
distance between the nodes gives the distance between the nodes. Multiple levels of
N and T are used to create different problem sizes, and three different data instances
are created for each problem size. The retailer demands and inventory cost parameters
are generated from standard distributions while other parameters take fixed values.
The details of the data generation process are given in Table 4.

Table 4 Parameter values used in the computational analysis

Parameters Data used/generated
Initial inventory Ii,0 = 0 ∀i
Initial backorders Bi,0 = 0 ∀i
Retailer demands Demi,t ∼ Uniform(50, 100)∀i, t
Inventory holding costs/unit/period h i ∼ Normal(0.1, 0.02) ∀i
Backordering cost/unit/period πi ∼ Normal(5, 0.5) ∀i
Inventory storage capacity Ci = 120 ∀i
Vehicle capacity Capv = 250 ∀v
Fixed transportation cost Fv = 10 ∀v
Variable transportation cost per unit distance 1 ∀v
Parameters Levels used
Number of retailers N = 5, 6, 7, 8, 9, 10
Periods T = 3, 5
Number of vehicles V =3
54 K. R. Preethi et al.

The proposed MILP models 1 and 2 were run on a computer with Windows 7 OS
and an Intel i5 processor and 8 GB RAM. Each instance was run for a maximum
CPU time of 2 h (7200 s). We have reported the upper and lower bounds obtained
from CPLEX’s implementation of the branch and cut algorithm in Table 5. As the
program runs, the upper and lower bounds converge to an optimum.
The optimality gap is a measure of the relative difference between the bounds and
is defined as (Upper Bound − Lower Bound)/Upper Bound. The algorithm stops and
gives the optimal solution when this gap becomes close to zero (≤10−6 ). If an optimal
solution is reached within the 2 h run time, we report the optimal solution (Lower
bound = Upper bound) and the CPU run time in seconds. If the run time exceeds
2 h, we report the best bounds that were obtained within the 2 h.
In Table 5, when the optimality gap is greater than 0, it means that an optimal
solution was not reached before 2 h. We find that the models were able to obtain
optimal solutions for most of the test cases. Both the models reach optimal solutions
for 26 out of the 30 tested instances within 2 h of run time. As the problem size
increases, the computational time to reach the optimum increases as well. We have
reported that the optimality gap is not more than 3% (0.03) in all the tested cases. We
also observe that the models do not dominate each other. They are rather comple-
mentary in terms of their computational performance. Both models offer competitive
solutions in a computationally efficient manner.

4 Heuristics for the Inventory Routing Problem

We know that IRP is an NP-hard problem. As evidenced by the computational exper-

iments above, the CPU time increases as the problem size increases. It takes a
huge computational effort to find the optimal solution to the problem. Therefore,
we propose two simple heuristics that provide near-optimal solutions to the IRP.
Both the proposed heuristics are two-phased. The first phase finds feasible delivery
schedules and delivery quantities. In the second phase, delivery routes are obtained.
The details of the heuristics are presented below.

4.1 Heuristic 1

The proposed heuristic is two-phased. In the first phase, we find the vehicle assign-
ments and retailer allocations using an MILP model. We obtain the information on
when the deliveries should be made, by which vehicle, and what the corresponding
delivery quantities are. In the second phase, we find the routes that should be followed
by the vehicles when making those scheduled deliveries. Given the delivery sched-
ules and quantities from the first phase, we just need to solve a travelling salesman
problem (TSP) for each assigned vehicle in every period. We know that the optimiza-
tion of routing cost is computationally costly. Therefore, a combination of simple
Table 5 Computational performance of the proposed MILP models
N T V MILP model 1 MILP model 2
Lower bound Upper bound CPU time (Sec) Optimality gap Lower bound Upper bound CPU time (Sec) Optimality gap
5 3 3 207.65 207.65 0.34 0 207.65 207.65 0.31 0
5 3 3 175.53 175.53 0.69 0 175.53 175.53 0.25 0
5 3 3 217.48 217.48 0.72 0 217.48 217.48 0.56 0
5 5 3 341.33 341.33 65.57 0 341.33 341.33 44.60 0
5 5 3 338.99 338.99 196.98 0 338.99 338.99 291.18 0
5 5 3 338.29 338.29 2.29 0 338.29 338.29 2.96 0
6 3 3 202.72 202.72 5.29 0 202.72 202.72 1.56 0
Mathematical Models and Heuristics …

6 3 3 250.63 250.63 1.34 0 250.63 250.63 1.65 0

6 3 3 220.65 220.65 8.39 0 220.65 220.65 7.41 0
6 5 3 309.54 309.54 2872.32 0 309.54 309.54 1282.44 0
6 5 3 408.40 408.40 339.88 0 408.40 408.40 339.57 0
6 5 3 392.27 392.27 415.24 0 392.27 392.27 6597.86 0
7 3 3 251.67 251.67 17.21 0 251.67 251.67 26.05 0
7 3 3 251.34 251.34 18.10 0 251.34 251.34 107.16 0
7 3 3 261.21 261.21 79.20 0 261.21 261.21 129.98 0
7 5 3 390.25 390.25 253.60 0 390.25 390.25 412.94 0
7 5 3 430.69 430.69 66.44 0 430.69 430.69 151.46 0
7 5 3 440.81 440.81 543.98 0 440.81 440.81 1276.45 0
8 3 3 243.87 243.87 55.05 0 243.87 243.87 125.94 0
8 3 3 287.23 287.23 19.64 0 287.23 287.23 139.75 0
(continued)
55
Table 5 (continued)
56

N T V MILP model 1 MILP model 2

Lower bound Upper bound CPU time (Sec) Optimality gap Lower bound Upper bound CPU time (Sec) Optimality gap
8 3 3 269.46 269.46 26.33 0 269.46 269.46 141.81 0
8 5 3 418.08 424.72 7200.00 0.02 411.91 424.72 7200.00 0.03
8 5 3 441.51 446.03 7200.00 0.01 441.15 446.03 7200.00 0.01
8 5 3 394.55 394.55 1480.17 0 389.17 394.55 7200.55 0.01
9 3 3 353.08 353.08 1299.27 0 353.08 353.08 162.37 0
9 3 3 312.51 321.88 7202.24 0.03 321.88 321.88 1635.56 0
9 3 3 330.24 330.24 168.09 0 330.24 330.24 76.53 0
10 3 3 362.09 362.09 448.71 0 362.09 362.09 597.03 0
10 3 3 1027.31 1031.91 7200.21 0 1019.92 1031.91 7202.94 0.01
10 3 3 753.92 764.50 7202.41 0.01 764.50 764.50 1167.93 0
K. R. Preethi et al.
Mathematical Models and Heuristics … 57

travelling salesman problem (TSP) heuristics is used to find the routes in the second
phase.

4.1.1 Phase 1

The following MILP gives delivery schedules and delivery quantities for the entire
planning horizon.


T 
V 
T 
N
 
Minimize Z H 1 = Fv v,t + h i Ii,t + πi Bi,t
t=1 v=1 t=1 i=1


T 
V 
N 
T 
V
+ δi,v,t di∗ + v,t d B∗ (38)
t=1 v=1 i=1 t=1 v=1


V
δi,v,t ≤ 1, ∀ i, t (39)
v=1


N
δi,v.t ≤ N v,t , ∀v, t (40)
i=1

Q i,v,t ≤ Mi δi,v,t , ∀i, v, t (41)


N
Q i,v,t ≤ Capv v,t ∀v, t (42)
i=1


V 
T 
T
Q i,v,t ≤ Demi,t ∀i (43)
v=1 t=1 t=1


V
Ii,t − Bi,t = Ii,t−1 − Bi,t−1 − Demi,t + Q i,v,t , ∀i, t (44)
v=1

Ii,t ≤ Ci ∀i, t (45)

where d B∗ is the minimum

  distance travelled to reach the vendor from any retailer,
d B∗ = mini = 1,..,N di,B , and di∗ is the minimum distance travelled
 to reach retailer
i from either the vendor or other retailers, di∗ = min j = B,1,..,N d j,i .
j = i
In this phase, we retain the vehicle assignment and retailer allocation variables
which determine which retailers are scheduled for delivery and which vehicles are
making the deliveries in each period. The inventory-related variables and the delivery
quantities are also included in the first phase. However, the precedence-based routing
58 K. R. Preethi et al.

variables X i, j,v,t and hence the routing decisions are eliminated from this part of the
analysis.
Here, the objective function (Z H 1 ) is the sum of fixed transportation cost, inventory
holding, and backordering costs and an approximation of the variable transportation
cost. The approximation is the sum of the minimum distance travelled to reach each
active node over the entire finite time horizon. It is given by the last two terms of the
objective function. It is worthwhile to note that the first phase also gives a meaningful
lower bound for the IRP.

4.1.2 Phase 2

We have the vehicle retailer assignments and the quantities to be delivered from phase
1. Now, constructing the tours for the vehicles in each period will give a feasible
solution to the IRP. Tours are constructed using two simple travelling salesman
problem (TSP) heuristics. First, we use the nearest neighbourhood heuristic to form
the tours. The tours are then improved by using the pairwise interchange heuristic.
The routing costs from this phase combined with the fixed transportation cost and
the inventory costs from phase 1 give a feasible solution (upper bound) for the IRP.
The nearest neighbourhood algorithm is implemented as follows. Let us just
consider a single vehicle and the set of retailers allocated to it, in the first period.
The vehicle starts at the vendor node, visits the nearest retailer node, and makes
the delivery. From there, the vehicle visits another unvisited retailer node which is
closest to the current node. The vehicle returns to the vendor node after all the allo-
cated deliveries are made. Repeat this for all the vehicles assigned for delivery in each
period. Once we have the nearest neighbourhood solution, pairwise interchanges of
the retailers are made to search for an improved solution.

4.2 Heuristic 2

4.2.1 Phase 1

In this heuristic, the first phase is MILP based in which we find the delivery schedules
and quantities. We extend the MILP formulation from Heuristic 1 by including addi-
tional variables and constraints. The additional variables help us keep track of the
position in which the retailers are visited by a vehicle. This provides extra informa-
tion on the routes and therefore, a better approximation for the variable transportation
cost. The additional variables in the model are presented in Table 6.
Like Heuristic 1, the phase 1 objective function consists of the fixed cost of
transportation, inventory holding, and backordering costs and a better approximation
of the variable transportation cost.
This approximation given by the last part of the objective function consists of two
parts. For retailers that are visited either in the first position or in the last position,
Mathematical Models and Heuristics … 59

Table 6 Decision variables used in Heuristic 2–Phase 1

Notations Definitions

λi,1,v,t 1, if vehicle v delivers first to retailer i in period t
0, otherwise

λi,2,v,t 1, if vehicle v does not visit retailer i in the first position in period t
0, othherwise

λi,3,v,t 1, if vehicle v delivers last to retailer i in period t
0, otherwise
 
v,t 1, if vehicle v serves only one retailer in period t
0, otherwise

the distance between the vendor and the retailer is added to the variable cost approx-
imation. For the remaining active retailers, the least distance travelled to reach that
retailer from any other retailer is added
 to the variable cost component. We have
defined di∗∗ = min d j,i .
j = 1, .., N
j = i
Consider a situation in which a vehicle v visits only one retailer in a given period
t. Here, the same retailer is visited in the first and the last position by the vehicle.
Therefore, we need both the values of λi,1,v,t and λi,3,v,t to be equal to 1 in that

case. To ensure this, we introduce another binary variable v,t in the model. This
variable takes a value of 1 if vehicle v serves only one retailer in period t.


T 
V 
T 
N
 
Minimize Z H 2 = Fv v,t + h i Ii,t + πi Bi,t
t=1 v=1 t=1 i=1


T 
V 
N
 
+ d B,i λi,1,v,t + di,B λi,3,v,t + λi,2,v,t di∗∗ (46)
t=1 v=1 i=1

In addition to the constraints (39) to (45), we add the following constraints to the
MILP model.


N
λi,1,v,t = v,t ∀v, t (47)
i=1

λi,1,v,t + λi,2,v,t = δi,v,t ∀i, v, t (48)

60 K. R. Preethi et al.


N
λi,3,v,t = v,t ∀v, t (49)
i=1


N
δi,v,t ≥ 2 − (N + 1)v,t ∀v, t (50)
i=1


N
 
δi,v,t ≤ 1 + (N + 1) 1 − v,t ∀v, t (51)
i=1

λi,1,v,t + λi,3,v,t ≤ δi,v,t + v,t ∀i, v, t (52)


N
  
N
λi,1,v,t + λi,2,v,t + λi,3,v,t = δi,v,t + v,t ∀v, t (53)
i=1 i=1

λi,1,v,t + λi,2,v,t + λi,3,v,t ≤ 2δi,v,t ∀i, v, t (54)

A vehicle should visit exactly one retailer in the first position if the vehicle is
assigned for delivery (Eq. 47). A retailer allocated for delivery is visited either in the
first position or other positions by a vehicle (Eq. 48). A vehicle should visit exactly

one retailer in the last position if it is assigned for delivery (Eq. 49). v,t takes a
value 1 if the vehicle v serves only one retailer in period t (Eqs. 50 and 51). Both
λi,1,v,t and λi,3,v,t are equal to 1 if retailer i is the only retailer visited by vehicle v
in period t (Eqs. 52 and 53). This enables us to add the distance from the vendor
to retailer i and back to the vendor to the variable cost approximation. To further
tighten the model, we add constraints (Eq. 54) in the model. Phase 1 of Heuristic 2
will also provide a meaningful lower bound to the problem.

4.2.2 Phase 2

Given the delivery schedules and quantities from Phase 1, we obtain the routes using
the nearest neighbourhood heuristic followed by the implementation of the pairwise
interchange heuristic. The routes together with the phase 1 delivery schedules and
delivery quantities provide a heuristic solution to the Inventory Routing Problem.
The routing cost from Phase 2 plus the fixed transportation cost and the inventory
costs from Phase 1 is the total supply chain cost and is a feasible solution (upper
bound) to the IRP.
Mathematical Models and Heuristics … 61

4.3 Computational Analysis and Discussion

The performance of the proposed heuristics is presented in Table 7. We used the same
set of generated data instances from Sect. 3 to evaluate the computational efficiency

Table 7 Computational performance of the heuristics

N T V Upper bound Heuristic 1 Heuristic 2
fromMILP Feasible Deviation (in Feasible Deviation (in
model 1 solution %) solution %)
5 3 3 207.65 264.54 27 262.06 26
5 3 3 175.53 199.14 13 197.24 12
5 3 3 217.48 246.35 13 246.31 13
5 5 3 341.33 367.47 8 379.78 11
5 5 3 338.99 373.74 10 373.74 10
5 5 3 338.29 361.52 7 369.14 9
6 3 3 202.72 238.37 18 238.75 18
6 3 3 250.63 287.76 15 284.41 13
6 3 3 220.65 239.43 9 246.46 12
6 5 3 309.54 323.08 4 344.49 11
6 5 3 408.40 483.26 18 441.34 08
6 5 3 392.27 406.77 4 430.94 10
7 3 3 251.67 322.8 28 302.45 20
7 3 3 251.34 306.93 22 299.04 19
7 3 3 261.21 303.17 16 311.73 19
7 5 3 390.25 448.92 15 444.68 14
7 5 3 430.69 503.19 17 537.18 25
7 5 3 440.81 481.93 9 517.33 17
8 3 3 243.87 287.21 18 278.73 14
8 3 3 287.23 341.22 19 335.18 17
8 3 3 269.46 286.85 6 295.87 10
8 5 3 424.72 515.79 21 519.23 22
8 5 3 446.03 549.1 23 575.19 29
8 5 3 394.55 469.11 19 488.62 24
9 3 3 353.08 428.02 21 445.57 26
9 3 3 321.88 383.81 19 352.34 9
9 3 3 330.24 406.16 23 390.36 18
10 3 3 362.09 439.53 21 436.34 21
10 3 3 1031.91 1075.58 4 1099.4 7
10 3 3 764.50 834 9 848.64 11
62 K. R. Preethi et al.

of the proposed heuristics. We show that both the heuristics generate near-optimal
solutions to the IRP.
Heuristic 1 gives good quality feasible solutions to the problem for all the tested
instances within a few seconds of run time (not more than 10 s). We compared the
performance of the heuristics against the upper bounds obtained from MILP model
1 (described in Sect. 2). The percentage deviation of the heuristic solutions from the
upper bounds obtained using MILP model 1 is reported in Table 7. We observe that
the maximum deviation is 28% among the tested cases. The computational efficiency
of heuristic 1 is more evident in larger problem sizes.
Heuristic 2 also obtains good quality solutions within reasonable CPU time. It
takes not more than 2 min (except in 1 case) to obtain the solution to the problem.
The maximum deviation of heuristic 2 solutions from the MILP model’s upper bound
is 29%. Again, the advantage of heuristic 2 over the optimal MILP model is more
pronounced in larger problems. The quality of heuristic 1 solutions is comparable
to that of heuristic 2’s solution, but the latter runs longer. This is expected as the
first phase of heuristic 2 has more decision variables and constraints than heuristic 1.
However, it is worthwhile to note that the lower bounds from heuristic 2 (first phase)
will be tighter. Both the heuristic solutions can be used as initial seeding solutions
for a metaheuristic.

5 Conclusion

5.1 Theoretical Contributions

We developed two efficient MILP models to find optimal solutions to the IRP with
backorders and no split deliveries. We showed that the models are computationally
efficient and can solve smaller instances easily. For larger instances, we were able to
obtain close bounds within 2 h of run time. The optimality gap was no more than 3%
in the tested data instances. We further proposed two MILP-based heuristics that give
near-optimal solutions to the NP-hard IRP. The heuristics can be used to generate
initial solutions to a metaheuristic.

5.2 Future Research

We do not account for delivery times in the model. We assume that all the deliv-
eries could be made within a given period. Therefore, there is potential for future
research which includes delivery times and delivery time windows. Future research
may also incorporate the vendor’s production decisions in the analysis. Alternatively,
the vendor’s inventory planning problem for ordering from an external supplier (a
three-echelon supply chain) can also be considered.
Mathematical Models and Heuristics … 63

Acknowledgements Note from the first author—I am grateful to Indian Institute of Technology,
Madras for funding the research. I would also like to thank Nanyang Technological University,
Singapore, for its support in carrying out the research.

References

Abdelmaguid, T. F., Dessouky, M. M., & Ordóñez, F. (2009). Heuristic approaches for the inventory-
routing problem with backlogging. Computers & Industrial Engineering, 56(4), 1519–1534.
Anily, S., & Federgruen, A. (1990). One warehouse multiple retailer systems with vehicle routing
costs. Management Science, 36(1), 92–114.
Archetti, C., Bertazzi, L., Laporte, G., & Speranza, M. G. (2007). A branch-and-cut algorithm for
a vendor-managed inventory-routing problem. Transportation Science, 41(3), 382–391.
Bertazzi, L., Paletta, G., & Speranza, M. G. (2002). Deterministic order-up-to level policies in an
inventory routing problem. Transportation Science, 36(1), 119–132.
Campbell, A. M., & Savelsbergh, M. W. (2004). A decomposition approach for the inventory-routing
problem. Transportation Science, 38(4), 488–502.
Campbell, A., Clarke, L., Kleywegt A., & Savelsbergh, M. (1998). The inventory routing problem.
Fleet management and logistics (pp. 95–113)
Coelho, L. C., & Laporte, G. (2013a). A branch-and-cut algorithm for the multi-product multi-
vehicle inventory-routing problem. International Journal of Production Research, 51(23–24),
7156–7169.
Coelho, L. C., & Laporte, G. (2013b). The exact solution of several classes of inventory-routing
problems. Computers & Operations Research, 40(2), 558–565.
Coelho, L. C., Cordeau, J. F., & Laporte, G. (2013). Thirty years of inventory routing. Transportation
Science, 48(1), 1–19.
Laporte, G. (1992). The vehicle routing problem: An overview of exact and approximate algorithms.
European Journal of Operational Research, 59(3), 345–358.
Preethi, K. R., Rajendran, C., & Viswanathan, S. (2018). Lower bounding methods for an inventory
routing problem with backorders. In: X. Xu, et al. (Ed.), Proceedings of International Conference
on Computers and Industrial Engineering 48th 2018 (Vol. 3, p. 2022).
Toth, P., & Vigo, D. (2002). The vehicle routing problem. In Society for industrial and applied
mathematics.
Viswanathan, S., & Mathur, K. (1997). Integrating routing and inventory decisions in one-warehouse
multiretailer multiproduct distribution systems. Management Science, 43(3), 294–312.
Waller, M., Johnson, M. E., & Davis, T. (1999). Vendor-managed inventory in the retail supply
chain. Journal of Business Logistics, 20, 183–204.
Modeling a Supply Chain
with Price-Dependent Stochastic Demand
and Discrete Transportation Lead Time

Susheel Yadav, Anil K. Agrawal, and Manu K. Vora

Abstract This article deals with the modeling of a supply chain where a single
manufacturer sells its product through multiple buyers. All the buyers face a price-
dependent stochastic demand. The manufacturer sends the item to the buyers in
multiple shipment of equal-sized sub-batches. A third-party logistics ships the item to
the buyers. The different types of the vehicles provided by this third party have buyer-
specific transportation cost per shipment and transportation lead time per shipment.
Under these conditions, the safety stock of the retailers and the discrete transportation
lead time become crucial for the supply chain, especially when a minimum service
level is defined for each retailer. The problem is formulated as a Mixed Integer
Nonlinear Programming (MINLP) model and is solved for the maximum supply
chain profit. Finally, important managerial insights are drawn for the safety factor
and service levels of the buyers.

Keywords Price-sensitive stochastic demand · Discrete transportation lead time ·

Service level · MINLP · SCM

1 Introduction

Today, the businesses are competing along the supply chain (Li et al., 2006) and this
makes supply chain management (SCM) an important task to be handled judiciously.
The presence of a large number of supply chain partners associating themselves
into various activities and functions makes the chain complex (Arshinder et al.,
2008). Efficient integration among the supply chain partners emerged as an important
strategy for SCM as the competition proliferated (Ahmadizar et al., 2015). Arshinder
et al. (2011) mentioned joint consideration of the functions and processes by supply

S. Yadav (B) · A. K. Agrawal

Department of Mechanical Engineering, Indian Institute of Technology (BHU), Varanasi 221005,
India
e-mail: [email protected]
M. K. Vora
Business Excellence Inc, Naperville, IL 60567-5585, USA
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 65
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_4
66 S. Yadav et al.

chain partners as an important aspect of the supply chain. Joint consideration of the
inventories across the supply chain is one such school of thought where the inventory
policies of the supply chain partners are decided by keeping the total supply chain
in view and not the one of them.

2 Literature Review

2.1 Joint Economic Lot Size (JELS) Models

Goyal (1976) proposed Joint Economic Lot Size (JELS) model that aimed to mini-
mize the joint inventory-related costs for the case of a single vendor and single buyer
(SVSB). He showed that a considerable saving can be achieved by determining the
economic lot size jointly. Later, this model was extended along different consid-
erations. This paper was the building block for many integrated inventory models
that were developed by various researchers along different considerations. Banerjee
(1986) in his JELS model considered the supplier as a manufacturer supplying
product to a buyer on lot-for-lot policy. Goyal (1988) extended this and modified the
lot-for-lot supply policy to the policy of supplying multiple sub-batches of equal size
after the lot has been completely produced. Lu (1995), and Agrawal and Raju (1996)
proposed the strategy of supplying equal lot size batches even during the produc-
tion, to further reduce the joint inventory cost. Hill (1997, 1999), Goyal (2000), and
Goyal and Nebebe (2000) extended the basic JELS model with the consideration of
the shipment in unequal-sized batches or in combination of equal- and unequal-sized
sub-batches. A lot of research work is available on joint consideration of inventory
of supply chain partners. A review of the same is available in the work of Goyal and
Gupta (1989), Ben-Daya et al. (2008), and Glock (2012a).

2.2 Inventory Models with Deterministic Demand

The classical inventory models assume the demand as constant (Baker and Urban,
1988); however, it need not be so in practice (Gupta, 1994). There are certain factors
that affect the demand; one such factor is the selling price of the item which regulates
the customer’s demand (Das Roy and Sana, 2011; Agrawal and Yadav, 2020; Yadav
et al., 2020). In the early work of Whitin (1955), the demand of the product was taken
as a function of the product price. The models which assume the price–demand rela-
tionship with certainty are known as deterministic inventory models (Bushuev et al.,
2011). This price–demand relationship serves as the input parameter in proposing
various kinds of inventory policies. Later, Urban (1992), Abad (1994), Goyal and
Gunasekaran (1995), Petruzzi and Dada (1999), You (2005), and Wu et al. (2009)
considered the price-dependent demand in their studies. These studies in essence
Modeling a Supply Chain with Price-Dependent Stochastic Demand … 67

determine the inventory policies as well as the pricing decisions for the supply chain.
Initial research in JELS problem was concentrated on production shipment schedule
in terms of number and size of the batches, mostly considering the demand as exoge-
nous where it does not depend on any other parameter (Jokar and Sajadieh, 2009).
Later on, Sajadieh and Jokar (2009), Jokar and Sajadieh (2009), Zanoni et al. (2014),
Lin and Lin (2015), Taleizadeh et al. (2016), Taleizadeh and Noori-Daryan (2016),
and Chen and Sarker (2017) considered a deterministic price–demand relationship
in their study of JELS inventory problems.

2.3 Inventory Models with Stochastic Demand

When the deterministic demand assumption is relaxed, the model turns out to be
stochastic in nature (Whitin, 1954). Stochastic demand assumption has extensively
followed in single-period inventory models over the time. Sharafali and Co (2000) are
the first who considered the stochastic demand function in integrated inventory model
of a single vendor and a single buyer. Later on, deterministic demand assumption
is also relaxed by Ben-Daya and Hariga (2004) in JELS model as they considered
the lead time demand to follow normal distribution making the model stochastic
in nature. The stochastic demand assumption in JELS model is also followed by
Taleizadeh et al. (2010), Jha and Shanker (2013), AlDurgam et al. (2017), Tiwari
et al. (2018), and Wangsa and Wee (2018).

2.4 Inventory Models with Price-Dependent Stochastic

Demand

Though, the retail demand is not only price-sensitive but also stochastic (Ray et al.,
2005). Mills (1959) was the first to study the effect of price and randomness on the
demand. Later, several authors have studied the effect of price-dependent stochastic
nature of demand on pricing and inventory control policies (Petruzzi and Dada, 1999;
Ray et al., 2005; Lau et al., 2007; Jadidi et al., 2017). But there is hardly any literature
available on the effect of a price-dependent stochastic nature of demand on pricing
and inventory policies in the integrated vendor–buyer environment. This study tries
to fill this research gap by considering the demand as price-dependent stochastic
JELS problem environment.
68 S. Yadav et al.

2.5 Inventory Models with Lead Time, Transportation Time,

and Their Control

When the demand is assumed as stochastic in nature, lead time becomes an important
concern and its control leads to several benefits (Ouyang et al., 2004). Giri and Roy
(2015) asserted that lead time reduction can bring lower safety stock level, low
stock-out loss, and improved customer service level, leading toward an efficient
supply chain. Liao and Shyu (1991) were the first who argued that lead time can be
controlled at an additional cost and considered lead time as a variable rather than
assuming it as uncontrollable parameter. Ben-Daya and Raouf (1994) considered
lead time and ordered quantity as variable and further classified the additional cost on
reducing the lead time into three categories: administrative cost, transportation cost,
and supplier’s speedup cost. Liao and Shyu (1991), and Ben-Daya and Raouf (1994)
both agreed that transportation time is an important factor in such studies; however,
an explicit consideration of transportation time was not included in their models.
Glock (2012b) studied the lead time reduction strategies for a single-vendor and
single-buyer integrated inventory model with explicit consideration of transportation
time. He considered transportation time is related to the total nonproductive time
and expressed it as a fraction of the total nonproductive time. Mou et al. (2017)
revisited Glock (2012b) and relaxed the assumption that the transportation time is
a fraction of total nonproductive time. They expressed the transportation time and
total nonproductive as two independent variables. Chang and Lo (2009) considered
delivery lead time explicitly in their study; they considered that the total lead time is
composed of continuous lead time and discrete lead time. They assumed that delivery
time depends on the mode of transportation and can be reduced only by changing the
mode of transportation making this reduction as discrete. Multanen (2011) studied
empirically the effect of reduction in transportation time on inventory levels and
costs. He found that it is not necessary that increasing the transportation cost to
reduce the transportation time will increase the total cost of the supply chain because
at the same time due to the reduction in the transportation time, savings can be gained
in inventory cost. Ng et al. (1997) emphasized any attempt to reduce the cycle time
must consider transportation time also, and suggested a number of ways to reduce
it. Lead time can be reduced by investing more in advanced transportation systems.
The transportation time is a major factor in lead time reduction issues and requires
more recognition. Braglia et al. (2016) studied the safety stock for a single-vendor
and single-buyer supply chain for a case of controllable lead time. They developed
both exact and approximate algorithms.

2.6 Inventory Models with Service-Level Constraint

In the situations when the demand exceeds the quantity available, stock-out occurs.
Ouyang et al. (1996) considered this stock-out in their inventory model by assuming
Modeling a Supply Chain with Price-Dependent Stochastic Demand … 69

that a fraction of the unmated demand can be backordered at an additional cost while
the remaining will be lost. Thus, a stock-out cost is introduced in the inventory model.
However in practice, it is very difficult to estimate a stock-out cost, so dealing stock-
out through the consideration of a service-level constraint is an alternative (Aardal
et al., 1989). Ouyang and Wu (1997) also considered a service-level constraint in
their inventory model to deal with stock-out. Jha and Shanker (2009a, b) considered
a service-level constraint as a substitute of stock-out cost in a single-vendor and
single-buyer integrated inventory problem for decaying and non-decaying items,
respectively. Li et al. (2011) analyzed two inventory models for a single-vendor
and single-buyer supply chain with the consideration of service-level constraint; the
first model is a decentralized model based on Stackelberg game, and the second is
an integrated model. Jha and Shanker (2013) considered the minimum service-level
constraint in a single-vendor and multiple buyers’ integrated inventory problem also.
A large number of inventory models could be found in the literature, and most
of them are bounded by their defined conditions. Thus, a model defined for one
environment cannot be applied in a different environment. This study is inspired by
the problem of a manufacturing company which sells its product through its retailers
and uses a third party for the transportation of the item. Each of the vehicle types
provided by the third-party logistics has a buyer-specific transportation cost and a
transportation lead time. In the literature, a number of inventory models are available
which considers the lead time as continuous and can be reduced at the expense
of crashing cost. But in this problem, the transportation lead time is considered as
discrete and different lead time values can be selected for a specific buyer at the
expense of a different transportation cost. Additionally, a minimum service level is
defined for each of the retailers; i.e., each of the retailers has to maintain a certain level
of safety stock in order to fulfill the minimum service-level constraint. The demand
of the retailers is defined as price-dependent stochastic. We present this problem as
an Mixed Integer Nonlinear Programming (MINLP) mathematical model and solve
for the maximum supply chain profit.
The organization of the remaining paper is as follows: Section 3 describes the
notations and assumptions involved. The model considered has been described in
Sect. 4. The mathematical formulation of the problem is presented in Sect. 5. Section 6
presents illustrative numerical examples and their results. Sensitivity analysis has
been presented in Sect. 7. Section 8 summarizes the results, provides some important
managerial insights, and suggests future research directions.

3 Assumptions and Notations

The problem also gets characterized by the following assumptions and notations.
1. The planning has been considered for infinite time.
2. Replenishment cycle for each sub-batch for all the buyers is of same length.
3. Each of the buyers has a minimum service-level constraint.
70 S. Yadav et al.

4. Transportation lead time for a buyer depends upon the type of the vehicle and on
the buyer itself. The vehicle type selected by the buyer has a transportation lead
time L i j , and the demand during this period willbe normally distributed with
mean D L i = Di L i j and a standard deviation of σεi L i j . Thus, the transportation
  
lead time demand can be expressed as X ∼ N Di L i j , σεi L i j .
5. Third-party logistics provider can provide any number of vehicles to the buyers.
For shipment of a sub-batch from the manufacturer to the buyer, any number of
a particular type of vehicle can be used by the buyers.
6. Inventory carrying cost rate for a supply chain partner may be different from
others.

αi Proportion of the demand not fulfilled from the stock, thus (1 − αi ) service level
c Per unit cost incurred by the manufacturer ($)
i Index for a buyer
j Index for a vehicle type
N Total number of buyers
P Annual production rate of manufacturer (units/year)
hi Annual inventory holding cost rate incurred for ith buyer
hm Annual inventory holding rate incurred for manufacturer
r pi Reorder point
si Safety stock
Cj Capacity of vehicle type j (units)
Fi j Per shipment transportation cost for transporting the items to ith buyer using single unit of
vehicle type j ($)
m Number of sub-batches during production cycle of the manufacturer
w Unit purchase price to buyers ($)
x Ratio of annual production rate to annual cumulative demand rate
D Annual demand for the manufacturer being equal to cumulative demand of all the buyers
(units/year)
T Total annual supply chain profit ($/year)
X Demand during the transportation lead time
C̄ Total annual cost to manufacturer ($/year)
P̄ Total annual profit of manufacturer ($/year)
P̂ Total annual profit of all the buyers ($/year)
Ai Ordering cost per order of ith buyer ($/order)
Am Setup cost per setup of manufacturer ($/setup)
α Profit margin of the manufacturer as a fraction of unit cost of the manufactured item
ki Safety factor
n Number of equal-sized sub-batches per order
t Common replenishment cycle time for all the buyers (year)
Q Production lot size of the manufacturer (units)
(continued)
Modeling a Supply Chain with Price-Dependent Stochastic Demand … 71

(continued)
pi Unit selling price of ith buyer ($)
Ni j Number of the vehicles of type j employed in supplying the item to the ith buyer

4 The Model

The supply chain considered in this study is a single-manufacturer and multiple

buyers’ supply chain which deals in one single item. The cost of third-party logistics
service provider is borne by the buyers. The third-party logistics provider offers
a number of vehicle types with different capacities to the buyers and charges the
transportation cost from the particular buyer on the basis of the vehicle type selected.
Each vehicle of a particular type provided by the third-party logistics provider has a
transportation cost and a transportation time specific to the buyer, and thus a same type
of vehicle may have a different transportation cost and a different transportation time
for different buyers. Due to this, the model assumes the nonidentical transportation
lead time for the buyers. Inventory at the buyers’ end is reviewed continuously, and
for each of the buyers, a service-level constraint is specified; as in many practical
situations, the shortage cost cannot be exactly quantified due to intangible losses
(Choudhary and Shankar, 2013). Service-level constraint also limits the stock-out
level per cycle for each buyer. The buyers sell the item in the market where the ith
buyer faces an average demand rate of (Di ), per unit time which is price-sensitive
and stochastic in nature, and since the manufacturer  fulfills the demand
 of all the
N
buyers so the manufacturer faces a demand rate of D = i=1 Di per unit time.
The demand function of the ith buyer can be expressed as

Di = y( pi ) + εi (1)

where y( pi ) is a linear decreasing function of the per unit price pi and can be
expressed as y( pi ) = ai − bi pi , where ai , bi , and pi , in Eq. (1), respectively,
represent the maximum demand, price elasticity constant, and selling price of the
item and εi is a continuous random variable for the ith buyer. The above expression is
in the same line with that of Petruzzi and Dada (1999), Ray et al. (2005), and Jadidi
et al. (2017). Thus,

Di = ai − bi pi + εi . (2)

This is also assumed that εi follows a normal distribution with mean μεi and
variance as σε2i , i.e., N μεi , σε2i . Thus, the demand rate of the ith buyer can be
 
expressed as Di ∼ N λ( pi ), σε2i , where λ( pi ) = y( pi ) + μεi , and if μεi = 0, ∀i,
then λ( pi ) = y( pi ).
72 S. Yadav et al.

A synchronization policy as of the second model of Hoque (2008), and Jha and
Shanker (2013) is followed in this study to achieve the synchronization in delivery and
replenishment among the manufacturer and the buyers. The manufacturer produces
the item at a finite production rate (P) and produces a lot of size of Q units in one
setup. However, these Q units are shipped to all the buyers in n number of equal-sized
sub-batches. These sub-batches are shipped to the buyers even during the production
of the lot. A buyer can select any number of the vehicles of a particular type for the
transportation of the sub-batch from the manufacturer to it. Although the shipment
from the manufacturer to all the buyers is made at the same time due to the vehicle
and buyer-specific transportation lead times, the buyers may receive the orders at
different times (Fig. 1). From the sub-batch of size q = Qn , the ith buyer receives a
quantity qi = DDi q which is in the ratio of its annual demand (Di ) to the total annual
 N 
demand D = i=1 Di of all the buyers. The ith buyer consumes this quantity qi
at a rate of Di in time t = Dqii . For each buyer in the network, the replenishment cycle
time will be t = Dq11 = Dq22 = . . . = Dq NN . The above philosophy for the production
and shipment to the buyers can be understood from Fig. 1 for a single-manufacturer

Fig. 1 Inventory pattern of the single-manufacturer and multiple buyers’ supply chain
Modeling a Supply Chain with Price-Dependent Stochastic Demand … 73

and three-buyer case. It can be observed from Fig. 1 that

Q
t= (3)
nD

5 Mathematical Model Formulation

5.1 Annual Profit of the Manufacturer

The manufacturer charges a profit markup of α on its incurred total annual costs and
supplies to the buyers at a price of w. The manufacturer incurs the following costs.

5.1.1 Annual Cost on Finished Products Incurred by the Manufacturer

Since the manufacturer faces a total annual demand of D units and incurs a per unit
cost c which include all the related costs such as unit raw material purchase cost,
unit raw material holding cost, and unit processing cost, thus the manufacturer will
incur an annual cost of

Pm = cD. (4)

5.1.2 Annual Inventory Holding Cost Incurred by Manufacturer

This cost is taken in proportion to the average inventory that can be determined by
dividing the manufacturer’s inventory cycle area by the cycle length. Inventory held
by the manufacturer in one cycle can be calculated by determining the area under the
inventory cycle curve of the manufacturer. From Fig. 1, the area under the inventory
cycle curve of the manufacturer (Cm ) can be determined as

q q q P q P q
Cm = + +q −1
2P 2 D D D D

q P q P q
+ ... + +q − 1 (m − 1)
2 D D D D
Q − (m − 1)q DP − q Q − (m − 1)q DP − q
+
2 P
  
P q Q − (m − 1)q DP − q
+ Q − (m − 1)q − q −
D D P
74 S. Yadav et al.
  q 
+ {1 + 2 + 3 + . . . + (n − m − 1)} q .
D
Average inventory of the manufacturer (Im ) can be calculated by dividing the area
under the inventory cycle curve of the manufacturer (Cm ) or inventory held by the
manufacturer in one cycle by the cycle time (D/Q). Hence, average inventory of the
manufacturer will be

q q q P q P q
Im = + +q −1
2P 2DD D D
q P q P q
+ ... + +q − 1 (m − 1)
2DD D D
Q − (m − 1)q DP − q Q − (m − 1)q DP − q
+
2 P
  
P q Q − (m − 1)q DP − q
+ Q − (m − 1)q − q −
D D P
  q  D
+ [1 + 2 + 3 + . . . + (n − m − 1)] q (5)
D Q

After simplification,
 
Q 1 (n − 1)2
Im = 2 + (n − m − 1)(n − m) + m(2n − m − 1) − . (6)
2n P/D P/D

Therefore, the annual inventory holding cost of the manufacturer (Hm ) can be
calculated by multiplying the average inventory by inventory holding cost rate (rm )
and per unit cost (c).
Thus,

Hm = rm cIm

After putting the value of Im from Eq. (6) in the above, we would get
 
Q 1 (n − 1)2
Hm = 2 + (n − m − 1)(n − m) + m(2n − m − 1) − rm c
2n P/D P/D
(7)

5.1.3 Annual Setup Cost of the Manufacturer

Since the manufacturer produces Q units per setup and satisfies an annual demand
of D units annually, so the setup cost can be expressed as
Modeling a Supply Chain with Price-Dependent Stochastic Demand … 75

D
Sm = Am (8)
Q

So, the total annual cost of the manufacturer (C̄) will be the sum of annual purchase
cost (Pm ), annual inventory holding cost (Hm ), and annual setup cost (Sm ).
Thus,

C̄ = Pm + Hm + Sm

Substituting the values of Pm , Hm , and Sm into the above equation from their
respective Eqs. (4), (7), and (8), we get

Q 1
C̄ = cD + 2
+ (n − m − 1)(n − m)
2n P/D

(n − 1)2 D
+m(2n − m − 1) − cm c + Am . (9)
P/D Q

Since the manufacturer charges a markup (α) on his total incurred cost and supplies
the item to the buyers at per unit selling price of w, thus the selling price (w) of the
manufacturer will be

C̄
w = (1 + α) (10)
D

where C̄D would represent per unit production cost on the item by the manufacturer.
Manufacturer’s profit can be calculated by multiplying the markup factor to his
incurred total cost. Therefore, the annual profit of the manufacturer ( P̄) will be

P̄ = α C̄. (11)

5.2 Annual Profit of the ith Buyer

Profit of the ith buyer can be calculated by subtracting the total annual cost incurred
by the buyer from his total annual revenue (Ri ). Various costs incurred by this buyer
will be annual purchase cost (Pi ), inventory holding cost (Hi ), ordering cost (Oi ),


and transportation cost (Ti ). So, the annual profit of ith buyer will be:
 

Pi = Ri − (Pi + Hi + Oi + Ti ).
 

Pi = Di pi − (Pi + Hi + Oi + Ti ). (12)
76 S. Yadav et al.

Various cost elements experienced by the buyers are described and developed
below.

5.2.1 Annual Purchase Cost to the ith Buyer

Buyer i purchases annually a quantity (Di ) equal to its annual demand and pays w
per unit to the manufacturer; thus, annual purchase cost for the ith buyer will be

Pi = Di w. (13)

5.2.2 Annual Inventory Holding Cost to ith Buyer

The buyer takes the services of the third-party logistics provider which offers a
number of types of vehicle to transport the units from the manufacturer to the buyer
where each vehicle takes a buyer-specific transportation lead time and has a buyer-
specific cost per shipment. A buyer can select any number of the vehicles of a
particular type for transportation. The reorder point for the ith buyer is r pi which is
equal to the sum of demand during the transportation lead time and the safety stock.
Thus,

r pi = demand during the transportation lead time + safety stock

or

r pi = Di L i j + σεi Li j

where L i j is the transportation lead time for the ith buyer transporting the items with
the jth type of vehicle. The average inventory of the ith buyer  will be the mean of
the inventory just before the receipt of the ordered quantity  r pi − Di L i j and just
after the receipt of the ordered quantity qi + r pi − Di L i j . Thus, average inventory
of the ith buyer is
 
r pi − Di L i j + qi + r pi − Di L i j
Ibi =
2
   
Di L i j + σεi L i j − Di L i j + qi + Di L i j + σεi L i j − Di L i j
Ibi =
2
   
σεi L i j + qi + σεi L i j
Ibi =
2
Modeling a Supply Chain with Price-Dependent Stochastic Demand … 77
q  
i
Ibi = + σεi Li j
2
Annual holding cost for the ith buyer is

Hi = Ibi ri w
q  
i
Hi = + σεi L i j ri w
2

Since qi = Q Di
n D

Q 
Hi = Di + σεi L i j ri w (14)
2n D

5.2.3 Annual Ordering Cost to the ith Buyer

Annual ordering cost to the ith buyer can be expressed as

D
Oi = Ai (15)
Q

5.2.4 Annual Transportation Cost to the ith Buyer

The buyer takes the services of the third-party logistics provider which offers a
number of types of the vehicle where each vehicle takes a specific transportation
lead time to transport the units from manufacturer to the buyer and has a specific
cost per trip. A buyer can select any number of the vehicles of a particular type for
transportation. However, each of the buyers has a minimum service-level constraint
(1 − αi ), where αi is the proportion of the demand not fulfilled from the stock. Thus,
a buyer has to select a vehicle keeping its cost, capacity, and transportation lead time
in view. Thus, the transportation cost of the ith buyer will be

D 

N
Ti = n Ni j Fi j ∀i (16)
Q j=1

The vehicle type selected by the buyer has a transportation lead time L i j , and the
demand during this period will be normally distributed with mean D L i = Di L i j and
a standard deviation of σεi L i j . Thus, the transportation lead time demand can be
  
expressed as X ∼ N Di L i j , σεi L i j . Since in this study, any stock-out cost has
not been considered at the buyers’ end, rather a service-level constraint is defined
78 S. Yadav et al.

for each of the buyers. The stock-out cost at the buyers’ end can be replaced by
a service-level constraint assuming that the buyer has to fulfill a minimum fill rate
(Ouyang et al., 1996; Jha and Shanker, 2009a, b, 2013). Thus, the expected shortages
at the buyer’s end cannot exceed the given value αi .

Shortages at the end of the inventory cycle of ith buyer

≤ αi
Ordered quantity per cycle of ith buyer

The expected shortages at the end of the cycle of ith buyer in such environment are
also given as σεi L i j ψ(ki ) by Ravindran et al. (1987), and Jha and Shanker (2013).
where

ψ(ki ) = ϕ(ki ) − ki [1 − φ(ki )]

Here, ϕ, φ represent the standard normal probability density function and cumu-
lative distribution function, respectively, and ki is the safety factor of the ith
buyer.
or

σεi L i j ψ(ki )
≤ αi ∀i
qi

where qi = Q Di
n D

Dσεi L i j ψ(ki )
≤ αi ∀i (17)
Q Di /n

5.3 Annual Profit of the Supply Chain

Total supply chain profit is the sum of manufacturer’s profit and the profit of all the
buyers. Thus, the total supply chain profit can be computed from the following


N 

T = P̄ + Pi . (18)
i=1
Modeling a Supply Chain with Price-Dependent Stochastic Demand … 79

5.4 Formulation as a Mathematical Programming Problem

Because of integer requirements on the values of ‘m’ and ‘n’, the objective function is
discontinuous and classical optimization approach cannot be used. Thus, the problem
of maximization of overall supply chain profit is mathematically expressed as
N 

Maximize T = P̄ + i=1 Pi
subject to
Equations (1), (2), (3), (4), (5), (6), and

Ni j C j ≤ qi ∀i∀ j (19)

ai
pi ≤ ∀i (20)
bi

pi ≥ w ∀i (21)


N
D= Di (22)
i=1
√
σi L i ψ(ki )
≤ αi ∀i (23)
qi

α, t, Q, D, w, P̄, C̄, Im ≥ 0 (24)

 

ai bi , Di , Hi , Oi , pi , Pi , Ri , Pi , Ti ≥ 0 ∀i (25)

n, m, ≥ 0 and integer (26)

Ni j ≥ 0 and integer ∀i∀ j (27)

Equation 19 represents the constraint that for each of the buyers, the capacity of the
selected vehicles must be equal to or larger than the size of its sub-batch. Equation 20
represents the maximum selling price of the retailers so that the demand cannot be
negative, while Eq. 21 represents that this selling price must be greater than the
wholesale price of the manufacturer. Equation 22 represents that the demand faced
by the manufacturer is the sum of the individual demands of the retailers. Equation 23
represents the service-level constraint for each of the buyers. Finally, Eqs. 24, 25,
26, and 27 represent the non-negativity and integer constraint for various parameter
and decision variables.
80 S. Yadav et al.

The objective function of the problem is a nonlinear function and involves a

large number of continuous and discrete variables for which it is very difficult to
investigate the concavity property analytically (Mandal and Giri, 2015), although
several numerical methods can be used to optimize the such problems (Pattnaik,
2013). In this paper, we propose the use of optimization software package LINGO
17.0 to solve the developed mathematical model. Hariga et al. (2007) also emphasized
to make use of the high speed of computers and efficient solvers to such complex
problems.

6 Illustrative Example

The use of proposed model is being illustrated through an illustrative numerical

problem. The example problem is solved using an optimization software LINGO
17.0 on an Intel (R) Core (TM) i7-6700 CPU @ 3.40 GHz processor system with
8 GB RAM. The illustrative example considered here is of a single manufacturer and
three retailers (N = 3). The manufacturer-related data are: P = 12,000 units/year, c
= $ 60, Am = $ 1000, and rm = 0.10. The buyers-related data are given in Table 1.
The data for the third-party logistics are given in Tables 2 and 3. Third-party
logistics service provider offers three types of vehicles, namely Types I, II, and III,
with capacities as 20, 40, and 60 units, respectively. Their buyer-specific cost is
given in Table 2, while the transportation lead time data for the same are provided in
Table 3.
Table 4 summarizes the results obtained from the solution from the LINGO 17.0
solver. Table 5 shows the vehicle mix selected by the buyers for the maximum supply

Table 1 Example problem data for buyers

Buyer (i) ai bi σεi Ai ri αi
1 1500 10 400 25 0.20 0.05
2 1800 12 450 23 0.20 0.08
3 1400 11 300 22 0.20 0.06

Table 2 Transportation cost

Vehicle
data for example problem
Type I ( j = 1) Type II ( j = 2) Type III
( j = 3)
Buyer (i) Transportation Transportation Transportation
cost cost cost
1 16 30 42
2 20 25 30
3 12 22 30
Modeling a Supply Chain with Price-Dependent Stochastic Demand … 81

Table 3 Transportation lead

Vehicle
time data for example
problem Type I ( j = 1) Type II ( j = 2) Type III
( j = 3)
Buyer (i) Transportation Transportation Transportation
lead time (days) lead time (days) lead time (days)
1 4 6 8
2 5 7 10
3 3 5 7

Table 4 Optimal values of the variables from the model

Variable Value Variable Value Variable Value Variable Value
T 35,768.47 w 62.08 α 0 q3 40
P̄ 0 D 1991.85 p2 85.04 k1 1.13


P1 11,307.07 D1 692.82 p3 80.03 k2 0.95



P2 16,071.00 D2 779.42 Q 920.00 k3 0.96



P3 8390.39 D3 519.61 q1 53.33

t 0.0769 n 6.00 q2 60

Table 5 Number of vehicles

Buyer (i) Vehicles
with their types selected by
the buyers Type I ( j = 1) Type II ( j = 2) Type III ( j = 3)
1 3 0 0
2 3 0 0
3 2 0 0

chain profit. The prices of all the buyers are shown in Table 4 along with the safety
factors of all the buyers with the selected vehicles. The sub-batch sizes for the buyers
are also in the same table which shows that FTL is not necessary for the coordinated
supply chain because one of the buyers (buyer 1) selects the capacity of the vehicles
which is slightly greater than its sub-batch size. The model also shows that the buyers
prefer faster mode of transportation in order to reduce the transportation lead time.

7 Sensitivity Analysis

7.1 Sensitivity Analysis with Respect to Service Level

In this section, the effect of the service level on the inventory policy is analyzed by
changing the service level. For this analysis, we have used the same data as of given in
82 S. Yadav et al.

Table 6 Effect of service level on safety factor and total supply chain profit
Service level (%) Safety factor q1 q2 q3 n D T
k1 k2 k3
95.0 1.14 1.19 1.06 53 60 40 6 1920 35,579
97.5 1.46 1.51 1.40 53 60 40 6 1920 35,088
98.5 1.68 1.73 1.62 53 60 40 6 1920 34,758
99.0 1.84 1.89 1.79 53 60 40 6 1920 34,512
99.5 2.11 2.15 2.05 53 60 40 6 1920 34,119

Table 7 Effect of active and inactive service level on transportation lead time
Service level Safety factor Transportation lead time
Buyer 1 (%) Buyer 2 (%) Buyer 3 (%) k1 k2 k3 Buyer 1 Buyer 2 Buyer 3
95.0 92.0 94.0 1.14 0.95 0.97 4 5 3
95.0 50 94.0 1.14 0.00 0.97 4 10 3
50 50 94.0 0.00 0.00 0.97 8 10 3

the example section except that for the service level. First, all the buyers are assumed
to have a minimum service level of 95%; then, it is gradually increased for all the
buyers up to a value of 99.5%. It can be observed from Table 6 that as the service level
of the buyers is increased the safety factor for the buyers is also increased which is in
the line of the available literature that when the service level is increased the buyers
have to maintain a higher safety inventory. The result of maintaining a higher safety
inventory can be seen on the total supply chain profit which is decreasing with the
increase of the service level. However, the lot size and number of deliveries remain
constant.
Buyers of a supply chain may have different service levels; in that scenario, a
service level may not be active for all the buyers. This analysis has been conducted
by considering different service levels for different buyers as given in Table 7. The
first scenario where all the buyers have a nonzero safety factor shows that the service-
level constraint is active for all the buyers and all the buyers choose the transportation
lead time considering this constraint. However, the service-level constraint is inactive
for buyer 2 in scenario 2 and for buyers 1 and 2 in scenario 3. The safety factor for the
inactive constraint condition is zero. Also, for inactive constraint the buyers choose
comparatively cheap and slow transportation modes.

7.2 Sensitivity Analysis with Respect to σεi

The effect of demand uncertainty will be another important analysis for this study
from the realistic applications. For this analysis, the values of σεi for different buyers
Modeling a Supply Chain with Price-Dependent Stochastic Demand … 83

Table 8 Effect of σεi on safety factor and total supply chain profit
Change in σεi (%) Safety factor q1 q2 q3 n D T
k1 k2 k3
−20 1.02 1.08 0.95 53 60 40 5 1593 47,677
−10 0.93 1.08 0.85 71 80 53 4 1793 42,906
−5 0.96 1.11 0.88 71 80 53 4 1892 39,570
0 1.14 1.19 1.06 53 60 40 6 1992 35,579
+5 1.16 1.22 1.09 53 60 40 6 2091 31,019
+10 1.04 1.18 0.96 71 80 53 5 2191 25,930
+20 0.90 1.05 0.83 100 113 75 4 2390 13,784

have been changed by the amount shown in Table 8. Thus, the first scenario of Table 8
shows that the values of σεi have been reduced by 20% for all the buyers, while the last
scenario shows that the same has been increased by 20% in the last scenario. A larger
value of σεi indicates the larger demand uncertainty. Thus, the model predicts a lower
total supply chain profit for higher demand uncertainty and vice versa. Regarding the
safety factor, there is no such trend that can be seen directly from the table because of
the change in the values of sub-batch size and the number of sub-batches in one lot.
Thus, the true effect of the σεi can be seen collectively on the safety factor, sub-batch
size, and the number of sub-batches in one lot which is increasing with the increase
in σεi values. Thus, a higher demand uncertainty results in higher safety stock and a
lower total supply chain profit and vice versa.

8 Conclusion

The paper deals with the problem of the case company and develops the mathematical
model for its integrated inventory and pricing problem. The model also takes care
for the transportation of the item, its lead time, and service-level requirement. The
developed model is an MINLP which is very complex and is solved by LINGO 17.0
solver. The results show that a higher service level requires higher safety factor and
faster modes of transportation. The model is also validated for different service levels
for different buyers. Another analysis shows that a higher demand uncertainty is not
in favor of the supply chain.
Efficient heuristics can be developed in the future to solve the model. The model
can be extended with perishable product. The transportation of the perishable product
with the different technology-based refrigerated trucks (different deterioration rates)
will be an interesting study.

Acknowledgements The authors are grateful to the editors Dr. Rajendran C and Dr. Vipin B and
two anonymous referees for their valuable and constructive comments and suggestions on the earlier
84 S. Yadav et al.

versions of this article. The same has helped the authors significantly in improving the readability
and the contents of the present paper considerably.

References

Aardal, K., Jonsson, Ö., & Jönsson, H. (1989). Optimal inventory policies with service-level
constraints. Journal of the Operational Research Society, 40, 65–73. https://doi.org/10.1057/
jors.1989.6.
Abad, P. L. (1994). Supplier pricing and lot sizing when demand is price sensitive. European Journal
of Operational Research, 78, 334–354.
Agrawal, A. K., & Raju, A. (1996). Improved joint economic lot size model for a purchaser and a
vendor. In M. K. Khan, C. S. Wright, & R. Whalley (Eds.), Advanced manufacturing processes,
systems, and technologies (pp. 579–587). London: ASME Books.
Agrawal, A. K., & Yadav, S. (2020). Price and profit structuring for single manufacturer multi-
buyer integrated inventory supply chain under price-sensitive demand condition. Computers &
Industrial Engineering, 139,.
Ahmadizar, F., Zeynivand, M., Arkat, J. (2015). Two-level vehicle routing with cross-docking in a
three-echelon supply chain: A genetic algorithm approach. https://doi.org/10.1016/j.apm.2015.
03.005.
AlDurgam, M., Adegbola, K., & Glock, C. H. (2017). A single-vendor single-manufacturer inte-
grated inventory model with stochastic demand and variable production rate. International
Journal of Production Economics, 191, 335–350. https://doi.org/10.1016/j.ijpe.2017.05.017.
Arshinder, Kanda, A., & Deshmukh, S. G. (2008). Supply chain coordination: Perspectives,
empirical studies and research directions. International Journal of Production Economics, 115,
316–335. https://doi.org/10.1016/j.ijpe.2008.05.011.
Arshinder, K., Kanda, A., Deshmukh, S. G. (2011). A review on supply chain coordination: Coor-
dination mechanisms, managing uncertainty and research directions. In: T. Choi, T. Cheng
(Eds.), Supply chain coordination under uncertainty (pp. 39–82). International Handbooks on
Information Systems. https://doi.org/10.1007/978-3-642-19257-9.
Baker, A. R. C., Urban, T. L. (1988). A deterministic inventory system with an inventory-level-
dependent demand rate published by : Palgrave Macmillan. Journals on behalf of the Operational
Research Society Stable URL : http://www.jstor.org/stable/2583525.
Banerjee, A. (1986). A joint economic-lot-size model for purchaser and vendor. Decision Sciences,
17, 292–311.
Ben-Daya, M., Darwish, M., & Ertogral, K. (2008). The joint economic lot sizing problem: Review
and extensions. European Journal of Operational Research, 185, 726–742. https://doi.org/10.
1016/j.ejor.2006.12.026.
Ben-Daya, M., & Hariga, M. (2004). Integrated single vendor single buyer model with stochastic
demand and variable lead time. International Journal of Production Economics, 92, 75–80. https://
doi.org/10.1016/j.ijpe.2007.01.018.
Ben-Daya, M., Raouf, A. (1994). Operational research society is collaborating with JSTOR to
digitize, preserve, and extend access to Journal of the Operational Research Society. ® www.jst
or.org. 45, 579–582.
Braglia, M., Castellano, D., & Frosolini, M. (2016). A novel approach to safety stock management
in a coordinated supply chain with controllable lead time using present value. Applied Stochastic
Models in Business and Industry, 32, 99–112. https://doi.org/10.1002/asmb.2126.
Bushuev, M. A., Guiffrida, A., Jaber, M. Y., Khan, M. (2011). A review of inventory lot sizing
review papers. Management Research Review, 38, 283–298. http://dx.doi.org/10.1108/MRR-09-
2015-0216.
Modeling a Supply Chain with Price-Dependent Stochastic Demand … 85

Chang, C. T., & Lo, T. Y. (2009). On the inventory model with continuous and discrete lead time,
backorders and lost sales. Applied Mathematical Modelling, 33, 2196–2206. https://doi.org/10.
1016/j.apm.2008.05.028.
Chen, Z., & Sarker, B. R. (2017). Integrated production-inventory and pricing decisions for a single-
manufacturer multi-retailer system of deteriorating items under JIT delivery policy. International
Journal of Advanced Manufacturing Technology, 89, 2099–2117. https://doi.org/10.1007/s00170-
016-9169-0.
Choudhary, D., & Shankar, R. (2013). Joint decision of procurement lot-size, supplier selection,
and carrier selection. Journal of Purchasing and Supply Management, 19, 16–26. https://doi.org/
10.1016/j.pursup.2012.08.002.
Das Roy, M., Sana, S. S. (2011). Article information. Journal of Advanced Management Research,
14, 408–424. http://dx.doi.org/10.1108/MRR-09-2015-0216.
Giri, B. C., & Roy, B. (2015). A single-manufacturer multi-buyer supply chain inventory model
with controllable lead time and price-sensitive demand. Journal of Industrial and Production
Engineering, 32, 516–527. https://doi.org/10.1080/21681015.2015.1086442.
Glock, C. H. (2012a). The joint economic lot size problem: A review. International Journal of
Production Economics, 135, 671–686. https://doi.org/10.1016/j.ijpe.2011.10.026.
Glock, C. H. (2012b). Lead time reduction strategies in a single-vendor–single-buyer integrated
inventory model with lot size-dependent lead times and stochastic demand. International Journal
of Production Economics, 136, 37–44. https://doi.org/10.1016/j.ijpe.2017.09.012.
Goyal, S. K. (1976). An integrated inventory model for a single supplier-single customer problem.
International Journal of Production Research, 15, 107–111. https://doi.org/10.1080/002075477
08943107.
Goyal, S. K. (1988). A joint economic-lot-size model for purchaser and vendor: A comment.
Decision Sciences, 19, 236–241. https://doi.org/10.1111/j.1540-5915.1988.tb00264.x.
Goyal, S. K. (2000). On improving the single-vendor single-buyer integrated production inventory
model with a generalized policy. European Journal of Operational Research, 125, 429–430.
https://doi.org/10.1016/S0377-2217(99)00269-6.
Goyal, S. K., & Gunasekaran, A. (1995). An integrated production-inventory-marketing model for
deteriorating items. Computer and Industrial Engineering, 28, 755–762.
Goyal, S. K., & Gupta, Y. K. (1989). Integrated inventory models: The buyer-vendor coordination.
European Journal of Operational Research, 41, 261–269.
Goyal, S. K., & Nebebe, F. (2000). Determination of economic production-shipment policy for a
single-vendor-single-buyer system. European Journal of Operational Research, 121, 175–178.
https://doi.org/10.1016/S0377-2217(99)00013-2.
Gupta, O. K. (1994). An inventory model with lot-size dependent ordering cost. Production Planning
& Control, 5, 585–587. https://doi.org/10.1080/09537289408919532.
Hariga, M. A., Al-Ahmari, A., & Mohamed, A. R. A. (2007). A joint optimisation model for
inventory replenishment, product assortment, shelf space and display area allocation decisions.
European Journal of Operational Research, 181, 239–251. https://doi.org/10.1016/j.ejor.2006.
06.025.
Hill, R. M. (1997). The single-vendor single-buyer integrated production-inventory model with a
generalised policy. European Journal of Operational Research, 97, 493–499. https://doi.org/10.
1016/S0377-2217(96)00267-6.
Hill, R. M. (1999). The optimal production and shipment policy for the single-vendor single-
buyer integrated production-inventory problem. International Journal of Production Research,
37, 2463–2475. https://doi.org/10.1080/002075499190617.
Hoque, M. A. (2008). Synchronization in the single-manufacturer multi-buyer integrated inventory
supply chain. European Journal of Operational Research, 188, 811–825. https://doi.org/10.1016/
j.ejor.2007.05.019.
Jadidi, O., Jaber, M. Y., & Zolfaghari, S. (2017). Joint pricing and inventory problem with price
dependent stochastic demand and price discounts. Computer and Industrial Engineering, 114,
45–53. https://doi.org/10.1016/j.cie.2017.09.038.
86 S. Yadav et al.

Jha, J. K., & Shanker, K. (2009a). A single-vendor single-buyer production-inventory model with
controllable lead time and service level constraint for decaying items. International Journal of
Production Research, 47, 6875–6898. https://doi.org/10.1080/00207540802398053.
Jha, J. K., & Shanker, K. (2009b). Two-echelon supply chain inventory model with controllable lead
time and service level constraint. Computer and Industrial Engineering, 57, 1096–1104. https://
doi.org/10.1016/j.cie.2009.04.018.
Jha, J. K., & Shanker, K. (2013). Single-vendor multi-buyer integrated production-inventory model
with controllable lead time and service level constraints. Applied Mathematical Modelling, 37,
1753–1767. https://doi.org/10.1016/j.apm.2012.04.042.
Jokar, M. R. A., & Sajadieh, M. S. (2009). Optimizing a joint economic lot sizing problem with
price-sensitive demand. Transaction E Industrial Engineering, 16, 159–164.
Lau, A. H. L., Lau, H. S., & Wang, J. C. (2007). Some properties of buyback and other related
schemes in a newsvendor-product supply chain with price-sensitive demand. Journal of the
Operational Research Society, 58, 491–504. https://doi.org/10.1057/palgrave.jors.2602167.
Li, S., Ragu-Nathan, B., Ragu-Nathan, T. S., & Rao, S. S. (2006). The impact of supply chain
management practices on competitive advantage and organizational performance. International
Journal of Management Science, 34, 107–124. https://doi.org/10.1007/s00125-009-1560-z.
Li, Y., Xu, X., & Ye, F. (2011). Supply chain coordination model with controllable lead time and
service level constraint. Computer and Industrial Engineering, 61, 858–864. https://doi.org/10.
1016/j.cie.2011.05.019.
Liao, C., & Shyu, C. (1991). An analytical determination of lead time with normal demand. Inter-
national Journal of Operations & Production Management, 11, 72–78. https://doi.org/10.1108/
eum0000000001287.
Lin, Y. J., & Lin, H. J. (2015). Integrated supply chain model with price-dependent demand and
product recovery. Journal of Applied Science and Engineering, 18, 213–222. https://doi.org/10.
6180/jase.2015.18.3.01.
Lu, L. (1995). A one-vendor multi-buyer integrated inventory model. European Journal of
Operational Research, 81, 312–323. https://doi.org/10.1016/0377-2217(93)E0253-T.
Mandal, P., & Giri, B. C. (2015). A single-vendor multi-buyer integrated model with stock- and
price-dependent demand under consignment stock policy. International Journal of Services and
Operations Management, 20, 228–245. https://doi.org/10.1504/IJSOM.2015.067478.
Mills, E. S. (1959). Uncertainty and price theory. The Quarterly Journal of Economics, 73, 116–130.
Mou, Q., Cheng, Y., & Liao, H. (2017). A note on “lead time reduction strategies in a single-
vendor-single-buyer integrated inventory model with lot size-dependent lead times and stochastic
demand”. International Journal of Production Economics, 193, 827–831. https://doi.org/10.1016/
j.ijpe.2017.09.012.
Multanen, H. (2011). Lower inventory levels and costs due to reduction of transportation time To
develop a framework to understand the effects of the transportation time to inventory. University
of GAVLE.
Ng, B., Ferrin, B. G., & Pearson, J. N. (1997). The role of purchasing/transportation in cycle time
reduction. International Journal of Operations & Production Management, 17, 574–591. https://
doi.org/10.1108/01443579710167267.
Ouyang, L. Y., & Wu, K.-S. (1997). Mixture inventory model involving variable lead time with a
service level constraint. Computers & Operations Research, 24, 875–882.
Ouyang, L. Y., Wu, K. S., & Ho, C. H. (2004). Integrated vendor-buyer cooperative models with
stochastic demand in controllable lead time. International Journal of Production Economics, 92,
255–266. https://doi.org/10.1016/j.ijpe.2003.10.016.
Ouyang, L. Y., Yeh, N. C., & Wu, K. S. (1996). Mixture inventory model with backorders and lost
sales for variable lead time. Journal of the Operational Research Society, 47, 829–832. https://
doi.org/10.1057/jors.1996.102.
Pattnaik, M. (2013). Optimization in an instantaneous economic order quantity (EOQ) model incor-
porated with promotional effort cost, variable ordering cost and units lost due to deterioration.
Uncertain Supply Chain Manag., 1, 57–66. https://doi.org/10.5267/j.uscm.2013.06.005.
Modeling a Supply Chain with Price-Dependent Stochastic Demand … 87

Petruzzi, N. C., & Dada, M. (1999). Pricing and the newsvendor problem: A review with extensions.
Operations Research, 47, 183–194.
Ravindran, A., Phillps, D. T., & Solberg, J. J. (1987). Operations research: Principles and practice.
New York: Wiley.
Ray, S., Li, S., & Song, Y. (2005). Tailored supply chain decision making under price-sensitive
stochastic demand and delivery uncertainty. International Journal of Clinical and Experimental
Medicine, 5, 1873–1891.
Sajadieh, M. S., & Jokar, M. R. A. (2009). Optimizing shipment, ordering and pricing policies in
a two-stage supply chain with price-sensitive demand. Transportation Research Part E Logistics
and Transportation Review, 45, 564–571. https://doi.org/10.1016/j.tre.2008.12.002.
Sharafali, M., & Co, H. C. (2000). Some models for understanding the cooperation between the
supplier and the buyer. International Journal of Production Research, 38, 3425–3449. https://
doi.org/10.1080/002075400422734.
Taleizadeh, A. A., Kalantari, S. S., & Cárdenas-Barrón, L. E. (2016). Pricing and lot sizing for an
EPQ inventory model with rework and multiple shipments. Top, 24, 143–155. https://doi.org/10.
1007/s11750-015-0377-9.
Taleizadeh, A. A., Niaki, S. T. A., Shafii, N., Meibodi, R. G., & Jabbarzadeh, A. (2010). A particle
swarm optimization approach for constraint joint single buyer-single vendor inventory problem
with changeable lead time and (r, Q) policy in supply chain. International Journal of Advanced
Manufacturing Technology, 51, 1209–1223. https://doi.org/10.1007/s00170-010-2689-0.
Taleizadeh, A. A., & Noori-Daryan, M. (2016). Pricing, manufacturing and inventory policies for
raw material in a three-level supply chain. International Journal of Systems Science, 47, 919–931.
https://doi.org/10.1080/00207721.2014.909544.
Tiwari, S., Sana, S. S., & Sarkar, S. (2018). Joint economic lot sizing model with stochastic demand
and controllable lead-time by reducing ordering cost and setup cost. Revista de la Real Academia
de Ciencias Exactas, Físicas y Naturales. Serie A. Matemáticas, 112, 1075–1099. https://doi.org/
10.1007/s13398-017-0410-y.
Urban, T. L. (1992). Decisions 22, 85–93.
Wangsa, I. D., & Wee, H. M. (2018). An integrated vendor–buyer inventory model with trans-
portation cost and stochastic demand. International Journal of Systems Science: Operations &
Logistics, 5, 295–309. https://doi.org/10.1080/23302674.2017.1296601.
Whitin, T. M. (1954). Inventory control research: A survey. Management Science, 1, 32–40. http://
dx.doi.org/10.1287/mnsc.1.1.32Full.
Whitin, T. M. (1955). Inventory control and price theory. Management Science, 2, 61–68. https://
doi.org/10.1287/mnsc.2.1.61.
Wu, K., Ouyang, L., & Yang, C. (2009). Coordinating replenishment and pricing policies for non-
instantaneous deteriorating items with price-sensitive demand. International Journal of Systems
Science, 40, 1273–1281. https://doi.org/10.1080/00207720903038093.
Yadav, S., Agrawal, A. K., & Vora, M. K. (2020). A single manufacturer multiple buyers inte-
grated production-inventory model with third-party logistics. International Journal of Business
Performance and Supply Chain Modelling, 11(2), 91–127.
You, P. S. (2005). Inventory policy for products with price and time-dependent demands. Journal of
the Operational Research Society, 56, 870–873. https://doi.org/10.1057/palgrave.jors.2601905.
Zanoni, S., Mazzoldi, L., Zavanella, L. E., & Jaber, M. Y. (2014). A joint economic lot size model
with price and environmentally sensitive demand. Production & Manufacturing Research, 2(1),
341–354.
Optimal and Heuristic Profit Sharing
Using Sales Rebate Contract in a
Multi-level Supply Chain

I. Thomson, Raichel Elsa Thomas, K. Susithra, S. Shobana, Arshinder Kaur,

and Chandrasekharan Rajendran

Abstract A supply chain (SC) is an interdependent network of many SC members

who add value to the product successively until it gets to the final stage and attains
utility. One of the most important problems in any supply chain (SC) consisting of
organizations is the sub-optimization due to the fact that the SC decision making is
distributed over various players. An interesting solution to the issue, which leaves the
decision power and supply chain structure intact, is the use of contract mechanisms.
Ideally, contract mechanisms ensure that the whole SC is optimized as if it were
a single unit (coordination) and is designed such that all the players benefit from
working together through the coordination mechanism (win-win). Once coordination
is in place, the next challenge is to devise some mechanism to share the profits gained
due to coordination. To enable coordination among the SC members, sales rebate
contracts in a multi-level SC using fair share mechanism have been devised. In
the sales rebate contract, the supplier charges the buyer wholesale price per unit

I. Thomson
Reckitt Benckiser Arabia, Dubai 119481, UAE
e-mail: [email protected]
R. E. Thomas
Department of Mechanical Engineering, Indian Institute of Technology BHU, Varanasi 221005,
India
e-mail: [email protected]
K. Susithra · S. Shobana
Department of Industrial Engineering, College of Engineering Guindy, Anna University, Chennai
600025, India
e-mail: [email protected]
S. Shobana
e-mail: [email protected]
A. Kaur · C. Rajendran (B)
Department of Management Studies, Indian Institute of Technology Madras, Chennai 600036,
India
e-mail: [email protected]
A. Kaur
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 89
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_5
90 I. Thomson et al.

purchased but then gives rebate to the buyer per unit sold above some threshold.
The objective of this paper is to determine the optimal rebates given to downstream
SC members and threshold value of sales, which maximizes the SC profit as well
as fairly shares the profits among the SC members. This objective can be achieved
by minimizing the deviation between proportion of each member’s value addition
and proportion of each member’s respective share in the total profit by coordinating
through sales rebate contract. An analytical model is developed to formulate the
supply chain coordination problem and simultaneously formulate the optimization
problem to achieve fair distribution of profit among the SC members. Three heuristic
search techniques based on genetic algorithm and simulated annealing have been
proposed to solve the problem.

Keywords Multi-level supply chain · Supply chain coordination · Sales rebate

contracts · Profit sharing · Analytical model · Optimization · Genetic algorithm ·
Simulated annealing

1 Introduction

A supply chain is termed a coordinated supply chain, when all the supply chain
members work toward a common objective of maximizing the total supply chain
profitability. Generally, the supply chain members being primarily concerned with
optimizing their own objectives tend to sell/buy an order quantity which is optimum
for them and that self-serving focus often results in poor performance of the whole
supply chain.
Supply chain management (SCM) is a set of approaches utilized to effectively
integrate suppliers, manufacturers, warehouses and retailers which are interdepen-
dent entities. The current challenges of globalization and product proliferation in
supply chain demands supply chains to be considered as a single system to effec-
tively manage the flow of resources and information between supply chain members.
The supply chain members are expected to align their respective objectives with that
of the whole supply chain to minimize the system-wide cost and satisfy the service-
level requirement (Simchi-Levi et al., 2008).
SCM is a philosophy of managing the supply chain (externally) as well as man-
aging a company (internally). Nalla (2008) observed SCM as the coordination of
different business entities in the supply chain to reduce waste (costs), create value
to customers and thus enhance revenues. Coordination in the above definition refers
to managing challenges due to inter-dependencies among business entities by align-
ing goals and process/functions. According to Nalla (2008), organizations seek to
achieve coordination through different approaches. The first step toward establish-
ing coordination might be to share information between the entities in the SC. The
sharing of information is indeed a necessary condition but may not be sufficient to
achieve coordination and improve overall SC performance.
Besides sharing of information, Nalla (2008) observed organizations could use
two other approaches to achieve coordination. The first approach is to modify the
Optimal and Heuristic Profit Sharing Using Sales Rebate … 91

governance structure of the trading relationship, for example, by modifying the own-
ership (i.e., “who owns what”) and/or by modifying decision rights (i.e., “who decides
what”). Modifying the governance structure works only when the owner gets the deci-
sion rights over the functional people. This approach is most difficult to implement
within and across companies in SC. The second approach for achieving coordina-
tion within the SC is to modify the terms of trade. The modification of the terms
is achieved through incentive schemes or contracts over certain trade parameters
(variables). This approach aims to achieve coordination among business entities by
providing incentives to share risks and/or rewards. There are several contract mecha-
nisms that can be designed and used to make sure that the independent decisions made
by business entities optimize the overall performance of the whole chain (in such a
case, we say that the mechanism coordinates the chain). The contracts must enable
all independent business entities to improve their performance when compared to an
uncoordinated situation. In such a case, we say that the contract mechanism leads to
a win-win situation.
In supply chain coordination, the whole supply chain arrives at a consensus of
how much to order (optimal order quantity of whole supply chain). However, firms
lack the incentive to implement those actions. To create that incentive, the firms can
adjust their terms of trade via a contract that establishes a transfer payment scheme.
This paper explores profit sharing between supply chain members in a centralized
supply chain. A systematic framework for an analytical model is built to find out
the optimal rebate and threshold values that ensure a transparent and trustworthy
supply chain coordination. Contrary to the initial agreement where a decentralized
supply chain exists, here a centralized supply chain is considered. The profit is being
distributed among the supply chain members based on the value they add to the
product.
The remainder of the paper is organized as follows: In Sect. 2, we give details
of various coordination mechanisms available in the literature and introduce the
research problem. In Sect. 3, an analytical model of multi-level (stage) supply chain
is introduced and a three-level supply chain is discussed in detail. In Sect. 4, we
present the overall problem of determining the decision variables and sharing rewards
among supply chain members using a heuristic technique. In Sect. 5, meta-heuristic
search techniques such as genetic algorithm and simulated annealing are discussed
in order to solve the problems. In Sect. 6, various experiment settings are discussed,
and finally in Sect. 7 the study is concluded.

2 Literature Review

Cachon (2003) defined supply chain coordination as follows: A contract is said

to coordinate the supply chain if the set of supply chain optimal actions is in
Nash equilibrium; i.e., no firm has a profitable unilateral deviation from the set
of supply chain optimal actions. Tsay (1999) defined supply chain contracts for-
mally as the rules for transactions between the supply chain actors and later utilized
92 I. Thomson et al.

incentives (risks and rewards) to make the SC members’ decisions coherent with
each other. A number of different contract types are identified in the literature such
as buyback contracts (Sang, 2016), revenue sharing contracts, quantity flexibility
contracts and sales rebate contracts (Emmons & Gilbert, 1998; Koulamas, 2006;
Lian & Deshmukh, 2009’ Krishnan et al., 2004; Taylor, 2002). Generally, the analy-
sis of contracts offers guidance in negotiating the terms of the relationship between
the supplier and the buyer. The contracts are designed to sort out conflicts that may
crop up in the future.
The focus of this paper is sales rebate contract where the supplier charges a
wholesale price per unit purchased and gives the retailer a rebate per unit sold above
a threshold fixed by the supplier. The objective of this paper is to explore the appli-
cability of sales rebate contract in a multi-level supply chain. Sales rebate contracts
can be implemented as promotional trade incentives and especially as alternative
sales promotion to price discounts (Arcelus et al., 2008). Rebates in a multi-level
supply chain have been less explored, as increase in the number of SC members
increases the complexity in optimizing the key parameters that are involved for the
intermediate members in the SC.
A rebate is different from an order quantity discount as it only applies to items
sold to end users. Hence, a rebate contract is more efficient than an order quantity
discount because it provides a direct incentive for retailers to increase sales (Wong et
al., 2009). The supplier needs to know the exact quantity sold by the retailer in order
to pay the rebate, but difficulties arise when the supplier cannot acquire the retailer’s
sales data directly. On the other hand, the data obtained from the retailer may not be
authentic as the retailer may claim more rebates than what the actual sales allow. The
ability of the manufacturer to arrive at the best price and rebate is conditioned by
the level of information that the manufacturer has about the uncertainty faced by the
retailer as well as the retailer’s cost and demand function (Arcelus, 2007). Krishnan
et al. (2004) proposed a buyback with the sales rebate contract to coordinate the
news vendor with a fixed price but effort-dependent demand. The rebate induces the
retailer to price too low (in an effort to generate sales above the rebate threshold),
but a buyback induces the retailer to price too high, so it is possible to counteract the
deleterious effects of the rebate on price. Taylor (2002) discussed a rebate contract
for a supply chain composed of one supplier and one retailer, in which retailers
were allowed to determine their order quantities consistent with the optimal order
quantities so as to optimize the global supply chain’s profit. In the auto industry,
channel rebates are termed as dealer incentives.
In the literature on supply chain contract mechanisms, the main focus seems to
be on a situation with only two decision makers: a buyer and a supplier, however
in reality a SC usually consists of more than two entities. Whereas, in reality a SC
usually consists of more than two entities. A natural question therefore is how the
fundamental ideas underlying contract mechanisms can be generalized to settings
with a multi-echelon supply chain (van der Rhee et al., 2010). There is a need
to capture the problem of multi-level supply chains and to analyze how contract
decision variables can be determined at each interface between supply chain entities.
The complexity of managing multi-level supply chain and hence the contract decision
Optimal and Heuristic Profit Sharing Using Sales Rebate … 93

variables at two interfaces has been captured in the analytical model proposed in the
next section.
In the literature, the emphasis has been given to achieve supply chain coordination
with uniform order quantity, with increase in profits of whole supply chain as a
performance indicator of coordinated supply chain. Even if the supply chain contract
helps in improving the system-wide (total supply chain) profits, the challenge still
remains in determining the fair share of the benefits among supply chain entities. In
this paper, we have analyzed the sales rebate contracts in multi-level supply chain
considering all the members to be a part of the supply chain. To analyze the utility of
sales rebate contracts, we compare the expected profits of all supply chain members as
well as total expected supply chain profits with the case when supply chain members
act independently. The savings gained by adopting sales rebate contracts have to
be shared fairly among supply chain members based on the value addition done by
respective supply chain members. The model proposed in this paper simultaneously
determines the decision variables of the sales rebate contracts and the fair share of
savings among supply chain members. Further, the fair share of profits is determined
along with the decision variables of rebate contracts in a three-level supply chain
using meta-heuristic techniques; genetic algorithm (GA) and simulated annealing
(SA) have been adopted.
Genetic algorithms have the ability to handle nonlinear interactions in the model
to arrive at the near-optimum values of the decision variables. GAs have been applied
in various supply chain problems like supply chain design problems (Chang, 2010),
determination of base stock values in serial supply chain (Daniel & Rajendran, 2006),
information sharing in supply chain, vendor managed inventory (VMI) in supply
chain (Nachiappan & Jawahar, 2007) and comparison of periodic review order up to
policy and continuous review policy comparison in a serial supply chain (Sethupathi
et al., 2014). The application of GAs in SC contract area seems to be less explored
in the literature.
Simulated annealing (SA) heuristic generates near-optimal solutions to combi-
natorially intractable problems. The heuristic, however, is generic and has to be
modified in the context of the specific problem under study. The main reason for
which SA heuristic is supposed to give a good solution is due to the fact that it also
accepts inferior solutions (of course, with a certain acceptance criterion) in search
of a good solution. Many papers have reported successful applications of simulated
annealing in scheduling, and the same can be extended to SC contracts.

3 Model for Sales Rebate Contracts in a Multi-level Supply

Chain

Contracts have been widely explored, and they help in coordinating the supply chain
as well as improving the performance of the whole supply chain. Most of the contracts
have been explored for two-level supply chain. In general, a typical supply chain has
more than two members dealing with each other. When we consider more than two
94 I. Thomson et al.

Fig. 1 Supply chain model

members in the supply chain system, the decisions at one interface are related to the
decision at the other interface of the supply chain (Fig. 1).

3.1 Assumptions

The sequence of events in end customer supply chain model are as follows; the most
downstream member entity 1 of the supply chain based on the end customer demand
decides to order q units from entity 2. The same order size is passed on from entity
2 to entity 3 all the way to entity N . Subsequently, the amount q is shipped from
entity N all the way to entity 1. Finally, the customer demand occurs and the buyer
sells the amount min(q, customer demand) to the end customer. We now build an
N-echelon model where we can imagine the entities in the supply chain to be retailer,
distributor, manufacturer, supplier and so on.
We consider the following basic assumptions:
• An N-level supply chain is considered comprising a supplier, a manufacturer, a
distributor, a retailer and so on.
• A single period model with short life cycle product is assumed where the end of
period inventory must be disposed off.
• The demand is stochastic in nature.
• Cost/price data are known in advance.
• A single product flows through the supply chain.
• Information lead time is negligible or zero.
• Base stock level at every installation takes discrete integer values.
• There is no lot size or discount policy for any installation.
• There are no holding cost, shortage cost and transportation cost.
• All installations have infinite capacity.
• The most upstream installation is linked to an infinite supply source.
• The end customer demand information is shared across the SC.
Optimal and Heuristic Profit Sharing Using Sales Rebate … 95

3.2 Notations and Terminologies

All prices are expressed in standard unit cost

P: Purchase cost
αi P: Price at which units are sold by entity i to entity i − 1 such that
αi−1 ≥ αi ∀i ∈ {1, 2, 3...N }
s: Salvage value at retailer
qd : Quantity ordered by entity 1 to entity 2, which is passed to entities upstream
qd∗ : Decentralized optimal order quantity; optimal order quantity of entity 1
qc : Order quantity of SC, where all the supply chain members are considered
to be a part of one system
qc∗ : Centralized optimal order quantity; optimal order quantity of SC system
ti : Threshold fixed by entity i to entity i − 1
ri : Rebate paid by entity i to entity i − 1 for each unit sold above the threshold
X: Denotes the following scenarios—Case A—decentralized ,
Case B—centralized , Case C—coordinated
πiX : Expected profit of entity i in case X {X ∈ A, B, C}
πX: Expected profit of whole supply chain in case X {X ∈ A, B, C}
i: Entity in the supply chain {i ∈ 1, 2, 3 . . . N }
P Si : Proportion of increase in profit of entity i due to coordination w.r.t. total
profit gain of SC
Vi : Proportion of value addition by the SC entity i w.r.t. the total value addition
of SC
S(q): Expected sale.

3.3 Analytical Model

The expected sales S(q) as per Cachon (2003) are given as follows:

q ∞
S(q) = x · f (x)dx + q f (x)dx
0 q
q
= q(1 − F(q)) + x f (x)dx
0
q
S(q) = q − F(x)dx
0

S  (q) = 1 − F(q) = F(q), (1)

96 I. Thomson et al.

where x is stochastic end customer demand, f (x) is probability density function of

end customer demand and F(x) is cumulative distribution function of end customer
demand. The demand is assumed to follow a uniform distribution with parameters
a, b, though not restrictive. This is done for generalized understanding. The cumu-
lative distribution is, without loss of generality, assumed to be the following:
⎧
⎪
⎨0 if x < a
(x−a)
F(x) = ) if a ≤ x ≤ b
⎪ (b−a)
⎩
1 if x > b

The expected sales is

q
S(q) = q − F(x)dx
a
q
x −a
=q− dx
b−a
a
 
(q − a)2
=q−
2(b − a)

Therefore, the expected sales, when there is decentralized decision making, qd , are
written as follows:
 
(qd − a)2
S(qd ) = qd − (2)
2(b − a)

Similarly, the expected sales when there is centralized decision making of the
order quantity from supply chain perspective, qc can be written as in Eq. (3)
 
(qc − a)2
S(qc ) = qc − (3)
2(b − a)

3.4 Model Formulation

In this paper, the proposed model describes three cases: First is a case where there is
a decentralized decision making, wherein the retailer decides the order quantity, i.e.,
the decentralized optimal order quantity qd∗ ; second where there is a centralized deci-
sion making wherein the entities in the supply chain order a common quantity, i.e.,
centralized optimal order quantity qc∗ ; and thirdly where supply chain coordination
is facilitated with the installation of sales rebate contracts between every consecutive
pair of entities to order the centralized optimal order quantity. The expected profits
Optimal and Heuristic Profit Sharing Using Sales Rebate … 97

of each member and the whole supply chain for three different cases are analyzed
below.

3.4.1 Case A—Decentralized Supply Chain

The profit optimization of entity 1 alone is considered in this case. This function is
proved to be an increasing concave function in (qd ) using the first- and second-order
derivatives. The optimum order quantity is computed from this function. Each entity
of the supply chain orders the optimal order quantity of entity 1 which is called
decentralized order quantity qd∗ .
The profit function of retailer is

π1A (qd ) = α1 P S(qd ) − α2 Pqd + s(qd − S(qd ))

It can be observed that the function in equation is concave. The optimum order
quantity of entity 1, qd∗ is computed from the above function. Entity 2, entity 3, all
the way up to entity N ship qd∗ , and their profits are based on this order quantity.
 
α1 P − α2 P
qd∗ =F −1
(4)
α1 P − s

where qd∗ is positive. The profit function of entity 1 with qd∗ of entity 1 can be written
as shown below

π1A (qd∗ ) = α1 P S(qd∗ ) − α2 Pqd∗ + s(qd∗ − S(qd∗ ))

The profit functions w.r.t. entities 2 to N can be written as follows:

π2A (qd∗ ) = α2 Pqd∗ − α3 Pqd∗

π3A (qd∗ ) = α3 Pqd∗ − α4 Pqd∗ , so on upto,

π NA−1 (qd∗ ) = α N −1 Pqd∗ − α N Pqd∗

π NA (qd∗ ) = α N Pqd∗ − Pqd∗

Total expected profit of the supply chain in Case A is as follows:

N
πiA (qd∗ ) = π1A (qd∗ ) + π2A (qd∗ ) + π3A (qd∗ ) · · · + π NA−1 (qd∗ ) + π NA (qd∗ )
i=1

π (qd∗ ) = α1 P S(qd∗ ) − Pqd∗ + s(qd∗ − S(qd∗ ))

A
(5)
98 I. Thomson et al.

In this case of decentralized decision making, the optimal order quantity is deter-
mined as per the interest of entity 1 and not in the interest of whole supply chain. In
the next subsection, another case is analyzed, where each SC member is considered
to be a part of a single supply chain system.

3.4.2 Case B—Centralized Supply Chain

In this case, the SC members can be considered as a part of one organization and
work to maximize the total supply chain profit. It is assumed in this case that SC
members will coordinate and determine the centralized optimal order quantity, which
is optimal to the whole supply chain system. The expected profit of entities 1 to N
(when they all agree to order (qc )) is as follows:

π1B (qc ) = α1 P S(qc ) − α2 Pqc + s(qc − S(qc ))

π2B (qc ) = α2 Pqc − α3 Pqc

π3B (qc ) = α3 Pqc − α4 Pqc , so on upto,

π NB−1 (qc ) = α N −1 Pqc − α N Pqc

π NB (qc ) = α N Pqc − Pqc

The total expected profit of the supply chain will be as presented in the below
equation

N
πiB (qc ) = π1B (qc ) + π2B (qc ) + π3B (qc ) · · · + π NB−1 (qc ) + π NB (qc )
i=1

π B (qc ) = α1 P S(qc ) − Pqc + s(qc − S(qc ))

In Case B, we have an increasing concave function. The optimum order quantity

of whole supply chain is given in Eq. (6). Each member of the supply chain orders
the optimal order quantity for the whole supply chain given by the below equation
 
α1 P − P
qc∗ =F −1
(6)
α1 P − s

where qc∗ (optimal order quantity of the supply chain) is positive. Substituting qc∗ in
all the profit equations, we get the expected profits of all SC entities. The expected
profit functions for entities in the supply chain are given in the following equations.

π1B (qc∗ ) = α1 P S(qc∗ ) − α2 Pqc∗ + s(qc∗ − S(qc∗ ))

Optimal and Heuristic Profit Sharing Using Sales Rebate … 99

π2B (qc∗ ) = α2 Pqc∗ − α3 Pqc∗

π3B (qc∗ ) = α3 Pqc∗ − α4 Pqc∗ , so on upto,

π NB−1 (qc∗ ) = α N −1 Pqc∗ − α N Pqc∗

π NB (qc∗ ) = α N Pqc∗ − Pqc∗

Total expected supply chain profit is

N
πiB (qc∗ ) = π1B (qc∗ ) + π2B (qc∗ ) + π3B (qc∗ ) · · · + π NB−1 (qc∗ ) + π NB (qc∗ )
i=1

π B (qc∗ ) = α1 P S(qc∗ ) − Pqc∗ + s(qc∗ − S(qc∗ )) (7)

Lemma 1 The centralized optimal order quantity qc∗ is greater than the decentral-
ized optimal order quantity qd∗ .
We build the following expression/constraints

α1 P − P α1 P − α2 P
> since P < αi P for i ∈ {1, . . . N }
α1 P − s α1 P − s

Using Eqs. (4), (5), (6) and (7), we get

f (qc∗ ) > f (qd∗ )

Therefore, when we calculate the quantities using inverse, we will have

qc∗ > qd∗ (8)

Thus, the qc∗ is greater than the qd∗ .

Lemma 2 The expected sale of centralized optimal order quantity S(qc∗ ) is greater
than the expected sale of decentralized optimal order quantity S(qd∗ ).
We build the following inequality which is valid from Lemma 1:
   
(qc∗ − a)2 (qd∗ − a)2
qc∗ − > qd∗ − since qc∗ > qd∗
2(b − a) 2(b − a)

In the above inequality, substitutions can be made using Eqs. (2) and (3), resulting
in the following relationship:
100 I. Thomson et al.

S(qc∗ ) > S(qd∗ ) (9)

Lemma 3 The optimal order quantity qx∗ is greater than or equal to the expected
sale of optimal order quantity S(qx∗ ) where x is either c, centralized order quantity
qc∗ , or d, decentralized order quantity qd∗ .
This can be inferred from the following equations:
 
(qc∗ − a)2
qc∗ > S(qc∗ ) considering S(qc∗ ) = qc∗ −
2(b − a)
 
(qd∗ − a)2
qd∗ > S(qd∗ ) considering S(qd∗ ) = qd∗ −
2(b − a)

Therefore, we have

qc∗ > S(qc∗ ) (10)

qd∗ > S(qd∗ ) (11)

The expected sales of SC are greater with qc∗ than the expected sales with the qd∗ .
On comparing the expected profits of all the SC members for Case A and Case B,
we note that the retailer may incur losses as per Lemma 1, but there is some scope
for improving the sales as seen from Lemma 2. Whereas the expected profits of the
entity 2 and the entity 3 may improve in Case B as compared to Case A, the expected
profits of the whole supply chain is also more in Case B. The retailer may not agree
to coordinate with other SC members as she/he may incur losses by ordering qc∗ . To
encourage the retailer to order qc∗ , the upstream members can devise some mechanism
so that no SC member would incur losses and at the same time centralized supply
chain system should also not be compromised. In the next scenario Case C, sales
rebate contracts will be introduced and the coordination achieved will be analyzed.

3.4.3 Case C—Supply Chain Coordination with Centralized Optimal

Order Quantity and Sales Rebate Contracts

In this case, each member of the supply chain orders qc∗ . The upstream member will
fix some threshold value of sales above which the downstream member can avail
rebates from the upstream member. If entity i − 1 sells more than the threshold ti
fixed by entity i, entity i pays to entity i − 1 a rebate for each unit sold that is above
the threshold limit. The profit functions with sales rebate contracts for the entities in
the SC are as follows.
The profit functions are as follows:
+
π1C (qc∗ ) = α1 P S(qc∗ ) − α2 Pqc∗ + s qc∗ − S(qc∗ ) + r2 S(qc∗ ) − t2
Optimal and Heuristic Profit Sharing Using Sales Rebate … 101

+ +
π2C (qc∗ ) = α2 Pqc∗ − α3 Pqc∗ − r2 S(qc∗ ) − t2 + r3 S(qc∗ ) − t3

+ +
π3C (qc∗ ) = α3 Pqc∗ − α4 Pqc∗ − r3 S(qc∗ ) − t3 + r4 S(qc∗ ) − t4 ,

so on up to,
+
π NC −1 (qc∗ ) = α N −1 Pqc∗ − α N Pqc∗ − r N −1 (S(q∗c ) − t N −1 )+ + r N S(qc∗ ) − t N

+
π NC (qc∗ ) = α N Pqc∗ − Pqc∗ − r N S(qc∗ ) − t N

Total expected profit of the supply chain is

N
πiC (qc∗ ) = π1C (qc∗ ) + π2C (qc∗ ) + π3C (qc∗ ) · · · + π NC −1 (qc∗ ) + π NC (qc∗ )
i=1

π C (qc∗ ) = α1 P S(qc∗ ) − Pqc∗ + s qc∗ − S(qc∗ ) (12)

An interesting observation here is that the expected profits in scenarios B and C

are equal.
Here, the total supply chain profit is the same as that obtained in Eq. (7). The
transfer pay between the supply chain entities in this case will be related to sales
rebates and threshold values, which are internal to the supply chain. Hence, it shows
that total expected supply chain profits will be the same, but the way that supply chain
entities share risks and rewards is the question. In this case, we have the following
decision variables: ri ∀{i ∈ 2, 3, . . . N }, ti ∀{i ∈ 2, 3, . . . N }. Another decision vari-
able in this problem is to determine the share of each supply chain entity, due to gain
in the expected profits of the whole supply chain.
In this paper, we propose a model which determines the decision variables pertain-
ing to the sales rebate contracts (rebates and threshold values) as well as a fair share
of gain in the total SC expected profits among SC members due to the rebates. Each
supply chain entity adds value to the product in different forms throughout the supply
chain. One of the methods to fairly share the gain in the total supply chain expected
profit is based on the value addition to the product (difference between selling price
and cost incurred) by the respective SC entities. The profit sharing objective is to
minimize the deviation between the proportion of profit gained by SC entities and
the proportion of value addition by the same SC entities.

3.5 Objective Function

The profits gained by using sales rebate contracts for the whole SC must be fairly
shared among the entities based on their respective value addition to the product
(difference between selling price and cost incurred).
102 I. Thomson et al.

N
Objective: Minimize Z = |P Si − Vi |
i=1
= |P S1 − V1 | + |P S2 − V2 | + |P S3 − V3 | · · · + |P S N −1 − VN −1 | + |P S N − VN |
(13)

where

π1C − π1A π2C − π2A π3C − π3A

P S1 = , P S 2 = , P S 3 = ...,
πC − π A πC − π A πC − π A
π C − π NA−1 π NC − π NA
P S N −1 = N −1 , P S N =
πC − π A πC − π A
α1 P − α2 P α2 P − α3 P α3 P − α4 P
V1 = , V2 = , V3 = ...,
α1 P − P α1 P − P α1 P − P
α N −1 P − α N P αN P − P
VN −1 = , VN =
α1 P − P α1 P − P

3.6 Constraints

Since the SC entities will order the qc∗ only if their profits are more in Case C than in
Case A and the rebates help improve the profits of each entity while ordering qc∗ , we
have the following inequalities which have to be satisfied to coordinate the supply
chain.

πC ≥ π A (14)
πiC ≥ πiA where i ∈ {1, . . . N } (15)

From constraint 14, we have

πC ≥ π A

α1 P S(qc∗ ) − Pqc∗ + s qc∗ − S(qc∗ ) ≥ α1 P S(qd∗ ) − Pqd∗ + s qd∗ − S(qd∗ )

α1 P(S(qc∗ ) − S(qd∗ )) − P(qc∗ − qd∗ ) + s(qc∗ − S(qc∗ ) − qd∗ + S(qd∗ )) ≥ 0

From constraint 15, for entity 1 we have

Optimal and Heuristic Profit Sharing Using Sales Rebate … 103

π1C ≥ π1A
α1 P S(qc∗ ) − α2 Pqc∗ + s(qc∗ − S(qc∗ ))
+
+r2 S(qc∗ ) − t2 ≥ α1 P S(qd∗ ) − α2 Pqd∗ + s(qd∗ − S(qd∗ ))
α1 P(S(qc∗ ) − S(qd∗ )) − α2 P(qc∗ − qd∗ ) + s(qc∗ − S(qc∗ ) − qd∗ + S(qd∗ ))
+
+r2 S(qc∗ ) − t2 ≥0
+
r2 S(qc∗ ) − t2 ≥ (α2 P − s)(qc∗ − qd∗ ) − (α1 P − s)(S(qc∗ ) − S(qd∗ ))

From constraint 15, for entity 2 we have

π2C ≥ π2A
+ +
α2 Pqc∗ − α3 Pqc∗ − r2 S(qc∗ ) − t2 + r3 S(qc∗ ) − t3 ≥ α2 Pqd∗ − α3 Pqd∗

+ +
α2 P{qc∗ − qd∗ } − α3 P(qc∗ − qd∗ ) − r2 S(qc∗ ) − t2 + r3 S(qc∗ ) − t3 ≥0

+ +
r3 S(qc∗ ) − t3 ≥ r2 S(qc∗ ) − t2 − (α2 P − α3 P)(qc∗ − qd∗ )

From constraint 15, for entity 3 we have

π3C ≥ π3A
+ +
α3 Pqc∗ − α4 Pqc∗ − r3 S(qc∗ ) − t3 + r4 S(qc∗ ) − t4 ≥ α3 Pqd∗ − α4 Pqd∗

+ +
α3 P(qc∗ − qd∗ ) − α4 P(qc∗ − qd∗ ) − r3 S(qc∗ ) − t3 + r4 S(qc∗ ) − t4 ≥0

+ +
r4 S(qc∗ ) − t4 ≥ r3 S(qc∗ ) − t3 − (α3 P − α4 P)(qc∗ − qd∗ )

From constraint 15, for entity N−1 we have

π NC −1 ≥ π NA−1
+
α N −1 Pqc∗ − α N Pqc∗ − r N −1 S(qc∗ ) − t N −1
+
+r N S(qc∗ ) − t N ≥ α N −1 Pqd∗ − α N Pqd∗
+
α N −1 P(qc∗ − qd∗ ) − α N P(qc∗ − qd∗ ) − r N −1 S(qc∗ ) − t N −1
+
+r N S(qc∗ ) − t N ≥0
+ +
r N S(qc∗ ) − t N ≥ r N −1 S(qc∗ ) − t N −1
−(α N −1 P − α N P)(qc∗ − qd∗ )

From constraint 15, for entity N we have

104 I. Thomson et al.

π NC ≥ π NA
+
α N Pqc∗ − Pqc∗ − r N S(qc∗ ) − t N ≥ α N Pqd∗ − Pqd∗

+
α N P(qc∗ − qd∗ ) − P(qc∗ − qd∗ ) − r N S(qc∗ ) − t N ≥0

+
r N S(qc∗ ) − t N ≤ (α N P − P)(qc∗ − qd∗ )

From the above inequalities, we get

(α2 P − s)(qc∗ − qd∗ ) − (α1 P − s)(S(qc∗ ) − S(qd∗ ))

+
≤ r2 S(qc∗ ) − t2 ≤ (α2 P − P)(qc∗ − qd∗ ) (16)
(α3 P − s)(qc∗ − qd∗ ) − (α1 P − s)(S(qc∗ ) − S(qd∗ ))
+
≤ r3 S(qc∗ ) − t3 ≤ (α3 P − P)(qc∗ − qd∗ ) (17)
(α N −1 P − s)(qc∗ − qd∗ ) − (α1 P − s)(S(qc∗ ) − S(qd∗ ))
+
≤ r N −1 S(qc∗ ) − t N −1 ≤ (α N −1 P − P)(qc∗ − qd∗ ) (18)
(α N P − s)(qc∗ − qd∗ ) − (α1 P − s)(S(qc∗ ) − S(qd∗ ))
+
≤ r N S(qc∗ ) − t N ≤ (α N P − P)(qc∗ − qd∗ ) (19)

It is evident that the generalized expression for the above is

(αi P − s)(qc∗ − qd∗ ) − (α1 P − s)(S(qc∗ ) − S(qd∗ ))

+
≤ ri S(qc∗ ) − ti ≤ (αi P − P)(qc∗ − qd∗ )∀i ∈ {2, 3, . . . N } (20)

In order to meet the above-said objective, subject to the given constraints, an

appropriate technique is required to find the values of the decision variables ri ∀{i ∈
2, 3, . . . N }, ti ∀{i ∈ 2, 3, . . . N } and determine the fair share of profits gained due to
coordination by sales rebate contracts. It can be observed from the above equations
and inequalities that the objective function is linear but the constraints are nonlinear
in nature. Considering computational complexity of determining rebate values and
threshold values in N-level supply chain, meta-heuristic techniques, namely genetic
algorithm (GA) and simulated annealing (SA), are adopted. The objective of the
problem is to find the rebate and threshold values which, when installed in the supply
chain, will provide a fair share of profit to all the SC members involved.

4 Proposed Heuristic

Researchers have confirmed through experimentation that it is important and good to

start the search technique with a solution constructed by a heuristic than to start the
search technique with a randomly generated solution because such a heuristic seed
Optimal and Heuristic Profit Sharing Using Sales Rebate … 105

solution serves to enhance the process of convergence in the search process (Daniel
& Rajendran, 2006). Therefore, the choice of an appropriate starting solution is
important due to the computational complexity and because convergence to an extent
depends on the initial solution.
Bound-Based Sampling Heuristic
In the bound-based sampling heuristic, we sample a value of threshold from between
the lower and upper bounds of the order quantity; i.e., we chose a random value based
on the bounds. Entity 2 has a special constraint as threshold limit has to be set less
than the expected value of sales. For other entities since the quantity shipped between
them is fixed, the threshold limit can be up to to the maximum order quantity. The
lower limit for threshold can be greater than or equal to zero. A value of ri is sampled
in the interval (riLL , riUL )

tiLL ≥ 0 (21)
tiU L ≤ S(q∗sc ) for all {i ∈ 3, 4 ... N } (22)

The upper and lower limits for thresholds and sample values {t2 , t3 , . . . t N } are
used to evaluate the upper and lower bounds for rebates represented by riU L and riL L
respectively.
The rebate values ri ∀{i ∈ 2, 3, . . . N } thus sampled, coupled with the threshold
limits ti ∀{i ∈ 2, 3, . . . N }, constitute the initial seed solution for heuristic search algo-
rithms. The solution obtained has relatively higher fitness than the solution obtained
by the random procedure and is definitely a feasible solution.

5 Proposed Meta-Heuristics

Three different meta-heuristic algorithms have been proposed to identify the sales
rebate and threshold values for all the supply chain entities to equally share the supply
chain profit:
1. Parallel weighted genetic algorithm
2. Parallel simulated annealing
3. Parallel weighted GA followed by parallel SA.
The mechanics of each of these algorithms are modeled to be able to generate the
sales rebate and threshold values for the N entities in the supply chain. An initial
feasible solution is generated by using the bound-based sampling heuristic proposed
in the previous section.
106 I. Thomson et al.

Fig. 2 Chromosome structure

5.1 Parallel Weighted Genetic Algorithm

GA is a search algorithm based on the mechanics of natural selection and natural

genetics. Daniel and Rajendran (2006) developed different variants of GA for a supply
chain. Each solution is represented in the form of a chromosome. A chromosome
uses value encoding to represent the threshold and rebate values. Figure 2 shows the
representation of a chromosome and an example.
The length of the chromosome depends on the number of entities in the supply
chain; i.e., if N is the supply chain length, then chromosome length is 2(N − 1).
The genes in positions 1 to (N − 1) represent the rebate given to the supply chain
entities. r2 which is in the first position denotes the rebate given by entity 2 to entity
1 and so on. The genes in positions N to 2(N − 1) represent the threshold for the
supply chain entities. t2 which is in the N position denotes the threshold set by entity
2 to entity 1 and so on. In a parallel GA, two sets of initial populations A and B each
containing equal number of parent chromosomes are generated and mutated until the
termination criterion is attained.
For each of A and B:
• All the chromosomes in the initial population are evaluated, and their respective
objective function values (Z ) are obtained by evaluating Eq. (13). Fitness value
( f ) of each chromosome is computed using the formula f = 1+Z 1
(since Z is to
be minimized).
• A child chromosome (offspring) is created by taking the weighted mean (with
respect to relative fitness values of chromosomes) of the values of respective genes
of parent chromosomes. In addition to this, all the parent chromosomes in the initial
population as well as the new child chromosome are mutated with a perturbation
proportion PP. The best chromosomes according to their fitness function from
the set of (initial parent population, mutated parent population and mutated child
chromosome) are chosen as the next parent population.
The best chromosome from the next generation of both A and B replaces the worst
chromosome of the other one, and next parent population is formed. This ensures
communication between the two parallel populations A and B. The above-mentioned
steps are repeated. After a certain number of generations, the algorithm is terminated
and the best chromosome at the end of termination provides the rebates and the
threshold values for the respective members in the supply chain (Fig. 3).
Optimal and Heuristic Profit Sharing Using Sales Rebate … 107

Fig. 3 Parallel weighted GA illustration

5.1.1 Notations

no_gen: Number of generations

len: Length of the chromosome (number of genes)
pop_size: Population size Sik :] ith gene of the kth chromosome for
i ∈ {1, 2, 3, . . . len} and k ∈ {1, 2, 3, . . . pop_size}
ini_pop: Initial population
ri : Rebate given by entity i to entity i − 1, i ∈ {2, 3, . . . N }
ti : Threshold limit set by entity i to entity i − 1 i ∈ {2, 3, . . . N }
riUL : Upper limit of rebate given by entity i to entity i − 1, i ∈ {2, 3, . . . N }
riLL : Lower limit of rebate given by entity i to entity i − 1, i ∈ {2, 3, . . . N }
tiUL : Upper limit of rebate set by entity i to entity i − 1, i ∈ {2, 3, . . . N }
tiLL : Lower limit of rebate set by entity i to entity i − 1, i ∈ {2, 3, . . . N }
f: Fitness value of the chromosome
r el f : Relative fitness value of the chromosome
u: Uniform random number in the interval (0, 1)
P P: Perturbation proportion
c_chrom: Offspring obtained from the weighted mean of parent population
m_pop: Resultant population consisting of offspring after mutation.

5.1.2 Step-by-Step Procedure of the Parallel Arithmetic GA

(Considering a 3-Member Supply Chain)

• Step 1 : Initialize no_gen = 0.

• Step 2 : Generate two sets A and B of initial population (ini_pop) each of size
(pop_size) equal to 5 (set arbitrarily for the purpose of demonstrating the applica-
bility of GA to the current problem) with each chromosome representing a set of
rebates and thresholds of distributor and manufacturer. The first two genes carry
 108 I. Thomson et al.

the rebate values, and the last two carry the threshold values.
For set A:
if i ∈ {1, 2}
Sik = riLL + (riUL − riLL ) ∗ u
if i ∈ {3, 4}
Sik = tiLL + (tiUL − tiLL ) ∗ u
For set B, the same is repeated, but instead of the random number u its antithetic
1 − u is used. Note that ini_pop w.r.t A and B refers to populations A and B at the
beginning of every generation.
• Step 3: For each of the sets A and B, do the following:
– 3.1 Evaluate every chromosome in ini_pop (by evaluating the objective func-
tion Eq. 13), and obtain the value of Z . Obtain the fitness value f k for every
chromosome k.
– 3.2 Relative fitness of each of the parent chromosomes is calculated as f f where
 f is the sum of the fitness functions of all the parent chromosomes.
– 3.3 A child chromosome is formed by taking the weighted mean of the ini_pop.
c_chromi = 5k=1 (rel f k ∗ Sik ) for i ∈ {1, 2, 3, . . . len}
– 3.4 All the chromosomes in ini_pop and c_chrom are mutated with a perturba-
tion proportion P P = 0.2
m_popik = min(max((Sik ∗ (1 − P P) + Sik ∗ 2 ∗ P P ∗ u), riLL [or tiLL ]), riUL
[or tiUL ]) for i ∈ {1, 2, 3, . . . len}
c_chromi = min(max((c_chromi ∗ (1 − P P) + c_chromi ∗ 2 ∗ P P ∗ u), riLL
[or tiLL ]), riUL [or tiUL ]) for i ∈ {1, 2, 3, . . . len}
– 3.5 The best 5 chromosomes from ini_pop, m_pop and c_chrom are chosen
and set as the next ini_pop.
• Step 4: The best chromosome from the ini_pop of both sets A and B combined
is chosen, and it replaces the worst chromosome of the ini_pop of the other pop-
ulation. This communication establishes parallelism between the two sets A and
B.
• Step 5: Set no_gen = no_gen + 1.
• Step 6: Repeat steps 2, 3, 4 and 5 until no_gen = 500 which is the termination
criterion.
• Step 7 : The best chromosome among the resultant 10 chromosomes of A and B
constitutes the solution to the problem.

5.2 Parallel Simulated Annealing Algorithm

In the proposed parallel SA, the idea of parallelism is drawn from Janakiram et
al. (1996) and the implementation is drawn from Daniel and Rajendran (2005).
Janakiram et al. (1996) suggested a clustering algorithm for SA in which there are n
nodes of the network that run using different initial solutions. After a fixed number
of iterations, they exchange their partial results to get the best one. All the nodes
Optimal and Heuristic Profit Sharing Using Sales Rebate … 109

accept the best partial solution and start applying the SA technique for that best
partial result. This process is repeated until the termination criterion is attained.
Approaches of these kinds are often highly problem-dependent, and one variant of
this algorithm that suites the problem discussed here is the parallel SA algorithm.
In parallel SA, there are two parallel nodes A and B corresponding to two different
solutions.
For each of A and B, the following are done:
• The initial node is perturbed, and the objective function (Z ) is obtained by evalu-
ating Eq. (13).
• If (Z ) of the perturbed solution is better than the (Z ) of the initial solution, then
the initial solution is updated, else the initial solution is updated with a predefined
probability.
• If the solution is not improving even after two consecutive iterations, the following
are done in a cyclic manner:

– Reheating is done. Temperature T is increased to 500.

– The so far best solution of the parallel node replaces the current best solution of
the node under consideration.
– Both reheating and adoption of solution from the parallel node take place.

The above steps are repeated until the termination criterion is reached which is
the minimum temperature possible (Fig. 4).

5.2.1 Notations

N: Number of supply chain members

Z: Objective function value (Eq. 13)
S: Initial solution sequence
S: Perturbed solution sequence
rf: Temperature reduction factor (0 < r f < 1)

Fig. 4 Parallel SA illustration

 110 I. Thomson et al.

T: Temperature (T > 0)
len: Length of the sequence:]- 2*(N-1)
ri : Rebate given by entity i to entity i − 1, i ∈ {2, 3, . . . N }
ti : Threshold limit set by entity i to entity i − 1, i ∈ {2, 3, . . . N }
riUL : Upper limit of rebate given by entity i to entity i − 1, i ∈ {2, 3, . . . N }
riLL : Lower limit of rebate given by entity i to entity i − 1, i ∈ {2, 3, . . . N }
tiUL : Upper limit of threshold set by entity i to entity i − 1, i ∈ {2, 3, . . . N }
tiLL : Lower limit of threshold set by entity i to entity i − 1, i ∈ {2, 3, . . . N }
it_count: Iteration count
: absolute percentage change in the objective value function
u: Uniform random number in the interval (0, 1).

5.2.2 Step-by-Step Procedure of the Parallel Simulated Annealing

Algorithm (Considering a 3-Member Supply Chain)

• Step 1: Initialize it_count = 0.

• Step 2: Generate two nodes (candidate solution) A and B with initial solutions
(S), each of length (len = 4). The first two values are the rebate values, and the
last two are the threshold values.
For set A:
if i ∈ {1, 2}
Si = riLL + (riUL − riLL ) ∗ u
if i ∈ {3, 4}
Si = tiLL + (tiUL − tiLL ) ∗ u
For set B, the same is repeated, but instead of the random number u its antithetic
1 − u is used.
• Step 3: Set T = 500 and i = 0 and r f = 0.2.
• Step 4: For 500 ≥ T ≥ 17, do the following:
• 4.1: For each of the nodes A and B and for 0 ≤ j ≤ 10, do:
– 4.1.1: Perturb S as
Si = min(max(Si ∗ (1 − r f ) + 2 ∗ r f ∗ u, riLL ), riUL ) ∀ i ∈ {1, 2}
Si = min(max(Si ∗ (1 − r f ) + 2 ∗ r f ∗ u, tiLL ), tiUL ) ∀ i ∈ {3, 4}

)−z(S)
– 4.1.2: Compute  = z(S z(S) ∗ 100
if  < 0: Update S = S  .
−
else if e T ≥ u: Update S = S  .
– 4.1.3: If the value of the objective function is not improving even after two iterations
and
it_count ≤ 10, 000, do the following in a cyclic manner:
· Reheating: Set T = 500 and it_count = it_count + 1
· Update the best solution S (obtained so far) with respect to the current
node with the best solution S (obtained so far) of the parallel node. Set
it_count = it_count + 1
 Optimal and Heuristic Profit Sharing Using Sales Rebate … 111

· Do both the above steps: reheating and adoption of the best solution
(obtained so far) from the parallel node.
– 4.1.4: Set j = j + 1.
– 4.2: Set T = T ∗ r .
• Step 5: The node with a smaller objective function Z , between A and B, con-
stitutes the solution to the problem.

5.3 Parallel Genetic Algorithm Followed by Parallel

Simulated Annealing Algorithm

Experiments show that a good initial solution for simulated annealing improves
the quality of the solution. Genetic algorithm alone often fails to provide optimal
solutions as its quality is heavily dependent on the quality of the initial population.
Considering these two concepts, an algorithm is proposed that feeds the best chromo-
somes from the final child population of a parallel GA as the initial solutions of the
two nodes for a parallel SA. By doing so, the quality of the solution is tremendously
increased.

5.3.1 Step-by-Step Procedure of the Parallel GA-SA Algorithm

• Step 1: Do steps 1 to 6 of the parallel GA listed in 1.

• Step 2: Set the best chromosome from population A as the initial solution for node
A and the best chromosome from population B as the initial solution for node B
of the parallel SA.
• Step 3: Do steps 1 to 5 of the parallel SA listed in 2.

6 Experiments and Discussion

To compare the efficiency of various algorithms proposed above, we take a three-stage

supply chain and determine the objective function (equation 13) and the computa-
tional time for each algorithm. For a more real-time perspective of the model, we
refer to entity α_1 as retailer, entity α_2 as distributor and entity α_3 as manufac-
turer and discuss the expected profits and contract parameter decisions (i.e., rebates
and thresholds). We perform the two sets of experiments on the model described in
Sect. 5. Experimental settings have been identified by varying the cost settings for
the exchange of goods between the entities and also varying the demand patterns of
the end customer.
112 I. Thomson et al.

Table 1 Supply chain experiment 1—fixed cost parameter subjected to varied demand

In the first experiment, we have a fixed cost setting between the three entities,
namely retailer, distributor and manufacturer. The costs in experiments E1 are fixed;
assuming the purchase cost P for manufacturer is 40, she/he sells it to the distributor
at 52 (adds 30% value over the purchase cost). The distributor purchases it from the
manufacturer at 52 and sells it at 64 (adds 60% value over the purchase cost). Finally,
the retailer purchases it from the distributor at 64 and sells it at 76 (adds 90% value
over the purchase cost). They exchange the goods at a predetermined price but are
subjected to varied demand. The demand patterns used in this particular test are 1–
100, 101–200, 201–300 and 301–400. These model settings are described in Table 1.
The terms α3 , α2 and α1 signify the quantum of value addition that has been done by
manufacturer, distributor and retailer, respectively. Table 1 presents the experimental
setting E1_1, E1_2 and E1_3 in detail.
Tables 1.1, 1.2 and 1.3 give the results obtained for the setup in experiment 1 using
the three different meta-heuristics. All the meta-heuristic algorithms were run for a
computational time of around 1 s. The results show that the best value of the objective
function is attained using parallel GA-SA. Parallel weighted GA is the least accurate
due to a restricted computational time, while parallel SA performs considerably well
and produces solutions comparable to that of parallel GA-SA.
Based on the experimental results, the profits of the supply chain entities when a
GA-based heuristic was used are presented in Figs. 5, 6 and 7. First observation that
Optimal and Heuristic Profit Sharing Using Sales Rebate … 113

Fig. 5 Results of supply chain experiment 1

we are able to make is that there is a significant increase in the total profit in Case B
and Case C compared to Case A. This is primarily due to the increase in supply chain
order quantity to centralized order quantity qc from the decentralized order quantity
qd . We will notice that as the demand increases/stabilizes, the profits in Case B and
Case C tend to match that in Case A. We notice that in Case B of E1_2, the retailer
goes into a loss of 312 units. In Case B of E1_2, profits are not evenly shared and
the retailer faces most of the risk. To enable the retailer to participate in ordering
the higher-order quantity, the sales rebate contract plays a pivotal role. We notice
in Case C that the profits have been shared between various supply chain entities
based on the value addition to the product, thereby ensuring that no member faces
an excessive loss while trying to increase the overall supply chain profit.
In experiment 2, we have a varied cost setting with a fixed demand. In experiments
E2_1, assuming the purchase cost P for manufacturer is 50, she/he sells it to the
distributor at 75 (adds 50% value over the purchase cost). The distributor purchases
it from the manufacturer at 75 and sells it at 90 (adds 80% value over the purchase
cost). Finally, the retailer purchases it from the distributor at 90 and sells it at 95
(adds 90% value over the purchase cost). Experiments E2_2 and E2_3 also drive
scenarios where the value addition can be read in the above manner. Table 2 presents
the experimental setting E2_1, E2_2 and E2_3 in detail.
Tables 2.1, 2.2 and 2.3 give the results obtained for the setup in experiment 2 using
the three different meta-heuristics. These results show that in terms of giving the
optimum rebate and threshold values, both parallel SA and parallel GA-SA perform
well but as the number of members in the supply chain increases, parallel SA-GA is
expected to give a better solution as compared to parallel SA algorithm. Parallel GA
requires a greater computational time to produce efficient results; hence in this time
setting, it is not performing well.
114 I. Thomson et al.

Fig. 6 Results of supply chain experiment 1

Fig. 7 Results of supply chain experiment 1

Presented in Figs. 8, 9 and 10 are the profits obtained by various entities when a
GA-based heuristic was used to generate the rebate and threshold values of the sales
contract for experiment 2. In this experiment, we notice that the retailer is faced with
a loss in Case B where we have a centralized decision making but no coordination.
In Case C, we notice that coordination enables profit sharing between all the supply
chain entities, thereby encouraging the retailer to take the risk by ordering more
quantity.
An interesting observation in this experiment is with regard to the profits shared
in Case C. The total profits remain same in Case B and Case C, but in Case C as
we have installed sales rebate contract to ensure coordination, the profits are shared
Optimal and Heuristic Profit Sharing Using Sales Rebate … 115

Fig. 8 Results of supply chain experiment 2

Fig. 9 Results of supply chain experiment 2

Fig. 10 Results of supply chain experiment 2

116 I. Thomson et al.

Table 2 Supply chain experiment 2—varied cost parameters subjected to constant demand

based on the value added by the entity with respect to the purchase price. Considering
the example of setting E2_1, we notice that the SC is able to generate an additional
profit of 3000 units in Case B and Case C as compared to Case A, which is split
between the three entities based on their value addition in Case C, while in Case B
the retailer is at a loss.

7 Conclusion

This paper presented a profit sharing mechanism using sales rebate contracts in a
multistage coordinated supply chain. The generic N-echelon model was mathemat-
ically described, and the reasons why it cannot be solved have been listed. Several
meta-heuristics have been developed and have been tested on a 3-echelon model. It
was observed that the total SC profit has increased in Case B when compared to Case
A because of centralization, but the increase in profits is not proportionally shared
by every member according to their respective value addition. A single member
(retailer) is subjected to more risk. The introduction of contracts has compensated
for the retailer’s risk, and the increased profit due to supply chain coordination is
Optimal and Heuristic Profit Sharing Using Sales Rebate … 117

fairly shared among the members, as proportional to their value addition to the prod-
uct. Here, the deviation between the proportion of profit shared and the proportion of
their value addition is minimized using genetic algorithm and simulated annealing.
Analyzing the quality of the optimum solution obtained, it can be concluded that
both parallel GA-SA and parallel SA meta-heuristic techniques can be used and the
better of the two can be chosen for determining the sales rebate and threshold values.
Future work can be done on the application of particle swarm optimization and ant
colony optimization algorithms in determining the rebate and threshold values.

Acknowledgements This work becomes a reality with the kind support and help of many individ-
uals. We would like to extend our sincere thanks to all of them. We are overwhelmed by the quick
review and response we received related to the publication of this work. Our heartfelt gratitude to
the editor and the reviewers for their diligent work.

References

Arcelus, F. J., Kumar, S., & Srinivasan, G. (2007). Pricing and rebate policies for the newsvendor
problem in the presence of a stochastic redemption rate. International Journal of Production
Economics, 107(2), 467–482.
Arcelus, F. J., Kumar, S., & Srinivasan, G. (2008). Pricing and rebate policies in the two-echelon sup-
ply chain with asymmetric information under price-dependent, stochastic demand. International
Journal of Production Economics, 113(2), 598–618.
Cachon, G. P. (2003). Supply chain coordination with contracts. Handbooks in Operations Research
and Management Science, 11, 227–339.
Chang, Y.-H. (2010). Adopting co-evolution and constraint-satisfaction concept on genetic algo-
rithms to solve supply chain network design problems. Expert Systems with Applications, 37(10),
6919–6930.
Daniel, J. S. R., & Rajendran, C. (2005). Determination of base-stock levels in a serial supply chain:
a simulation-based simulated annealing heuristic. International Journal of Logistics Systems and
Management, 1(2/3),
Daniel, J., & Rajendran, C. (2006). Heuristic approaches to determine base-stock levels in a serial
supply chain with a single objective and with multiple objectives. European Journal of Opera-
tional Research, 175(1), 566–592.
Emmons, H., & Gilbert, S. M. (1998). Note the role of returns policies in pricing and inventory
decisions for catalogue goods. Management Science, 44(2), 276–283.
Janakiram, D., Sreenivas, T.H., & Subramaniam, K. G. (1996). Parallel simulated annealing algo-
rithms. Journal of Parallel and Distributed Computing, 37(121), 207–212 (1996)
Koulamas, C. (2006). A newsvendor problem with revenue sharing and channel coordination.
Decision Sciences, 37(1), 91–100.
Krishnan, H., Kapuscinski, R., & Butz, D.A. (2004). Coordinating contracts for decentralized supply
chains with retailer promotional effort. Management Science, 50(1), 48–63 (2004)
Lian, Z., & Deshmukh, A. (2009). Analysis of supply contracts with quantity flexibility. European
Journal of Operational Research, 196(2), 526–533.
Nachiappan, S. P., & Jawahar, N. (2007). A genetic algorithm for optimal operating parameters of
vmi system in a two-echelon supply chain. European Journal of Operational Research, 182(3),
1433–1452.
Nalla, V.R. (2008). Contract mechanisms for coordinating operational and marketing decisions
in a supply chain: models and analysis. Ph. D. Thesis, Nyenrode Business Universiteit, The
Netherlands.
118 I. Thomson et al.

Sang, S. (2016). Buyback contract with fuzzy demand and risk preference in a three level supply
chain, 46, 518–526.
Sethupathi, P. V. R., Rajendran, C., & Ziegler, H. (2014). A comparative study of periodic-review
order-up-to (T, S) policy and continuous-review (s, S) policy in a serial supply chain over a finite
planning horizon. Supply Chain Strategies: Issues, and Models. Springer.
Simchi-Levi, D., Kaminsky, P., & Simchi-Levi, E. (2008). Designing and managing the supply
chain. Managing the Supply Chain.
Taylor, T. A. (2002). Supply chain coordination under channel rebates with sales effort effects.
Management Science, 48(8), 992–1007.
Tsay, A. A. (1999). The quantity flexibility contract and supplier-customer incentives. Management
Science, 45(10), 1339–1358.
van der Rhee, B., van der Veen, J. A. A., Venugopal, V., & Nalla, V. R. (2010). A new revenue
sharing mechanism for coordinating multi-echelon supply chains. Operations Research Letters,
38(4), 296–301.
Wong, W.-K., Qi, J., & Leung, S. Y. S. (2009). Coordinating supply chains with sales rebate contracts
and vendor-managed inventory. International Journal of Production Economics, 120(1), 151–
161.
A Deterministic Heuristic Algorithm
to Minimize the Length
of a Manufacturing Line
in Transformation of Jobshops
into Flowshops

J. Krishnaraj

Abstract This research paper focuses on the development of a deterministic

heuristic algorithm for solving the problem of transforming jobshops into flowshops
with the aim of minimizing a manufacturing line length. The considered problem is
a variant of the familiar stringology problem of the shortest common supersequence
(SCS) which finds the shortest string length. The proposed deterministic heuristic
algorithm generates seed solutions from the concatenation pattern of forward reduc-
tion, inverse reduction, and machine elimination techniques, in addition to a new
local search scheme called two-pair adjacent interchange scheme (TPAIS). The best
seed sequence is improved by three concatenation patterns of the Job Index-Based
Insertion Scheme (JIS) and Modified Job Index-Based Swap Scheme (MJSS), thereby
leading to three variants of the suggested heuristic algorithm (called HAV1 to HAV3).
An extensive computational analysis is carried out to illustrate the performance of the
three variants of the proposed deterministic heuristic algorithm. The computational
results show that the HAV3 performs well in terms of minimizing a manufacturing
line length and number of duplicate machines saved in comparison with the existing
algorithm.

1 Introduction

In jobshop scheduling problems, the process sequence (machine routings) of the

jobs is not the same, and machines that can able to perform similar operations are
grouped (French, 1982). Hence, the flow of each job in jobshop scheduling is not
unidirectional. But in case of a flowshop, jobs follow a unidirectional flow as per
the respective job sequence through the shop. As per Knolmayer et al. (2002), the
conversion of jobshops into flowshops is significantly important in the situation
of reducing the manufacturing flow line (i.e., to achieve an efficient supply chain
management). It is noteworthy that the minimization of manufacturing flow line
length (i.e., flowshop length) leads to a reduced inventory level (Kimms, 2000).

J. Krishnaraj (B)
Department of Mechanical Engineering, MLR Institute of Technology, Hyderabad, Telangana,
India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 119
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_6
120 J. Krishnaraj

The minimization of a manufacturing flow line length is related to the familiar

classical SCS stringology problem (Framinan & Ruiz-Usano, 2002). Framinan
(2005) stated that although the SCS problem and jobshop transformation problem
appear similar, they are not in terms of assumptions made in the problem statement.
Timkovsky (1989) stated that in overall SCS problem is identified to be NP-hard.
From the literature studies, it is observed that some researchers (to cite a few, Fram-
inan and Ruiz-Usano (2002), Framinan (2005), Rajendran et al. (2010)) attempted
the problem of transmuting jobshops into flowshops with the aim of minimizing the
manufacturing flow line length. Mousavi et al. (2012), Dondi (2013), and Saifullah
and Islam (2016) attempted the SCS stringology problem; however, they are not
directly related to the transformation of jobshops into flowshops.
The SCS stringology problem has many applications in the fields of DNA analysis,
planning, artificial intelligence, query optimization, data compression in the database,
and various bioinformatics activities (Saifullah and Islam 2016). For a given set
of strings L, the SCS problem leads to find a minimal string length (X) that is a
supersequence of each string in L (strings in L are embedded). A string is a set of
zero or more characters from  (an alphabet). For example, consider a set of strings
L = {cba, abba, abc} over an alphabet  = {a, b, c}. The shortest supersequence
for L is abcba (i.e., X = 5). Exact algorithms like branch-and-bound (Fraser &
Irving, 1995) and dynamic programming (DP) (Jiang & Li, 1995) are proposed for
solving the SCS problem. These algorithms execute optimally in polynomial time
for small instances. But for large instances, DP uses lots of memory and branch-and-
bound consumes exponential time (Barone et al., 2001). Because of NP-hardness, the
exact algorithms failed to compute SCS for larger problem instances. Many heuristic
algorithms and meta-heuristic approaches were attempted to solve this problem.
The heuristic approaches include majority merge (Branke et al., 1998), reduce, and
expand (Barone et al., 2001) and deposition and reduction (DR) (Ning & Leong,
2006). The meta-heuristic approaches include an ant colony optimization algorithm
(Michel & Middendorf, 1999), an artificial bee colony algorithm (Noaman & Jaradat,
2011), enhanced beam search algorithm (Mousavi et al., 2012), and chemical reaction
optimization algorithm (Saifullah & Islam, 2016).
This work focuses on the development of a deterministic heuristic algorithm to
transform jobshops into flowshops with the aim of minimizing the manufacturing line
length. Note that deterministic heuristics are computationally faster than metaheuris-
tics always executes the same way and produces the same solution. The minimization
of manufacturing line is more relevant and meaningful for today’s dynamic produc-
tion environment which reduces the number of duplicate machines in the manufac-
turing line (i.e., a sequence of machine routings corresponds to the job sequence) and
helps to minimize the lead times during the production of jobs which also reduces
the inventory levels. This problem was attempted by very few researchers, Fram-
inan (2005) and Rajendran et al. (2010), and reported the best string length for
70 well-known benchmark problem instances. Rajendran et al. (2010) employed
a meta-heuristic algorithm which generally consumes higher computational time.
Hence in this work, a deterministic heuristic algorithm is proposed for the minimiza-
tion of length of the manufacturing line for the problem of transforming jobshops
A Deterministic Heuristic Algorithm to Minimize the Length … 121

into flowshop, and results are compared with the results yielded by Framinan (2005)
through deterministic heuristic. To illustrate the effectiveness and performance of
the proposed deterministic heuristic, the well-known benchmark problems (consid-
ered in Framinan, 2005) of 40 instances (LA01 ~ LA40) of eight different sizes
owing to Lawrence (1984), 10 problems (ORB1 ~ ORB10) due to Applegate and
Cook (1991), and 20 problems (SWV1 ~ SWV20) of Storer et al. (1992) are consid-
ered for the analysis of the transformation of jobshops into flowshops. An extensive
performance analysis on these 70 jobshop benchmark problems is carried out to
evaluate the proposed deterministic heuristic algorithm with the aim of minimizing
the manufacturing line length on the jobshop transformation problem.

2 Problem Description

The transformation of jobshops into flowshops with the aim of minimizing the length
of a manufacturing line may be described as follows:
Consider a jobshop of  m machines
 where n jobs have to be processed according to
a routing matrix M := Mi j , where Mi j indicates the jth operation to be performed
on job i. In this problem, the objective is to find a flowshop with the minimum number
of machines where all job operations can be completed in the order indicated in M.
The obtained solution, say shortest common supersequence, should be a feasible one
with a minimum length of manufacturing line.
Let us consider a job ordering J = { j1 − j2 − . . . − jn } of a jobshop is
transformed into a following flowshop:
 
S = M j1,1 − M j2,1 − . . . − M jn,1 − M j1,2 − M j2,2 − . . . − M j1,m − . . . − M jn,m

The above flowshop is a feasible one because the machines of all jobs are
sequenced as per the job ordering as well as satisfies the machine routing of jobs as
specified in the routing matrix M. In this case, the length of manufacturing line is
found as follows: L = n × m.
Let us consider a jobshop problem with 6 jobs and 4 machines (i.e., n = 6 and m
= 4) and its machine routings as given in Table 2 (refer Appendix 1). Assume the
job ordering as J= {1-2-3-4-5-6} gives S = {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-
1-3-3-4-1-2} and its length of the manufacturing line, L = 24. The flowshop, S, is
a feasible solution, but it is not a shortest common supersequence. For instance, let
us consider S = {1-2-3-4-1-3-2-4-3}. It is a feasible solution which follows all job
operations performed as per the order specified in machine routings. In this case, the
length of the manufacturing line, L = 9 and S, is one of the supersequence which
saved 15 duplicated machines of flowshop.
122 J. Krishnaraj

3 Proposed Deterministic Heuristic Algorithm

In this research work, three variants of the deterministic heuristic algorithm, namely
heuristic algorithm variant-1 (HAV1), heuristic algorithm variant-2 (HAV2), and
heuristic algorithm variant-3 (HAV3), are proposed. The notations and terminologies
used in the heuristic algorithm are first discussed in Sect. 3.1, and then the step-by-step
procedure of the heuristic algorithm and its variants are presented in Sect. 3.2.

3.1 Notations and Terminology

n number of jobs
m number of machines
S1 , S2 , S3 , S4 a sample of shortest common super seed sequences of machines
L 1, L 2, L 3, L 4 length of manufacturing line corresponds to S1 , S2 , S3 , S4
BS best shortest common supersequence found in a HAV
BL shortest string length (length of manufacturing line) corresponds
to BS found in a HAV.

3.2 The Step-by-Step Procedure of the Proposed

Deterministic Heuristic Algorithm and Its Variants
(HAV1 to HAV3)

Step 1: Generation of shortest common super seed sequence of machines 1.

Step 1.1: Initialize a super sequence, , based on the machine routing corre-
sponding to a job sequence {1-2-3-4-5-6-…-n} and assign the length of manufac-
turing line (string length) to .
Step 1.2: Apply Forward Reduction Technique (proposed by Framinan (2005);
See Appendix 1 for details) on , assign the obtained super sequence of machines
to and its string length to .

then
assign the super sequence of machines to and to
else
do not update and .
Step 1.3: Apply Inverse Reduction Technique (proposed by Framinan (2005); See
Appendix 2 for details) on , assign the obtained super sequence of machines to
and its string length to .

then
A Deterministic Heuristic Algorithm to Minimize the Length … 123

assign the super sequence of machines to and to

else
do not update and .
Step 1.4: Apply Machine Elimination Technique (proposed by Framinan and
Ruiz-Usano (2002); See Appendix 3 for details) on , assign the obtained super
sequence of machines to and its string length to .

then
assign the super sequence of machines to and to
else
do not update and .
Step 1.5: Invoke Two-Pair Adjacent Interchange Scheme (TPAIS) (proposed in
this work; See Appendix 4 for details) on , assign the obtained super sequence
of machines to and its string length to .

then
assign the super sequence of machines to and to .
else
do not update and .
Step 2: Generation of shortest common super seed sequence of machines 2.
Steps in the generation of and its string length differ from those in Step 1 in
terms of applying first the inverse reduction technique then forward reduction
technique and thereafter the machine elimination technique and TPAIS.
Step 3: Generation of shortest common super seed sequence of machines 3.
Step 3.1: Initialize a super sequence of machines, based on the machine routing
corresponds to a job sequence {n-…-6-5-4-3-2-1} (antithetic job sequence) and
assign the length of the manufacturing line (string length) to .
Step 3.2 to Step 3.5: Correspond to the application of forward reduction, inverse
reduction, machine elimination techniques, and TPAIS (see Steps 1.2 to 1.5).
Step 4: Generation of shortest common super seed sequence of machines 4.
Steps in the generation of and its string length differ from those in Step 3 in
terms of applying first the inverse reduction technique then forward reduction
technique and thereafter the machine elimination technique and TPAIS.
Step 5: Choose the best seed super sequence of machines among to based on
the lowest string length among to . Assign the best seed super sequence of
machines to BS and its string length to BL.
Step 6: Apply concatenation of JIS (used by Rajendran et al. (2010); See appendix
5 for details) and MJSS (modified job-index based swap scheme, originally used
by Rajendran and Ziegler (2004); See appendix 6 for details) such as JIS-MJSS-
JIS-MJSS-JIS-MJSS in case of HAV1, concatenation of JIS-JIS-MJSS-JIS-MJSS-
JIS in case of HAV2 and concatenation of JIS-JIS-JIS-MJSS-JIS-MJSS in case of
HAV3 on an initial job sequence {1-2-3-4-5-6-…-n}; Return the best shortest
common super sequence of machines (BS) and its string length (BL). STOP.

Note that on every perturbed job sequence obtained during the application of JIS
and MJSS, the supersequence of machines based on its machine routings is initialized
124 J. Krishnaraj

and the set of procedures followed in Step 1 to Step 5 is executed to update the best
job sequence, best shortest common supersequence of machines (BS), and shortest
string length (BL).
In the exploration of minimization of the length of manufacturing line, the
proposed variants (HAV1 to HAV4) employ the new local search scheme TPAIS
for the interchanging of adjacent machines to overcome the drawback present in the
earlier approaches like forward reduction, inverse reduction, and machine elimina-
tion techniques. Hence, the order of forward reduction–inverse reduction–machine
elimination–TPAIS is used in case of HAV1 and HAV3 and the order of inverse
reduction–forward reduction–machine elimination–TPAIS is used in case of HAV2
and HAV4.

4 Performance Evaluation of the Proposed Deterministic

Heuristic Algorithm

The performance of the proposed three variants of the deterministic heuristic algo-
rithm, namely HAV1, HAV2, and HAV3, is evaluated on the seventy well-known
jobshop benchmark instances (see Sect. 1 for details) considered by Framinan (2005)
to solve the jobshop problems by transforming them into flowshops with the aim
of minimizing the manufacturing line length. The results (best string length) of the
proposed heuristic algorithm variants are compared with the results reported by Fram-
inan (2005). It is to be noted that Framinan (2005) had consolidated and reported
the best string length as an outcome of the application of algorithms such as H2
and H3 employed by Branke et al. (1998), BS by Framinan and Ruiz-Usano (2002),
and TS by Framinan (2005). Note that the proposed deterministic heuristic algorithm
variants (HAV1 to HAV3) generate only 12n2 job sequences (including antithetic job
sequences) per variant during its search process to find the best string length which
corresponds to the machine routing or manufacturing flow line. The results of the
proposed heuristic algorithm variants HAV1 to HAV3 are presented in Table 1.
It is found that the proposed heuristic algorithm variants, HAV1, HAV2, and
HAV3, together yield better results than those reported by Framinan (2005) on 46
instances out of 70 benchmark jobshop problem instances considered. Note that
on remaining instances, the proposed HAV variants yield the same best manufac-
turing line length as reported by Framinan (2005). By considering all the results
(best manufacturing line length) reported by Framinan (2005), the proposed variants
HAV1, HAV2, and HAV3 produce best string length on 30, 31, and 33 instances,
respectively, out of 70 benchmark jobshop problem instances considered. Also, note
that the proposed HAV1, HAV2, and HAV3, respectively, saved a total of 89, 95, and
97 duplicate machines in the manufacturing flow line for the entire set of problem
instances. Based on the number of duplicate machines saved in the manufacturing
line (for all 70 benchmark problem instances considered) and the number of best
string length, the proposed HAV3 shows the better performance (refer to Table 1).
A Deterministic Heuristic Algorithm to Minimize the Length … 125

Table 1 Computational results

Jobshop problem n m String length String length String length String length
instance reported by obtained by obtained by obtained by
Framinan proposed proposed proposed
(2005) HAV1 HAV2 HAV3
La01 10 5 13 13(0) 13(0) 13(0)
La02 10 5 12 12(0) 12(0) 12(0)
La03 10 5 12 12(0) 12(0) 12(0)
La04 10 5 13 13(0) 13(0) 13(0)
La05 10 5 12 12(0) 12(0) 12(0)
La06 15 5 14 14(0) 14(0) 14(0)
La07 15 5 14 14(0) 14(0) 14(0)
La08 15 5 13 13(0) 13(0) 13(0)
La09 15 5 14 14(0) 14(0) 14(0)
La10 15 5 14 14(0) 14(0) 14(0)
La11 20 5 14 14(0) 14(0) 14(0)
La12 20 5 14 14(0) 14(0) 14(0)
La13 20 5 15 15(0) 15(0) 15(0)
La14 20 5 15 15(0) 15(0) 15(0)
La15 20 5 14 14(0) 14(0) 14(0)
La16 10 10 35 33(2) 33(2) 33(2)
La17 10 10 35 34(1) 34(1) 34(1)
La18 10 10 38 36(2) 36(2) 36(2)
La19 10 10 35 32(3) 32(3) 32(3)
La20 10 10 35 35(0) 35(0) 35(0)
La21 15 10 42 39(3) 39(3) 39(3)
La22 15 10 41 39(2) 39(2) 39(2)
La23 15 10 42 39(3) 39(3) 39(3)
La24 15 10 43 41(2) 41(2) 41(2)
La25 15 10 42 39(3) 39(3) 38(4)
La26 20 10 43 42(1) 43(0) 43(0)
La27 20 10 44 43(1) 42(2) 42(2)
La28 20 10 46 44(2) 44(2) 44(2)
La29 20 10 45 44(1) 44(1) 44(1)
La30 20 10 45 43(2) 44(1) 44(1)
La31 30 10 49 47(2) 46(3) 44(5)
La32 30 10 49 48(1) 46(3) 47(2)
La33 30 10 49 48(1) 45(4) 48(1)
La34 30 10 49 47(2) 48(1) 49(1)
La35 30 10 49 49(0) 48(1) 48(1)
La36 15 15 79 78(1) 76(3) 76(3)
(continued)
126 J. Krishnaraj

Table 1 (continued)
Jobshop problem n m String length String length String length String length
instance reported by obtained by obtained by obtained by
Framinan proposed proposed proposed
(2005) HAV1 HAV2 HAV3
La37 15 15 79 71(8) 73(6) 74(5)
La38 15 15 80 76(4) 73(7) 73(7)
La39 15 15 80 74(6) 75(5) 71(9)
La40 15 15 78 75(3) 74(4) 76(2)
orb01 10 10 28 27(1) 27(1) 27(1)
orb02 10 10 34 32(2) 32(2) 33(1)
orb03 10 10 20 20(0) 20(0) 20(0)
orb04 10 10 34 33(1) 33(1) 33(1)
orb05 10 10 26 26(0) 26(0) 26(0)
orb06 10 10 28 27(1) 27(1) 27(1)
orb07 10 10 34 32(2) 32(2) 33(1)
orb08 10 10 20 20(0) 20(0) 20(0)
orb09 10 10 34 33(1) 33(1) 33(1)
orb10 10 10 26 26(0) 26(0) 26(0)
swv01 20 10 29 28(1) 29(0) 29(0)
swv02 20 10 28 27(1) 27(1) 27(1)
swv03 20 10 28 27(1) 28(0) 27(1)
swv04 20 10 29 29(0) 29(0) 29(0)
swv05 20 10 30 29(1) 29(1) 29(1)
swv06 20 15 61 59(2) 58(3) 58(3)
swv07 20 15 59 57(2) 57(2) 56(3)
swv08 20 15 60 56(4) 55(5) 55(5)
swv09 20 15 57 58(−1) 56(1) 54(3)
swv10 20 15 57 57(0) 58(−1) 58(−1)
swv11 50 10 32 32(0) 32(0) 32(0)
swv12 50 10 33 32(1) 32(1) 32(1)
swv13 50 10 32 32(0) 32(0) 32(0)
swv14 50 10 33 32(1) 32(1) 32(1)
swv15 50 10 33 32(1) 32(1) 32(1)
swv16 50 10 52 51(1) 53(−1) 53(−1)
swv17 50 10 54 50(4) 51(3) 52(2)
swv18 50 10 54 53(1) 52(2) 52(2)
swv19 50 10 53 50(3) 51(2) 51(2)
swv20 50 10 55 53(2) 53(2) 52(3)
The improvement of HAV1, HAV2, and HAV3 with respect to the results reported by Framinan
(2005) is indicated in the bracket, and it shows the number of duplicate machines saved
A Deterministic Heuristic Algorithm to Minimize the Length … 127

5 Summary

The problem of transforming jobshops into flowshops is considered to minimize the

length of the manufacturing line and solved by using the three variants of determin-
istic heuristic algorithms (HAV1 to HAV3) in this work. The proposed algorithm
employs the concatenation of forward reduction, inverse reduction, and machine
elimination techniques along with a new local search scheme called two-pair adjacent
interchange scheme during seed solution generation on machine routings (manufac-
turing flow line) and during the main search process which uses the concatenation
patterns of JIS and MJSS on job sequence to obtain the reduced string length. The
computational evaluation of the three variants of the proposed deterministic heuristic
algorithm is carried out by comparing it with the existing reported work. The compu-
tational result shows that the HAV3 has performed well in terms of minimizing the
length of the manufacturing line (string length) and the number of duplicate machines
saved in most of the jobshop benchmark problem instances. In this work, the effec-
tiveness of the developed approach is tested with respect to minimization of the
length of manufacturing line corresponding to the problem of transforming jobshops
into flowshops, but its impact on the performance measures like makespan and total
flow time of jobs for the jobshop problems is not studied.

Acknowledgements The author is very much obliged to the reviewers and the editors for their
valuable suggestions and observations to improve the previous version of the paper. The author is
thankful to the Indian National Academy of Engineering (INAE) for the opportunity given to do
this research work under the mentoring of engineering teachers by the INAE Fellows scheme and
under Prof. C. Rajendran, INAE Fellow in IIT Madras.

Appendix 1

Forward Reduction Technique (Framinan, 2005)

Let us consider a jobshop problem with 6 jobs and 4 machines (i.e., n = 6 and m
= 4) and its machine routings as given below to illustrate various string reduction
techniques used in this work.
The initial job sequence is obtained by considering the jobs in the order as {1-2-
3-4-5-6} (or {6-5-4-3-2-1} in case of antithetic) in the generation of seed sequence,
and a supersequence of machines (S) is formed. In this forward reduction technique,
the supersequence of machines is scanned from left to right to check the machines
which are required to process the chosen job one at a time from the job sequence.
The feasibility of the supersequence of machines or manufacturing flow line is ascer-
tained for machine routings of every job, and reduced string length (L) is obtained.
The forward reduction technique is demonstrated in Table 3 through an example of
jobshop problem as given in Table 2.
128 J. Krishnaraj

Table 2 Machine routings of a jobshop (n = 6 and m = 4)

Job Machine routing
1 {1 – 2 – 3 – 4}
2 {1 – 3 – 2 – 4}
3 {1 – 4 – 2 – 3}
4 {2 – 3 – 1 – 4}
5 {2 – 4 – 1 – 3}
6 {3 – 4 – 1 – 2}

Table 3 Demonstration of forward reduction technique

Job sequence Supersequence of machines
{1-2-3-4-5-6} {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job Corresponding machines as per the respective machine routing of the job
Job 1 {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job 2 {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job 3 {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job 4 {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job 5 {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job 6 {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job sequence Shortest supersequence of machines
{1-2-3-4-5-6} {1-2-3-4-1-3-2-4-3}
Note that after the application of the forward reduction technique, the string length (or manufacturing
line) of the job sequence {1-2-3-4-5-6} is reduced to 9 from 24, and hence the saved number of
duplicated machines is 15

Appendix 2

Inverse Reduction Technique (Framinan, 2005)

The initial job sequence {1-2-3-4-5-6} (or {6-5-4-3-2-1}, in case of antithetic) is
considered in the generation of seed sequence, and a supersequence of machines
(S) is formed. In this inverse reduction technique, the supersequence of machines is
scanned from right to left to check the machines (i.e., machines in the order of right
to left from machine routings) which are required to process the chosen job one at a
time from reverse direction of the job sequence. The feasibility of the supersequence
of machines or manufacturing flow line is ascertained for machine routings of every
job, and reduced string length (L) is obtained. The inverse reduction technique is
demonstrated in Table 4 through an example of jobshop problem as given in Table 2.
A Deterministic Heuristic Algorithm to Minimize the Length … 129

Table 4 Demonstration of inverse reduction technique

Job sequence Supersequence of machines
{1-2-3-4-5-6} {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job Corresponding machines as per the respective machine routing of the job
Job 6 {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job 5 {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job 4 {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job 3 {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job 2 {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job 1 {1-2-3-4-1-3-2-4-1-4-2-3-2-3-1-4-2-4-1-3-3-4-1-2}
Job sequence Shortest supersequence of machines
{1-2-3-4-5-6} {1-2-3-1-4-2-4-1-3-4-1-2}
Note that after the application of the inverse reduction technique, the string length (or manufacturing
line) of the job sequence {1-2-3-4-5-6} is reduced to 12 from 24, and hence the saved number of
duplicated machines is 12. Also, note that the resultant supersequence of machines yielded from the
inverse reduction technique need not be the same as that of the resultant supersequence of machines
yielded by the forward reduction technique

Appendix 3

Machine Elimination Technique (Framinan & Ruiz-Usano, 2002)

This technique helps in further reduction of string length (L) by removing one
machine at a time from the supersequence of machines (S), and checking of feasi-
bility is made for the resultant supersequence of machines in terms of fulfilling the
machine routings of each job in the considered job sequence (say, {1-2-3-4-5-6}).
Let us consider the supersequence of machines {1-2-3-4-1-3-2-4-3} obtained from
the forward reduction technique for the application of the machine elimination tech-
nique which is demonstrated in Table 5 through an example of jobshop problem as
given in Table 2.

Appendix 4

Proposed Two-Pair Adjacent Interchange Scheme (TPAIS)

A new local search, two-pair adjacent interchange scheme (TPAIS), is proposed in
this work to explore a further decrease in string length. This local search is employed
for interchanging of adjacent machines found in the input supersequence of machines
(say, supersequence obtained from machine elimination technique employed during
seed generation, i.e., during Step 1 to Step 4 in Sect. 3.2). The step-by-step procedure
involved in TPAIS is explained below.
130 J. Krishnaraj

Table 5 Demonstration of machine elimination technique

Step no. Eliminated machine Resultant supersequence of Remarks
machines, S
1 Machine 1 found in {1-2-3-4-1-3-2-4-3} S does not fulfill the
position 1 machine routings of every
jobs; hence, machine 1 is
retained
2 Machine 2 found in {1-2-3-4-1-3-2-4-3} S does not fulfill the
position 2 machine routings of every
jobs; hence, machine 2 is
retained
3 Machine 3 found in {1-2-3-4-1-3-2-4-3} S does not fulfill the
position 3 machine routings of every
jobs; hence, machine 3 is
retained
4 Machine 4 found in {1-2-3-4-1-3-2-4-3} S does not fulfill the
position 4 machine routings of every
jobs; hence, machine 4 is
retained
5 Machine 1 found in {1-2-3-4-1-3-2-4-3} S does not fulfill the
position 5 machine routings of every
jobs; hence, machine 1 is
retained
6 Machine 3 found in {1-2-3-4-1-3-2-4-3} S fulfills the machine
position 6 routings of every jobs.
Hence, machine 3 is
eliminated and
supersequence of machines,
S, is updated with
{1-2-3-4-1-2-4-3}, L = 8
7 Machine 2 found in {1-2-3-4-1-2-4-3} Updated S does not fulfill
position 6 the required machine
routings of every jobs;
hence, machine 2 is retained
8 Machine 4 found in {1-2-3-4-1-2-4-3} S does not fulfill the
position 7 required machine routings
of every jobs; hence,
machine 4 is retained
9 Machine 3 found in {1-2-3-4-1-2-4-3} S does not fulfill the
position 8 required machine routings
of every jobs; hence,
machine 3 is retained and L
=8
Note that after the application of the machine elimination technique, the string length (or
manufacturing line) of the job sequence {1-2-3-4-5-6} is further reduced to 8 and hence the saved
number of duplicated machines is 16. Also, note that in some cases, for example, consider the
supersequence of machines {1-2-3-1-4-2-4-1-3-2}, and machine elimination technique does not
reduce the supersequence further
A Deterministic Heuristic Algorithm to Minimize the Length … 131

Step 1: Let us consider a supersequence of machines, obtained after machine

elimination technique during Step 1 to Step 4 in Section 3.2 (i.e., Assign the su-
persequence of machines in case of Step 1; in case of Step 2;
in case of Step 3 and in case of Step 4) and its corresponding string
length as (say, in case of Step 1; in case of Step 2; in
case of Step 3 and in case of Step 4).
Step 2: Set .
Step 3:
then
go to Step 10
else
proceed to Step 4.
Step 4:
then
proceed to Step 5
else
go to Step 6.
Step 5:
Step 5.1: Do the following for
{
Swap the machines appeared in positions and in ;
Apply forward reduction, inverse reduction, and machine elimi-
nation techniques on the perturbed sequence. Call the resultant
supersequence of machines obtained as and its string length as
.
}
Step 5.2: set , where,
Step 5.3:
then
assign the reduced supersequence of machines to and its
string length to , and proceed to step 6
else
go to step 6.
132 J. Krishnaraj

Step 6: Swap the machines found in positions and in sequence ;

Apply forward reduction, inverse reduction, and machine elimination
techniques on the perturbed sequence. Call the resultant supersequence of
machines obtained as and its string length as .

then
go to step 8
else
proceed to step 7.
Step 7:
Step 7.1: Do the following for
{
Swap the machines appeared in positions and in ;
Apply forward reduction, inverse reduction, and machine elimi-
nation techniques on the perturbed sequence. Call the resultant
supersequence of machines obtained as and its string length
as .
}
Step 7.2: set , where,
Step 7.3:
then
assign the reduced supersequence of machines to and
, and proceed to Step 8
else
go to Step 8.
Step 8:
then
assign the reduced supersequence of machines to and its
string length to , and proceed to Step 9
else
go to Step 9.
Step 9: Set and return to Step 3.
Step 10: Assign the finally returned supersequence of machines to and its
corresponding string length to in case of Step 1; similarly, to
and to in case of Step 2; to and to in case of Step 3; to
and to in the case of Step 4.

Appendix 5

Job Index-Based Insertion Scheme (JIS) (used by Krishnaraj et al., 2012, 2014,
2019)
A Deterministic Heuristic Algorithm to Minimize the Length … 133

To explore further reduction in supersequence of machines, the initial job sequence

(consider J S = {1 − 2 − 3 − . . . − n}) is subjected to JIS with a performance
measure of shortest string length. Note that whenever a perturbed job sequence is
obtained (i.e., supersequence of machines is initialized based on the perturbed job
sequence), the full set of procedures followed in Step 1 to Step 5 in Sect. 3.2 is
applied to get the reduced supersequence of machines and its string length which in
turn maximize the number of duplicate machines saved.
Let [k] represent the index of the job at position k of the initial job sequence J S
under consideration. The working of local search scheme is given below:

then insert job in position of the job sequence under consid-

eration and adjust the sequence accordingly by not changing the
relative positions of the other jobs; execute the set of procedures
followed in Step 1 to Step 5 as in Section 3.2 by initializing the
supersequence of machines based on the perturbed job sequence
to explore the decreased supersequence of machines, and its
string length, .
}
select the best job sequence among these sequences based on
minimum string length; if the string length is less than or equal to the
string length of the considered sequence (i.e., sequence ), then inter-
change the sequence by the best one obtained and update the superse-
quence of machines, and its string length, .
Also update the best shortest common supersequence of machines
and its shortest string length if necessary (i.e., )
}

Appendix 6

Modified Job Index-Based Swap Scheme (MJSS) (originally used by Rajendran

and Ziegler 2004)
Let JS be the job sequence under consideration and [k] be the index of the job at
position k of this sequence. The working of local search scheme is given below:
134 J. Krishnaraj

then swap job with job , with remaining jobs in unal-

tered; execute the set of procedures followed in Step 1 to Step 5
as in Section 3.2 by initializing the supersequence of machines
based on the perturbed job sequence to explore the reduced su-
persequence of machines, and its string length, .
Also update the job sequence if the string length corresponds
to the supersequence based on the perturbed job sequence is less
than the string length, . (Note that this updating of job sequence
is new in this work)
}
select the best job sequence among these sequences based on
minimum string length; if the string length is less than or equal to the
string length of the considered sequence (i.e., sequence ), then inter-
change the sequence by the best one obtained and update the superse-
quence, and its string length, .
Also update the best shortest common supersequence and its short-
est string length if necessary (i.e., )
}

References

Applegate, D., & Cook, W. (1991). A computational study of the job-shop scheduling problem.
ORSA Journal on Computing, 3(2), 149–156.
Barone, P., et al. (2001). An approximation algorithm for the shortest common super-sequence
problem: An experimental analysis. In: Proceedings of the 2001 ACM symposium on applied
computing, ACM.
Branke, J., Middendorf, M., & Schneider, F. (1998). Improved heuristics and a genetic algorithm
for finding short supersequences. Operations Research-Spektrum, 20(1), 39–46.
Dondi, R. (2013). The constrained shortest common supersequence problem. Journal of Discrete
Algorithms., 21, 11–17.
Framinan, J. M. (2005). Efficient heuristic approaches to transform job shops into flow-shops. IIE
Transactions, 37(5), 441–451.
Framinan, J. M., & Ruiz-Usano, R. (2002). On transforming job-shops into flow-shops. Production
Planning and Control, 13(2), 166–174.
Fraser, C. B., & Irving, R. W. (1995). Approximation algorithms for the shortest common
supersequence. Nordic Journal of Computing., 2(3), 303–325.
French, H. (1982). Sequencing and scheduling—an introduction to the mathematics of the job shop.
Ellis Hoewood John Wiley & Sons.
Jiang, T., & Li, M. (1995). On the approximation of shortest common supersequences and longest
common subsequences. SIAM Journal on Computing Archive., 24(5), 1122–1139.
A Deterministic Heuristic Algorithm to Minimize the Length … 135

Kimms, A. (2000). Minimal investment budjets for flow line configuration. IIE Transactions, 32(4),
287–298.
Knolmayer, G., Mertens, P., & Zeier, A. (2002). Supply chain management based on SAP systems.
Berlin: Springer.
Krishnaraj, J., Pugazhendhi, S., Rajendran, C., & Thiagarajan, S. (2012). A modified ant-colony opti-
mization algorithm to minimize the completion time variance of jobs in flowshops. International
Journal of Production Research, 50(20), 5698–5706.
Krishnaraj, J., Pugazhendhi, S., Rajendran, C., & Thiagarajan, S. (2014). A heuristic algorithm to
minimize the total flowtime of jobs in permutation flowshops. International Journal of Industrial
and Systems Engineering, 4(4), 511–532.
Krishnaraj, J., Pugazhendhi, S., Rajendran, C., & Thiagarajan, S. (2019). Simulated annealing algo-
rithms to minimize the completion time variance of jobs in permutation flowshops. International
Journal of Industrial and Systems Engineering, 31(4), 425–451.
Lawrence, S. (1984). Supplement to resource constrained project scheduling: An experimental
investigation of heuristic scheduling techniques. Pittsburgh, PA: Graduate School of Industrial
Administration, Carnegie-Mellon University.
Michel, R., & Middendorf, M. (1999). An ACO algorithm for the shortest common super-sequence
problem.
Mousavi, S. R., Bahri, F., & Tabataba, F. S. (2012). An enhanced beam search algorithm for the
shortest common supersequence problem. Engineering Applications of Artificial Intelligence,
25(3), 457–467.
Ning, K., & Leong, H. W. (2006). Towards a better solution to the shortest common super-sequence
problem: The deposition and reduction algorithm. BMC Bioinformatics, 7(Suppl 4), S12).
Noaman, M. M., & Jaradat, A. S. (2011). Solving shortest common supersequence problem using
artificial bee colony algorithm. International Journal of ACM Jordan., 2(3), 180–185.
Rajendran, S., Rajendran, C., & Ziegler, H. (2010). An ant-colony algorithm to transform
jobshops into flowshop: A case of shortest-common-supersequence stringology problem. In 5th
International ICST conference, BIONETICS 2010, Boston, USA, December 2010.
Rajendran, C., & Ziegler, H. (2004). Ant-colony algorithms for permutation flowshop scheduling
to minimize makespn/total flowtime of jobs. European Journal of Operational Research, 155(2),
426–438.
Saifullah, K. C. M., & Islam, M. R. (2016). Chemical reaction optimization for solving the shortest
common supersequence problem. Computational Biology and Chemistry, 64, 82–93.
Storer, R. H., Wu, S. D., & Vaccari, R. (1992). New search spaces for sequencing problems with
applications to job-shop scheduling. Management Science, 38(10), 1495–1509.
Timkovsky, V. G. (1989). Complexity of common subsequences and supersequences problems and
related problems. Cybernetics and Systems Analysis, 25(5), 565–580.
 Data Analytics, Qualitative,
and Simulation Approaches to Operations
and Supply Chain Issues
A Classification Algorithm Based
on Linear Regression and Linear
Programming for Predicting the Breast
Cancer

Sakthivel Madankumar

Abstract In this study, we consider the classification problem in the healthcare

domain, and the objective of the problem is to diagnose the breast cancer (binary
classification: malignant/benign) based on the number of different features with
respect to tumor characteristics. We propose an algorithm to solve this classification
problem. The proposed algorithm first eliminates the linear dependency between the
input features and then focuses on enriching the feature set based on these indepen-
dent attributes in order to capture the curvilinear function between the classes. The
proposed algorithm, subsequently, uses the multiple linear regression (MLR) model
on the training data set to capture the relationship between the response/class variable
(malignant/benign) and the set of features (from the enriched feature set). Thereafter,
the proposed linear programming (LP) model makes use of the set of relatively influ-
ential attributes from the MLR based on the absolute values of the coefficients, to
find the classification function/expression for predicting the breast cancer using the
training data set. As a part of this process, LP model also finds the initial thresholds
where b and b + 1 are thresholds for malignant and benign classes, respectively.
In the final phase, the proposed algorithm fine-tunes the thresholds obtained in the
LP model through a search process, to determine the exact boundary between the
target classes (malignant/benign) using the validation data set. In order to evaluate
the performance of the algorithm, we use the Wisconsin diagnostic breast cancer
(WDBC) data set from the University of California—Irvine machine learning repos-
itory, and we compare the performance with the existing algorithm in the literature.
From the results, we observe that the proposed algorithm performs better in terms of
the accuracy to predict the breast cancer.

Keywords Cancer diagnosis · Classification algorithm · Linear regression · And

linear programming

S. Madankumar (B)
Trimble Information Technologies India Private Limited, Chennai 600113, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 139
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_7
140 S. Madankumar

Notations

M: total number of instances/observations.

N0 : total number of input features/attributes in each observation.
N1 : total number of features retained after the elimination of multicollinearity
among the standardized features/attributes.
N2 : total number of features (after the augmentation) at the end of transforma-
tion using the exponential function.
N3 : total number of features (after the augmentation) at the end of transforma-
tion using the logistic function.
N4 : total number of features (after the augmentation) at the end of transforma-
tion using the cosine function.
N5 : total number of features after considering the contributions of attributes
from their transformations using higher order polynomials.
N6 : total number of features after considering the contributions based on the
interaction among original as well as transformed attributes with higher
order polynomials.
i: index for observations.
j, c: indices for features.
ai j : value of attribute/feature j for observation i.
fi j : value of feature j for observation i after the transformation.
 N0
ai j
xj: mean of attribute j; where x j = i=1
.
N0 √ M
i=1 (ai j −x j )×(ai j −x j )
sj: standard deviation of the attribute j; where s j = M
.
τ: threshold value (upper limit) for the Variance Inflation Factor (VIF).
λ: set of multipliers to be used in the exponential and logistic func-
tions, and for this study we assumed the following values for λ; λ =
{1, 2, 3, 4, 5, 6, 7, 8, 9, 10}.
|λ|: total number of elements in set λ.
λk : an element from set λ; and λk ∈ λ; for example, when k assumes value 1
then λk = λ1 = 1.
ε: a constant with a very small value 0.00001.
P: set of power terms considered with respect to the higher-order polynomials,
and for this study we assumed the following values for P; P = {2, 3, 4}.
Q: set of power terms considered with respect to the interaction effect of
attributes, and for this study we assumed the following values for Q;
Q = {1, 2, 3, 4}.
pk : an element from set P; and pk ∈ P; for example, when k assumes value 1
then pk = p1 = 2.
qk : an element from set Q; and qk ∈ Q; for example, when k assumes value 1
then qk = q1 = 1.
|P|: total number of elements in set P.
|Q|: total number of elements in set Q.
β0 : an intercept of the estimated regression expression.
A Classification Algorithm Based on Linear Regression … 141

βj: coefficient of each attribute j in the estimated regression expression.

yi : value of the response variable for observation i (binary classification: 0 for
malignant and 1 for benign).

yi : expression to predict the response for observation i.
yi : value of the estimated response for observation i.
i : error/deviation of yi from yi in MLR.
α: value for the regularization parameter; and in this study, we have set the
value as 10 for α.
D: a set of relatively-influential attributes based on the absolute value of the
coefficients (from MLR).
γj: a real (unrestricted) variable to capture the coefficient for attribute j in the
LP model.
errori : a real (non-negative) variable to capture the error in the expression in the
LP model.
b: a real (unrestricted) variable to capture the boundary/threshold in the LP
model.

1 Introduction

Health care is a field/domain that focuses on improving the health of the people
through the application of preventive measures, and the procedures/measures to diag-
nose the diseases at the right (or early) stage, and provide the treatments to cure the
diseases, in order to recover the infected or diseased people. The advancement in
the field of artificial intelligence (AI) and machine learning (ML) has given the hope
to leverage the respective methods and technologies in the healthcare field, in order
to aid the doctors/physicians to identify/predict the diseases at the right stage by
analyzing the medical records of the patients; then, the physicians can make use of
this prediction as an input to further analyze as well as provide treatments at the right
stage to the patients.
Cancer is a critical disease with high mortality rate in health care; Siegel et al.
(2012) studied the cancer statistics and estimated the total number of new cases
(1,639,910) and deaths (577,190) due to cancer in the United States for the year
2012. In their study, they also estimated that breast cancer was accounting for 29%
out of the new female cancer cases. Typically, mammograms are used by physicians
and radiologists to predict the cancer, but in general, there is a high variability in
the prediction by different radiologists. This was also evident in the study conducted
by Elmore et al. (1994), and only 3% of the cancers were identified. The advance-
ment in technologies (in the medical field and also in the data analytics) and the
computing power has led us to extract more tumor features and also apply the ML
algorithms to treat the diagnosis problem as a classification problem to predict the
breast cancer. Studies by earlier researchers (Wolberg et al., 1995; Pena-Reyes &
142 S. Madankumar

Sipper, 1999) proved that data mining techniques can be used to solve this problem
as a classification problem.
Researchers have traditionally used support vector machine (SVM) techniques to
solve this classification problem of predicting breast cancer (Bennett & Blue, 1998;
Huang et al., 2008; Akay, 2009; Prasad et al., 2010). Zheng et al. (2014) studied
the Wisconsin diagnostic breast cancer (WDBC) data set from the University of
California—Irvine machine learning repository, and they applied K-means algorithm
to reduce the number of features and then applied the SVM algorithm on the reduced
feature set to predict the breast cancer.
In this work, we apply operations research (OR) techniques to solve the classifica-
tion problem for predicting the breast cancer, and our proposed algorithm comprises
multiple linear regression (MLR) and linear programming (LP) techniques to solve
the classification problem. This is one of the few attempts in applying OR techniques
to solve the classification problem. Madankumar et al. (2017) proposed an LP-based
classifier for solving the classification problem, and the respective algorithm may
not work well on the high-dimensional data set due to the fact that the respective LP
classifier was defining a decision variable (in the LP model) for each feature, and
when the number of features is high, then the resulting number of LP decision vari-
ables will also be high which would make the model to be unsolvable in reasonable
CPU time.
In this study, we introduce a step where we apply the MLR to identify the features
that are promising, and then, we apply the LP on the reduced feature set to find the
boundary between the classes. We compare the efficiency of the proposed algorithm
with the results obtained by Zheng et al. (2014) using their K-SVM algorithm (K-
means followed by SVM), and we observe that our algorithm performs better in terms
of the accuracy to predict the breast cancer. Please note that the terms “features”
and “attributes” are used interchangeably in this paper, and they both represent the
measured inputs (the independent variables) of the problem.
This chapter is organized as follows: In Sect. 2, we present the summary of the
proposed algorithm, in Sect. 3, we present the algorithm in detail with the respective
implementation details for each step, and in Sect. 4, we present the experimental
results of the proposed algorithm where we compare the accuracy of the proposed
algorithm with the ML algorithm available in the literature for predicting the breast
cancer.

2 Proposed Methodology

In this study, we consider the classification problem in the healthcare domain. The
objective of the classification problem is to diagnose breast cancer (binary classi-
fication: malignant/benign) based on the number of different features with respect
to tumor characteristics. A set of instances are given with the respective diagnostic
results, and for each instance, the values with respect to the input features (different
tumor characteristics) and the corresponding classification of the tumor (whether it is
A Classification Algorithm Based on Linear Regression … 143

benign and malignant) are available in the data set. We propose an algorithm to solve
this classification problem (binary classification: malignant/benign), the summary of
the proposed algorithm is presented in this section, and we also present the overview
of the algorithm as a sequence of steps in Fig. 1.
The proposed algorithm first eliminates the linear dependency between the input
features/attributes with the help of Variance Inflation Factor (VIF) (Step 2 in Sect. 3);
and then focuses on enriching the feature set based on the resulting independent
attributes, by transforming the independent attributes using non-linear functions such
as exponential (Step 3.1 in Sect. 3), logistic (Step 3.2 in Sect. 3) and cosine (Step 3.3
in Sect. 3) functions; and then further transformation of attributes through higher-
order polynomials (Step 3.4 in Sect. 3), and then further enrichment by having a
set of additional features to track the interaction (Step 3.5 in Sect. 3) among the

Fig. 1 Overview of the proposed algorithm

144 S. Madankumar

independent attributes from their polynomials. These steps are carried out mainly to
capture the nonlinear/curvilinear boundary between the classes (two predetermined
classes: malignant and benign).
Then, the proposed algorithm splits the data set in to training data set, validation
data set, and test data set. Subsequently, the proposed algorithm uses the multiple
linear egression (MLR) (see Step 5 in Sect. 3) model on the training data set, in
order to capture the relationship between the response/class variable and the set of
features (from the enriched feature set). As a part of this step, the MLR also finds the
coefficient for each attribute based on its contribution toward the response variable.
The proposed algorithm uses the ridge regularization in order to avoid the effect of
over fitting on the training data set, so that the model can even work well for the
new/unseen samples when it is tested against test and validation data set.
Thereafter, the proposed linear programming (LP) model (see Step 6 in Sect. 3)
makes use of the set of relatively influential attributes from the MLR (based on
the absolute values of the coefficients) in order to find the classification func-
tion/expression for predicting the breast cancer using the training data set, and as
a part of this process, LP model also finds the initial thresholds where b and b +
1 are thresholds for malignant and benign classes, respectively. In the final phase,
the proposed algorithm fine-tunes the thresholds obtained in the LP model through a
search process (see Step 7 in Sect. 3), to determine the exact boundary between the
target classes (malignant/benign) using the validation data set. Finally, the proposed
algorithm makes use of the test data set to evaluate the model in terms of accuracy,
and the respective results are presented in Sect. 4.

3 Proposed Algorithm

Step 1: Standardization
The algorithm first standardizes each feature to transform the mean of the attribute
to 0 and standard deviation of the respective attribute to 1.
 
f i, j = ai j − x j /s j , i = 1, 2, . . . , M; and j = 1, 2, . . . , N0 . (1)

A Numerical Illustration
We consider a sample data set presented in Table 1 to numerically illustrate Step 1,
and this data set consists of four features/attributes (N0 = 4) and one target/dependent
variable for binary classification.
As prescribed by the algorithm, we apply Step 1 to this data set, and the corre-
sponding standardized features are presented in Table 2; Figure 2 also depicts the
respective computation for a sample value.
A Classification Algorithm Based on Linear Regression … 145

Table 1 Sample data set for numerical illustration

Classification (yi ) Feature 1 Feature 2 Feature 3 Feature 4
0 3.13 36.00 108.63 19.25
0 5.90 41.00 124.18 33.60
0 4.42 29.00 87.88 25.00
0 8.72 24.00 73.74 46.00
1 8.13 21.00 64.63 42.75
1 4.06 33.00 99.81 23.60
1 6.92 46.00 139.38 39.20
1 8.06 10.00 31.61 41.30
1 4.98 42.00 127.00 29.10
1 6.88 49.00 148.38 39.30

Table 2 Values of the

Feature 1 Feature 2 Feature 3 Feature 4
standardized attributes
−1.64 0.25 0.23 −1.68
−0.12 0.68 0.68 −0.04
−0.93 −0.35 −0.36 −1.02
1.43 −0.78 −0.77 1.39
1.10 −1.04 −1.03 1.01
−1.13 −0.01 −0.02 −1.18
0.44 1.11 1.12 0.61
1.06 −1.99 −1.99 0.85
−0.63 0.77 0.76 −0.55
0.42 1.37 1.38 0.62

Fig. 2 Effect of applying the standardization step for a sample value {0.25} with x j = 6.12 and
s j = 1.82

Step 2: Elimination of multicollinearity

In order to eliminate the multicollinearity among the standardized features/attributes,
this step calculates the variance inflation factor (VIF) for each feature/attribute,
and VIF describes the factor by which a respective attribute’s esti-
mated coefficient is inflated because of the correlation with other attributes
(O’brien, 2007). In this study, we make use of the Python library
“statsmodels.stats.outliers_influence.variance_inflation_factor” to calculate the VIF
146 S. Madankumar

for each attribute, and we follow an iterative procedure to eliminate the multi-
collinearity. In this process, we choose the first attribute which has the VIF greater
than the threshold, and we first eliminate the attribute from the feature set. Then, we
calculate the VIF for the remaining attributes, and we repeat the same process till
the remaining attributes satisfy the condition that the corresponding VIFs are less
than the threshold. In this study, we have considered the threshold (τ ) as 5 for VIF
so that an attribute is eliminated for further analysis when the respective attribute is
dependent on other remaining attributes with R 2 (coefficient of determination) value
greater than 0.8. Let us denote the number of features retained after the elimination
of multicollinearity among the attributes as N1 , and these set of features are used
for further analysis and consideration. This step is mainly carried out to reduce the
change in estimated regression coefficients when there is a linear dependency among
the attributes/features.
A Numerical Illustration
In this numerical illustration, we eliminate the multicollinearity among the standard-
ized attributes as prescribed in Step 2 for the standardized data set presented in Table
2; this step eliminates the standardized features (Feature-1 and Feature-2) through the
iterative procedure described in Step 2 because of their corresponding linear depen-
dency with Feature-3 and Feature-4, then, the remaining set of features (N1 = 2) are
considered for further analysis, and the respective attributes are presented in Table
3.
Step 3: Enrichment of feature set
The procedure then augments the feature set (by using the reduced set of features N1
obtained from Step 2) by considering the following aspects:
• Contribution of attributes from their transformations using nonlinear functions
such as exponential, logistic and cosine functions.
• Contributions of attributes from their transformations using higher-order polyno-
mials.

Table 3 Retained features

Feature 3 Feature 4
(N1 = 2) after the elimination
of multicollinearity 0.23 −1.68
0.68 −0.04
−0.36 −1.02
−0.77 1.39
−1.03 1.01
−0.02 −1.18
1.12 0.61
−1.99 0.85
0.76 −0.55
1.38 0.62
A Classification Algorithm Based on Linear Regression … 147

Fig. 3 Effect of applying the exponential function on the standardized feature

• Contributions based on the interaction among the attributes with their higher-order
polynomials.

Step 3.1: Transformation of attributes using the exponential function

Our procedure applies the exponential transformation for the standardized features
(N 1 ) retained in Step 2, with the consideration of set of λ values, and the function
used for this transformation is defined in Eq. (2); the total number of features (after
the augmentation) at the end of this transformation would be N2 , and it is defined
in Eq. (3); the effect of applying this transformation on the standardized attribute is
captured in Fig. 3 with varying values (from −3.0 to 3.0) of the standardized attribute.

f i,c = e−λk × fi, j i = 1, 2, . . . , M; j = 1, 2, . . . , N1 ; λk ∈ λ;

k = 1, 2, . . . , |λ|; and c = N1 + ( j − 1) × |λ| + k (2)

N2 = N1 + N1 × |λ|. (3)

A Numerical Illustration
As prescribed in Step 3.1, we apply the exponential transformation for the standard-
ized features retained (N1 = 2) in Step 2 (for the data set presented in Table 3). But
for ease of understanding, we only apply the transformation for the first observation
in the sample data set with the consideration of subset of λ = {1, 5, 10} values,
the corresponding transformed values with respect to first observation are presented
in Table 4, and if we consider set λ = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}, then the total
number of features at the end of this transformation is N2 = 22; Fig. 4 also depicts
the respective computation for a sample value.
148 S. Madankumar

Table 4 Effect of exponential transformation on the sample data set

Standardized feature Transformed value Transformed value Transformed value
when λ1 = 1 when λ2 = 5 when λ3 = 10
0.23 0.79 0.32 0.10
−1.68 5.38 4488.29 20, 144, 720.26

Fig. 4 Effect of applying the exponential function for a sample value {0.23} with λ1 = 1

Step 3.2: Transformation of attributes using the logisticfunction

The procedure applies the logistic transformation for the standardized features
(N 1 ) retained in Step 2, with the consideration of set of λ values, and the function
used for this transformation is defined in Eq. (4); the total number of features (after
the augmentation) at the end of this transformation would be N3 , and it is defined
in Eq. (5); the effect of applying this transformation (with λ1 = 1) on the standard-
ized attribute is captured in Fig. 5 with varying values (from −3.0 to 3.0) of the
standardized attribute.

i = 1, 2, . . . , M; j = 1, 2, . . . , N1 ; λk ∈ λ;
f i,c =  1 
k = 1, 2, . . . , |λ| and c = N2 + ( j − 1) × |λ| + k. (4)
f i, j
−λk × 6
1+e

N3 = N2 + N1 × |λ|. (5)

Fig. 5 Effect of applying the logistic function on the standardized feature

A Classification Algorithm Based on Linear Regression … 149

Table 5 Effect of logistic transformation on the sample data set

Standardized feature Transformed value Transformed value Transformed value
when λ1 = 1 when λ2 = 5 when λ3 = 10
0.23 0.51 0.55 0.60
−1.68 0.43 0.20 0.06

Fig. 6 Effect of applying the logistic function for a sample value {0.23} with λ1 = 1

A Numerical Illustration
As prescribed in Step 3.2, we apply the logistic transformation for the standardized
features retained (N1 = 2) in Step 2 (which are presented in Table 3). But for ease
of understanding, we only apply the transformation for the first observation in the
sample data set with the consideration of subset of λ = {1, 5, 10} values, and the
corresponding transformed values with respect to first observation are presented in
Table 5; if we consider set λ = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}, then the total number of
features at the end of this transformation is N3 = 42; Fig. 6 also depicts the respective
computation for a sample value.
Step 3.3: Transformation of attributes using the cosine function
The procedure applies the cosine transformation for the standardized features (N 1 )
retained in Step 2, and in order to perform this transformation, this step first scales
the attribute to take the value between 0 and 1 and then multiply the respective value
by π (radians), and the function used for this transformation is defined in Eq. (6); the
total number of features (after the augmentation) at the end of this transformation
would be N4 , and it is defined in Eq. (7). The effect of applying this transformation
is captured in Fig. 7 with varying values (from −3.0 to 3.0) of the standardized
attribute.
⎛ ⎧   ⎫⎞
⎨ f i j − min f i j ⎬
f ic = cos⎝π × ⎠
1≤i≤M
   
⎩ max f i j − min f i j + ε ⎭
1≤i≤M 1≤i≤M

i = 1, 2, . . . , M; j = 1, 2, . . . , N1 ; and c = N3 + j. (6)

N4 = N3 + N1 . (7)
150 S. Madankumar

Fig. 7 Effect of applying the cosine function on the standardized attribute

A Numerical Illustration
As prescribed in Step 3.3, we apply the cosine transformation for the standardized
features retained (N1 = 2) in Step 2 (which are presented in Table 3). But for
ease of understanding, we only apply the transformation for the first observation
in the sample data set, the corresponding transformed values with respect to first
observation are presented in Table 6, and the total number of features at the end of
this transformation is N4 = 44; Fig. 8 also depicts the respective computation for a
sample value.
Step 3.4: Transformation of attributes using higher-order polynomials
The procedure applies this transformation for the standardized features (N 1 ) retained
in Step 2, in order to capture the contribution of attributes from their higher-order
polynomials, and the function used for this transformation is defined in Eq. (8); the
total number of features (after the augmentation) at the end of this transformation

Table 6 Effect of cosine

Standardized feature Transformed value
transformation on the sample
data set 0.23 −0.48
−1.68 1.00

 
Fig. 8 Effect of applying the cosine function for a sample value {0.23} with min f i j = −1.99
  1≤i≤M
and max f i j = 1.38
1≤i≤M
A Classification Algorithm Based on Linear Regression … 151

would be N5 , and it is defined in Eq. (9).

 p
f ic = f i j k ∀i = 1, 2, . . . , M; j = 1, 2, . . . , N1 ; k = 1, 2, . . . , |P|;
(8)
pk ∈ P; and c = N4 + ( j − 1) × |P| + k.

N5 = N4 + N1 × |P|. (9)

A Numerical Illustration
As prescribed in Step 3.4, we transform the values of the first observation in the
sample data set through their respective higher-order polynomials for the standard-
ized features retained (N1 = 2) in Step 2 (which are presented in Table 3), the
corresponding transformed values with respect to first observation are presented in
Table 7, the total number of features at the end of this transformation is N5 = 50;
and Fig. 9 also depicts the respective computation for a sample value.
Step 3.5: Transformation of attributes using the interaction between the
attributes
The procedure applies this transformation, in order to capture the contribution of
attributes from the interaction effect between the attributes/features (N 1 ) retained in
Step 2, and the function used for this transformation is captured in Eq. (10); the total
number of features (after the augmentation) at the end of this transformation would
be N6 , and it is defined in Eq. (11).
 q  q 
f ic = f i j k × f i j  k
∀i = 1, 2, . . . , M; j = 1, 2, . . . , (N1 − 1);
∀ j  = ( j + 1), ( j + 2), . . . , N1 where j < j  ;
k = 1, 2, . . . , |Q|; k  = 1, 2, . . . , |Q|; qk and qk  ∈ Q. (10)

Table 7 Effect on the contribution of attributes from their higher-order polynomials on the sample
data set
Standardized feature Transformed value Transformed value Transformed value
when p1 = 2 when p2 = 3 when p3 = 4
0.23 0.05 0.01 0.00
−1.68 2.83 −4.76 8.00

Fig. 9 Effect of higher-order polynomials for a sample value {0.23} with p1 = 2

152 S. Madankumar
 
(N1 − 1)
N6 = N5 + N1 × × |Q| × |Q|. (11)
2

A Numerical Illustration
As prescribed in Step 3.5, we transform the values of the first observation in the
sample data set based on the interaction effect between the standardized features
retained (N1 = 2) in Step 2 (which are presented in Table 3), the corresponding
transformed values with respect to first observation are presented in Table 8, and the
total number of features at the end of this transformation is N6 = 66; Fig. 10 also
depicts the respective computation for a sample set of values.
Step 4: Scaling
The algorithm finally scales each feature (including the features obtained in Step 3:
enrichment of the feature set) to take the value between 0 and 1, and the respective
function is defined in Eq. (12):
⎧ ⎫
⎪   ⎪
⎪
⎨ f i j − min fi j ⎪
⎬
1≤i≤M
fi j =   ∀i = 1, 2, . . . , M; and j = 1, 2, . . . , N6 .
⎪
⎪     ⎪
⎩ max f i j − min f i j + ε ⎪
⎭
1≤i≤M 1≤i≤M
(12)

Step 5: Multiple Linear Regression(MLR) using the training data set (Mtrain )
In this step, the algorithm applies MLR on the training data set to fit a regression
line between N6 independent attributes (that are obtained/transformed/scaled in the
previous steps: Step 2 to Step 4) and the dependent/response variable y (malignant
or benign). In order to avoid the effect of overfitting, the algorithm considers ridge
regularization as defined in Eq. (15). So, the objective function (zlr ) focuses on
minimizing the magnitude of the coefficients (β j ) in addition to the minimization

Table 8 Effect on the contribution of features based on their interactions on the sample data set
Standardized Standardized Power term qk Power term qk  Transformed Value
feature ( j = 1) feature ( j = 2)
0.23 −1.68 1.00 1.00 −0.39
0.23 −1.68 2.00 4.00 0.44

Fig. 10 Effect of interaction between the values {0.23, 1.681} with qk = 1 and qk  = 1
A Classification Algorithm Based on Linear Regression … 153

of errors (i ). The MLR finds the coefficient (β j ) for each attribute j based on the
respective attribute’s contribution toward target variable. Finally, the value of the
expression yi is captured in variable yi , and the value of variable yi and the values
of the coefficients β j are used as inputs in the LP model (in Step 6).


N5
 
yi = β0 + β j × f i j , ∀i = 1, 2, . . . , Mtrain . (13)
j=1

i = yi − yi , ∀i = 1, 2, . . . , Mtrain (14)


P
Minimize : zlr = i2 + α × β 2j (15)
j=1

yi = yi (16)

Step 6: Linear Programming(LP) model using the training data set (Mtrain )
In this step, we make use of the coefficients (β j ) obtained from the MLR (in
Step 5), and then, we select a set of D relatively influential attributes based on the
absolute values of the coefficients (β j ) for further analysis. In order to select D
relatively influential attributes, we sort the values of coefficients β j in descending
order, and then, we select the first D attributes where |D| is min {N6 , Mtrain }. In
addition, we also use the value yi obtained from the MLR (in Step 5), and then,
we solve the following minimization problem to minimize the errors (defined in
Eq. (17)). So as a part of this process, the proposed LP model finds the classification
function/expression for predicting the breast cancer using the training data set, by
varying/searching the coefficients (γ j ) of the relatively influential attributes in set
D. In this process, LP model also finds the initial thresholds where b and b + 1 are
thresholds for malignant and benign classes, respectively.


M train

Minimize : zlp = errori (17)

i=1

For each observation in training data set, and 

when the observation i is of cate-

  
gory/class 0 (malignant tumor), then expression yi + j∈D γ j × f i j − errori
should be less than the boundary (b):
 
yi + γ j × f i j − errori ≤ b. (18)
j∈D

when
 the observation i is of category/class
 1 (benign tumor), then expression

  
yi + j∈D γ j × f i j + errori should be greater than the boundary (b + 1):
154 S. Madankumar
 
yi + γ j × f i j + errori ≥ (b + 1). (19)
j∈D

Step 7: Determination of exact boundary for the target classes using the
validation data set (Mval )
In this step, the algorithm fine-tunes the thresholds obtained in the LP model
(Step 6) through a search process to determine the exact boundary between the
categories/classes (using validation data set Mval ). So, the algorithm increments the
value b from 0 to 1 with the help of step size (0.05), and for each value of b , the
algorithm performs the following:
1. For each observation in validation
 data set (Mval), the algorithm calculates the
  
value for the expression yi + j∈D γ j × f i j .
2. If the calculated value is less than (b + b ), then the corresponding observation
is classified as class/category 0 (malignant tumor), and if the calculated value
is greater than (b + b ), then the corresponding observation is classified as
class/category 1 (benign tumor).
3. Calculate the accuracy value for the given (b + b ) for the validation data set
(Mval ), choose the best (b + b ) based on the accuracy value for the validation
data set (Mval ), and it is defined as follows:

Accuracy
(number of observations predicted correctly in validation data set)
= × 100.
(number of observations in validation data set)
(20)

So through Steps 5 to 7, the classification model is trained and fine-tuned using

the training and validation data set, and then with the help of the classification func-
  
tion/expression yi + j∈D γ j × f i j obtained in Step 6 and the exact boundary
(b + b ) obtained in Step 7, the algorithm can classify/predict new observation to its
respective target class (malignant/benign tumor).
Step 8: Evaluation of the classifier using the test data set (Mtest )
In this step, algorithm makes use of the unseen/test data set (Mtest ) to evaluate
the model in terms of the accuracy to predict the target classes  (and it is defined

in Eq. (21)). It makes use of the classification expression (yi + j∈D γ j × f i j )
obtained in Step 6 and the exact boundary (b + b ) obtained in Step 7, to clas-
sify/predict each observation in the test data set to corresponding target class
(malignant/benign tumor).

(number of observations predicted correctly in test data set)

Accuracy = × 100
(number of observations in test data set)
(21)
A Classification Algorithm Based on Linear Regression … 155

4 Experimental Results

The proposed algorithm is evaluated with the help of the WDBC data set, and we
compare the performance of the proposed algorithm with the results of the K-SVM
algorithm proposed by Zheng et al. (2014). Please find the link to the data set: UCI
Machine LearningRepository: Breast Cancer Wisconsin (Diagnostic) Data Set.
The WDBC data set contains a set of instances/observations (569 observations)
for which the corresponding diagnostic results/classification of the tumor (whether
it is benign and malignant) are given; each observation contains 31 different tumor
characteristics (features).
In order to have a fair comparison of the proposed algorithm with the K-SVM by
Zheng et al. (2014), in this study, the proposed algorithm is also validated with the help
of tenfold cross validation as validated by Zheng et al. (2014). First, we apply tenfold
cross validation on the given data set to split the data into ten equal-sized groups,
and of the 10 groups, one group is treated as the test data set, and then, we again
apply tenfold cross validation to split the remaining groups (9 groups) into training
data set and validation data set. The same steps are repeated ten times/iterations so
that each of ten groups are used exactly once as the test data set.
We present the number of features that gets augmented at each stage of feature-
extraction process in Table 9, and in total, 31 features are enriched into 1573 features
through various steps (from Step 1 to Step 4). In Step 5, we apply the MLR technique
on these 1573 attributes to find the values of the variables yi and the coefficients
β j . These values are used in the LP model (in Step 6); as a part of this step, the
algorithm selects 460 relatively influential attributes (|D| = min{N6 , Mtrain } = min
(1573, 460) = 460), and from this set D, we present the first ten relatively influential
attributes (out of 460 attributes) in Table 10.
Zheng et al (2014) followed a different approach, their algorithm first focused
on reducing the number of features (from 31 to 6 features), and this was achieved
with the help of K-means algorithm by finding the representative clusters (with
the respective centroids) for both type of tumors; then, a membership function was
applied on the observations to find the similarity between each observation and the
identified centroids (for the representative clusters). Thus, the reduced number of

Table 9 Number of features over various stages of feature enrichment phase

Stages of feature enrichment N0 N1 N2 N3 N4 N5 N6
Number of features 31 13 143 273 286 325 1573

Table 10 Coefficients (β j ) of the first ten relatively influential attributes from the set D
Attributes 1 2 3 4 5 6 7 8 9 10
in set D
Coefficients 0.198 0.171 0.144 −0.132 0.118 0.092 −0.083 0.071 −0.070 0.067
(β j )
156 S. Madankumar

Table 11 Accuracy of the proposed classifier (for test data set) over the iterations in tenfold cross
validation
Iterations 1 2 3 4 5 6 7 8 9 10
Accuracy 96.55% 96.55% 98.24% 98.24% 98.24% 98.24% 100% 100% 100% 96.42%

features for each observation represented the similarity between this sample and the
identified clusters, these features then made use of to diagnose the cancer (through
SVM), and overall accuracy of their algorithm was 97.38%.
We also present the results of the proposed classifier for the test data set (Step 8),
across ten iterations (for tenfold cross validation) in Table 11, and we observe that
on average (across ten iterations of tenfold cross validation), our algorithm performs
better with the accuracy of 98.25%, whereas the K-SVM algorithm by Zheng et al.
(2014) reported 97.38% (overall accuracy) which is around 0.87% improvement, and
even 0.1% improvement is critical in healthcare domain (and particularly in breast
cancer diagnosis).

5 Conclusion

In this study, we proposed an algorithm for solving the classification problem which
internally used multiple linear regression (MLR) and linear programming (LP) to find
the nonlinear boundary between the target classes. We evaluated the performance of
the algorithm in terms of the accuracy, and the proposed algorithm was able to perform
better with the accuracy of 98.25%, while the K-SVM algorithm by Zheng et al.
(2014) reported 97.38% (overall-accuracy) for predicting the breast cancer. Thus,
the current work proposed an algorithm for solving the classification problem using
the OR techniques, which is in comparable with the ML algorithm available in the
literature for solving the classification problem (for predicting the breast cancer).
Future work can be in the direction of exploring the options to partially/completely
use the proposed algorithm in combination with the existing ML algorithms, to study
and improve the effect of the same in various problem domains and the respective
problem instances.

Acknowledgements I thank Professor C. Rajendran, Indian Institute of Technology Madras, India,

for his input and support. I am thankful to the reviewers and the editors for their valuable comments
and suggestions to improve the paper.

References

Akay, M. F. (2009). Support vector machines combined with feature selection for breast cancer
diagnosis. Expert Systems with Applications, 36(2), 3240–3247.
A Classification Algorithm Based on Linear Regression … 157

Bennett, K. P., & Blue, J. A. (1998). A support vector machine approach to decision trees. In 1998
IEEE International Joint Conference on Neural Networks Proceedings. IEEE World Congress
on Computational Intelligence (Cat. No. 98CH36227) (Vol. 3, pp. 2396–2401). IEEE.
Elmore, J. G., Wells, C. K., Lee, C. H., Howard, D. H., & Feinstein, A. R. (1994). Variability
in radiologists’ interpretations of mammograms. New England Journal of Medicine, 331(22),
1493–1499.
Huang, C. L., Liao, H. C., & Chen, M. C. (2008). Prediction model building and feature selection
with support vector machines in breast cancer diagnosis. Expert Systems with Applications, 34(1),
578–587.
Madankumar, S., Navya, P., Rajendran, C., Gupta, N. S., & Valarmathi, B. (2017). Linear program-
ming (LP)-based two-phase classifier for solving a classification problem with multiple objectives.
In Big Data Analytics Using Multiple Criteria Decision-Making Models (pp. 87–112). CRC Press.
O’brien, R. M. (2007). A caution regarding rules of thumb for variance inflation factors. Quality
and Quantity, 41(5), 673–690.
Prasad, Y., Biswas, K. K., & Jain, C. K. (2010). SVM classifier based feature selection using GA,
ACO and PSO for siRNA design. In International conference in swarm intelligence (pp. 307–314).
Springer.
Pena-Reyes, C. A., & Sipper, M. (1999). A fuzzy-genetic approach to breast cancer diagnosis.
Artificial Intelligence in Medicine, 17, 131–155.
Siegel, R., Naishadham, D., & Jemal, A. (2012). Cancer statistics, 2012. CA: A Cancer Journal for
Clinicians, 62(1), 10–29.
Wolberg, W. H., Street, W. N., & Mangasarian, O. L. (1995). Image analysis and machine learning
applied to breast cancer diagnosis and prognosis. Analytical and Quantitative Cytology and
Histology, 17, 77–87.
Zheng, B., Yoon, S. W., & Lam, S. S. (2014). Breast cancer diagnosis based on feature extrac-
tion using a hybrid of K-means and support vector machine algorithms. Expert Systems with
Applications, 41(4), 1476–1482.
Predicting Sustainable Supply Chain
Performance Based on GRI Metrics
and Multilayer Perceptron Neural
Networks

Devendra Singh, Krishnanand Lanka, and P. R. C. Gopal

Abstract The sustainable supply chain performance prediction model anticipates

the values of key performance indicators also known lagging metric, based on the
leading metrics. The prediction system helps decision-makers to identify the perfor-
mance gaps and helps to take necessary action plans to minimize the deviation
between the targets set and the outputs that are estimated by the model. This study
uses a worldwide accepted Global Reporting Initiative (GRI) metrics and evalu-
ates values of level 1 metrics that are effects of level 2 metrics by artificial neural
networks(ANN). The literature presents several sustainable supply chain perfor-
mance evaluation models; however, a prediction model based on a combination
of GRI and ANN is a fairly unexplored area. Multilayer perceptron neural network
model has the ability to adjust with the environment of use with the help of past
performance data unlike models present in the literature that require manual param-
eterization for updating and implementing them. MATLAB is used for the compu-
tational implementation of ANN models. From the results of Pearson correlation
coefficient, the correlation obtained among the targeted and forecasted performance
values for the GRI level 1 metrics is positive.

Keywords Sustainable supply chain management · GRI · Multilayer perceptron

D. Singh (B) · K. Lanka

Mechanical Engineering Department, National Institute of Technology Warangal, Warangal,
Telangana 506004, India
e-mail: [email protected]
K. Lanka
e-mail: [email protected]
P. R. C. Gopal
Department of School of Management, National Institute of Technology Warangal, Warangal,
Telangana 506004, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 159
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_8
160 D. Singh et al.

1 Introduction

Sustainable supply chain management (SSCM) is defined as the material and infor-
mation flow management and coordination between firms across the supply chain
while considering goals from all three dimensions of sustainable development, i.e.,
economic, environmental and social (Erol et al., 2011). Sustainability has become
the mainstream of the business world today (Gopal & Thakkar, 2015).
The combination of economic, social and environmental performances of a prac-
tice is defined as its sustainable performance (Bauman & Genoulaz, 2014). Sustain-
able supply chain performance (SSCP) assessment deals with the practices taking
place throughout the supply chain during operations of an organization and reflects
how well it is doing economically, environmentally and socially.
Firms started to move toward sustainable development because of increased envi-
ronmental concerns and labor exploitation. A sustainable development, by defini-
tion, is ‘a development that meets the needs of the present without compromising
the ability of future generations to meet their own needs’ (The World Commission
on Environment and Development, 1987). For sustainable development, Elkington
introduced the concept of the triple bottom line, TBL in 1994 (Elkington, 2004) that
consists of three dimensions which are economy, environment and society. These are
also known as 3Ps of TBL meaning profit, people, and planet.
Profit: The traditional measure of corporate profit.
People: Measures how socially responsible an organization has been throughout
its operations.
Planet: Measures how environmentally responsible a firm has been.
In TBL, the profits of an organization do matter but just not at the expense of
social and environmental concerns. Ignoring TBL can lead to the destruction of
the rainforest, exploitation of labor, damage to the ozone layer and other negative
impacts.
Profit of an organization is a monetary term that means it can be calculated,
whereas social and environmental responsibilities of an organization are subjective
terms and are not quantities. So Global Reporting Initiative, an international indepen-
dent standards organization founded in 1997, introduced GRI standards in 2000 to
measure sustainability. GRI standards quantified the environmental effects as GHG
emissions, waste disposed of, etc., and social effects as accidents reported in the
firm, training hours provided, etc. GRI standards are a systematic set of standards
of economic, environmental and social aspects. A GRI model can be constructed by
arranging these standards in a hierarchical manner into two levels: level 1 metrics
and level 2 metrics. Level 1 metrics can be termed as lagging metrics, and level 2
metrics are also known as leading metrics. In organizations, the term, key perfor-
mance indicators (KPIs), is used instead of level 1 metrics. This GRI model recom-
mends to analyze the relationship of cause and effect between the different levels of
the metrics as a way of trying to find underlying causes for bad performance. For
example, performance in respect of economic value distributed, a metric of level 1,
is an outcome of multifactor performance such as payments to providers of capital,
Predicting Sustainable Supply Chain Performance Based … 161

operating cost, payments to government by country and employee wages and bene-
fits. Therefore, the performance of level 1 metrics can be evaluated by using level 2
metrics.
This study proposes the idea to build a model that can estimate the cause and
effect relationship accurately among the different levels of metrics. This model can
be used for the prediction of lagging metrics (output variable) with the help of leading
metrics (input variable). This prediction model will help the decision-makers in the
organization to compare the targets they have set and values predicted by the model.
The deviation, if any, between the targets and predicted outcomes can be analyzed,
and action plans can be taken to minimize or eliminate it to achieve desired targets
by changing their inputs.
To estimate the causal relationship, the artificial neural network has been used as
it has the ability to adjust with a specific environment of use with the help of past
performance data. ANN has the advantage of modeling linear as well as nonlinear
relationships.
The literature presents various sustainable supply chain performance measure-
ment models (Ding, 2005; Erol et al., 2011, Buyukozkan & Çifçi, 2012, Uysal, 2012,
Govindan et al., 2013, Jakhar, 2015, Marioka & Carvahlo, 2016) based upon different
multi-criteria decision-making techniques, but the area of sustainable supply chain
performance prediction system is still untouched. The performance measurement
systems proposed in the literature reviewed have a drawback that corresponds to
the necessity of collecting the specialists’ judgments for parameterizing the decision
rules and linguistic terms of both the variables: input and output. One of the limita-
tions of these systems is that they are unable to review and adjust the relationship by
using past performance data between metrics. Secondly, parameterizing the system
with the help of judgments from specialists consumes a lot of time, whereas, these
limitations of the system that predict performance can be overcome by using it with
artificial neural networks.
So, in this study, a new prediction system for sustainable supply chain performance
has been proposed which estimates the lagging metrics values of the GRI model based
on leading metrics values by using artificial neural networks (as illustrated in Fig. 4).

2 Sustainable Supply chain Assessment and Management

Information, material and capital flow link suppliers, focal companies and customers
in a supply chain. In accordance with the product’s value, the burden of the environ-
ment and society comes resulting from various production stages. In light of this,
the supply chain’s focal companies might be held responsible for their suppliers’
environmental and social performance (Seuring & Muller, 2008). Improvement in
a supply chain’s service and goods quality can considerably be achieved by perfor-
mance management of the organization and its suppliers (Sweeney, 2011; SCC,
2012).
162 D. Singh et al.

There are reasons given by Lima Jr and Carpinetti (2019), based on Parker (2000),
why organizations should measure performance: “(1) for success identification, (2)
to check whether customer requirements are met, (3) to help them understand their
processes, (4) to identify where problems bottlenecks, waste, etc., exist and where
improvements are necessary, (5) to make sure decisions are fact-based and not
supposition, emotion, faith or intuition-based, (6) to show whether planned improve-
ments actually happened.” Performance measurement is an important aid for making
judgments and decisions (Parker, 2000).

2.1 GRI Model

GRI, acronyms for Global Reporting Initiative, is a set of standards developed in

the year 2000 by Global Reporting Initiative, an international standards organization
formed in 1997, to help businesses, governments and other organizations grasp and
convey their influence on issues like human rights, corruption and climate change.
The GRI framework enables organizations to identify, gather and report their envi-
ronmental, economic and social impacts. Annual sustainability reports are published
by many firms in order to be transparent about the effects of their processes on the
environment, society and economy. Reporting frameworks of GRI launched in 2018
are the most recent GRI standards.
These GRI standards have a modular and interrelated structure (GRI, 1997). In
light of this, these GRI standards have been arranged in a hierarchical structure into
two levels (level 1 and level 2) having a causal relationship to create a GRI model
for all three attributes of sustainability: economic, environmental and social. That
means for each attribute there is a set of level 1 metrics, and corresponding to each
level 1 metric, there are several level 2 metrics (Fig. 1).
Level 1 metrics are used as outputs, and level 2 metrics are used as inputs. The
GRI model involves both linear as well as nonlinear cause and effect relationships.
Therefore, the artificial neural network (ANN) has been used as a technique to esti-
mate this relationship so that the whole arrangement (GRI model and ANN) can be
used as a performance prediction system. The performance regarding level 1 metrics
can be achieved by using level 2 metrics as inputs. Studies reviewed in the literature
have not used GRI standards so far and that too with ANN technique so it remains a
new idea. ANN can obtain a relationship between cause and effect in any environ-
ment with the help of past performance data which is also known as the ability to
adjust to a particular environment of use.

2.2 Multilayer perceptron Neural Networks

Artificial neural networks (ANNs) are models of artificial intelligence (AI) inspired
by structure of the human brain and its learning skills. From literature, it is evident
Predicting Sustainable Supply Chain Performance Based … 163

Fig. 1 GRI model

that wide applications of ANNs in sustainable supply chain management are present.
Out of various ANN models, majorly multilayer perceptron (MLP) neural networks
are used because of its ability to deal with time forecasting and function approxi-
mation problems accurately. MLP networks have the ability to adjust to the required
environment with the help of previous data of performance by using a supervised
learning algorithm that utilizes input and output variables as a set of samples. In the
majority of applications, backpropagation algorithm is used.
Synaptic connections by interconnecting a set of artificial neurons form an MLP
network. It consists of input variables in an input layer, output variables in an output
layer and hidden neurons in one or more hidden layers. For each input variable, there
is a neuron in the input layer, and similarly each output variable there is a neuron in
the output layer. Data characteristics are extracted and stored by neurons in hidden
layer. Also, most of the operations of data processing are done by the hidden layer.
Figure 2 shows a general structure of an MLP network. In total, seven elements
together compose an artificial neuron: inputs, weights, activation threshold, linear
combiner, activation function, activation potential and outputs. The learning of a
perceptron is a process of iterations. The learning starts by assigning each weight a
random value. It then calculates the error function by comparing the corresponding
outputs with expected values and modifies the weights to minimize the error function.
As the training stage completes, an optimal weight matrix is so obtained that the
mean square error (MSE) among the values that are expected and the outputs that are
estimated by the network is minimized. Obtaining this optimal weight matrix is the
main objective of training. Learning rate, number of hidden layer neurons, hidden
layer count and activation function applied in output and hidden layers are some of
the parameters that influence the prediction accuracy of the models directly, and they
have to be chosen empirically. Studies (Lenard et al., 1995) suggest to use either
three-fourths of input variables or half of the combined output and input variables as
the count of neurons in hidden layer. Epochs count which is defined as the number
164 D. Singh et al.

Fig. 2 Multilayer perceptron neural networks

of times a network computes the subset of training samples is considered as both

efficiency parameter as well as network training stopping criteria.
An MLP network with a single hidden layer has the ability to estimate both linear
and nonlinear relationships, and also it can adjust to the required environment of
use by the help of past performance data. Because of its abilities, an integration of
GRI metrics with MLP neural networks can contribute to the literature as well as
to develop ANN systems that can help practitioners to enhance sustainable supply
chain performance.

3 Sustainable Supply Chain performance Prediction

System

A system for predicting performance allows the approximation of the results of

lagging metrics (output variables) with the help of leading metrics’ (input variables)
performance levels. The levels of performance regarding leading metrics are set
either by actual performance data or by estimation. Systems for prediction are based
majorly on the technique of artificial intelligence which estimates the cause and effect
relationship among metrics by using learning algorithms.
The prediction systems help managers to create systematic ways of predicting the
performance of various metrics and identifying issues in the company. Unahabhokha
et al. (2007) developed a performance prediction system and applied it in an industrial
printer manufacturer and in a textile company. Prediction systems can also support
decision-making in the context of the supply chain (Lima Jr & Carpinetti, 2019).
Though the literature available on performance prediction systems for the supply
Predicting Sustainable Supply Chain Performance Based … 165

Fig. 3 Sustainable supply chain performance prediction system

chain is less, the area of performance prediction systems for the sustainable supply
chain is yet to be discovered. The performance prediction system proposed in this
study is a combination of GRI metrics and ANN (Fig. 3). Based on past data of
performance, it can adjust to a particular environment of use. It also does not require
any judgments from the specialists for parameterization of various decision rules
unlike multi-criteria decision-making techniques (Lima Jr & Carpinetti, 2019).
Based on Lima Jr and Carpinetti (2019), a theoretical framework was developed
for performance prediction. The whole idea is to estimate the deviation between
predicted lagging metrics values and the target values; if any deviation is present,
then development and execution of action plans can be carried out in advance to
prevent or mitigate deviation from desired outcomes.
Figure 4 illustrates the proposed system for predicting performance which consists
of ten models of MLP neural network implemented on the hierarchically arranged
GRI standards into two levels. The output of ANN model 1 is used as an input to
ANN model 2 and also by studying the mathematical expressions present in the GRI
model to obtain level 1 metrics, it can be realized that nonlinear relationship among
input and output variables is present only in ANN model 3. It must be noted that the
whole idea proposed in this study was applied to the Tata Motors to put forward an
example (Sects. 3.1 and 3.2). The GRI reports of Tata Motors from 2010 to 2018
were analyzed, and only those metrics were extracted out which could be arranged
in two hierarchical levels. Hence, the number of parameters is different for the three
dimensions of sustainability.
The process of performance management presented in Fig. 3 can be carried out by
firstly selecting the set of GRI metrics that are in line with the company’s competitive
strategy and performance goals. Secondly, for each selected metric data must be
166 D. Singh et al.

Fig. 4 Sustainable supply chain performance prediction system

collected by the focal company. That means, the current values of input variables
(level 2 metrics) must be collected or estimated, and also performance targets must be
set for chosen metrics (level 1) for performing performance prediction tests. Now, the
normalization of collected values should be done by taking the universe of discourse
limits and Eq. 1.

X i − X min
xi = (1)
X max − X min

After normalizing the metrics values of level 2, the values are supplied to the
system for prediction, and by using them the trained ANN models measure the
expected level 1 metrics performance values. These measured values are obtained as
normalized values so they have to be converted back to the original form. This can
be done by taking the universe of discourse limits defined during the training stage
for each output metric. The targeted performance values can be compared with these
estimated performance outputs, and hence managers can analyze the deviation if any,
and therefore can identify the areas of gaps to put efforts on to ultimately minimize
or avoid this deviation.
Predicting Sustainable Supply Chain Performance Based … 167

For the training purpose of ANN models, the numerical values of metrics of both
the levels must be gathered by the focal company so that the prediction system can be
made adjustable to a particular environment of use. This data should be arranged as
input and output variables in a table for each ANN model. For each ANN model, a set
of networks depending upon various parameters (hidden and output layer activation
functions, neurons count in hidden layer) should be characterized as candidates. The
learning rate and training epochs count must also be selected. The total samples
should be divided into two sets: 70% of them are used for training purposes and 30%
for testing as well as validating purposes. For every candidate network, the accuracy
level is obtained by analyzing the mean square error (MSE) between the validation
subset and system estimated output values. The network with the least MSE should
be selected as the best network among all the candidate networks for respective ANN
model. In case the input values of variables lie outside the universe of discourse limits
defined earlier during the process of training, then the ANN models need updating
of new limits and hence retraining.

3.1 Illustrative Application

To evaluate the proposed system’s prediction accuracy along with the process of
grabbing the best candidate network for each ANN model, the development of an
illustrative application case was done. The GRI metrics adopted for the proposed
system are shown in Table 1 with measurement units. For each ANN model, six
candidate networks were defined by varying various parameters as shown in Table 2.
Neurons count in hidden layer was defined as per the quantity of variables used for
input (x n ) for every ANN model. Thus, x n , x n – 1, x n + 1 are the number of neurons
for which candidate networks were tested. A single hidden layer is sufficient to
map all linear as well as nonlinear relationships (Haykin, 2010; Silva et al., 2017).
Hyperbolic tangent and logistic in the hidden layer neurons are the two nonlinear
activation functions that were checked. Both of these functions are widely used
in ANNs. The learning procedure of neurons is different for different activation
functions. Hence, the trained networks will provide different accuracies for different
activation functions. As there is only one output in each ANN model, therefore a
single neuron is used in the output layer. The adoption of activation function as a
linear function was done for output layer neuron so that the output signals from the
neurons of hidden layer form a linear combination and it becomes easy to compute.
Therefore, based upon all these parameters (Table 2), 60 candidate networks (six for
each ANN model) were created. For validation stage, maximum value of 10E−03
was defined for MSE. Using MATLAB, these 60 candidate networks in total were
implemented, trained and validated. After training, the candidate network with least
MSE at validation stage out of six candidate networks for each ANN model was
selected as the best network for ANN model, hence for performance prediction.
168 D. Singh et al.

Table 1 Metric definitions

Model Variable Brief Description Measurement unit Universe of
discourse
ANN Model 1 x1 Operating cost: An INR Billion [350, 600]
organization can
measure the operational
costs as payments for
purchased goods,
product parts, facilities
and services rendered
outside the organization
x2 Employee wages and INR Billion [20, 50]
benefits: An
organization can
measure it as overall
payroll (which includes
salaries of employees
and amounts paid on
behalf of the workers to
government entities)
plus net benefits (which
excludes training,
safety gear expenses or
other things specifically
relevant to the job
function of the
employee)
x3 Payments to providers INR Billion [13, 35]
of capital: A company
can measure the
payments made to
capital providers as
dividends to all
shareholders, plus
interest fees paid to
loan providers
x4 Payments to INR Billion [30, 60]
government by country:
A company can
measure government
payments as all taxes of
the company plus
related sanctions paid at
the international,
national and local level.
Company taxes can
include corporate,
income and property
(continued)
Predicting Sustainable Supply Chain Performance Based … 169

Table 1 (continued)
Model Variable Brief Description Measurement unit Universe of
discourse
y1 Economic value INR Billion [413, 745]
distributed: It is the sum
of payments to capital
providers, operating
costs, benefits and
wages of employee,
community investments
and payments to
government by country
ANN Model 2 x5 Economic value INR Billion [400, 600]
generated: A company
can measure economic
value generated as total
sales plus revenues
from asset sales and
investments in
financing
x6 Economic value INR Billion [418.3056403,
distributed: It is the sum 736.8343429]
of payments to capital
providers, operating
costs, benefits and
wages of employee,
community investments
and payments to
government by country
y2 Economic value INR Billion [−283.13384,
retained: Direct 143.86489]
economic value
generated is the sum of
economic value
distributed and
economic value
retained
ANN model 3 x7 Direct energy GJ [900,000,
generated: Energy 3,000,000]
generated inside the
organization
x8 Indirect energy GJ [1,200,000,
generated: Energy 2,500,000]
purchased from outside
the organization
(continued)
170 D. Singh et al.

Table 1 (continued)
Model Variable Brief Description Measurement unit Universe of
discourse
x9 Renewable energy: GJ [110,000,
Renewable energy is 500,000]
the energy generated
from renewable sources
such as hydro, biomass,
wind, geothermal, solar
y3 % renewable energy of GJ [2.17277119,
total energy: How much 18.08650957]
part of total energy
consists of renewable
energy
ANN model 4 x 10 Scope 1: The tCO2 e [50,000,
greenhouse gas (GHG) 250,000]
emissions produced by
the consumption of
non-renewable fuels are
known as direct (Scope
1) GHG emissions
x 11 Scope 2: The use of tCO2 e [250,000,
purchased electricity, 500,000]
heating, cooling and
steam adds to the
indirect energy (Scope
2) emissions from the
company
y4 GHG Emissions: Total tCO2 e [300,000,
GHG emissions 750,000]
produced by the
organization, i.e., both
direct and indirect
ANN model 5 x 12 SOx : Oxides of sulfur Ton [50, 650]
resulted from the
manufacturing
processes of the
organization
x 13 NOx : Oxides of Ton [90, 250]
nitrogen resulted from
the manufacturing
processes of the
organization
x 14 SPM: Particulate matter Ton [120, 350]
is the sum of all solid
and liquid particles
suspended in air many
of which are hazardous
(continued)
Predicting Sustainable Supply Chain Performance Based … 171

Table 1 (continued)
Model Variable Brief Description Measurement unit Universe of
discourse
y5 Air emissions: Ton [260, 1250]
Manufacturing
processes contribute to
air emissions, namely
oxides of nitrogen
(NOx), sulfur dioxide
(SO2 ) and particulate
matter
ANN model 6 x 15 Municipal: Water m3 [2,000,000,
supplied from 8,500,000]
municipal (third party)
suppliers
x 16 Surface water: Water m3 [750,000,
that is naturally found 1,600,000]
on the surface of the
earth in icebergs, rivers,
bogs, ice caps, glaciers,
ice sheets, ponds,
streams and lakes
x 17 Groundwater: Water m3 [650,000,
that is held in an 1,800,000]
underground formation
and that can be
recovered from there
x 18 Rainwater: Water m3 [40,000,
harvested from rain and 650,000]
is used
y6 Water abstracted: The m3 [3,440,000,
total water used for 12,550,000]
operations in the
organization
ANN model 7 x 19 Hazardous waste Ton [5000, 7500]
generated: Hazardous
waste is waste that has
substantial or potential
threats to public health
or the environment
x 20 Non-hazardous waste Ton [70,000,
generated: All waste 150,000]
materials not
specifically deemed
hazardous under federal
law are considered
nonhazardous wastes
(continued)
172 D. Singh et al.

Table 1 (continued)
Model Variable Brief Description Measurement unit Universe of
discourse
y7 Waste generation: Total Ton [75,000,
waste generated in the 157,500]
organization (hazardous
and non-hazardous)
ANN model 8 x 21 Hazardous waste Ton [4500, 10,000]
disposed of: Complete
weight of hazardous
waste includes reuse,
incineration, deep well
disposal, recycle,
composting, landfill,
onsite storage, recovery,
etc.
x 22 Non-hazardous waste Ton [45,000,
disposed of: Complete 150,000]
weight of
non-hazardous waste
includes reuse,
incineration, deep well
disposal, recycle,
composting, landfill,
onsite storage, recovery,
etc.
y8 Total waste disposed of: Ton [49,500,
Total weight of waste 160,000]
disposed of (hazardous
and non-hazardous)
ANN model 9 x 23 Male hires: Total Dimensionless [900, 1500]
number of male
employees hired in the
organization
x 24 Female hires: Total Dimensionless [65, 160]
number of female
employees hired in the
organization
y9 Total hires: Total Dimensionless [965, 1660]
number of employees
hired (male and female)
in the organization
ANN model 10 x 25 Male attrition: Total Dimensionless [1500, 3000]
number of male
employees laid off in
the organization
x 26 Female attrition: Total Dimensionless [100, 180]
number of female
employees laid off in
the organization
(continued)
Predicting Sustainable Supply Chain Performance Based … 173

Table 1 (continued)
Model Variable Brief Description Measurement unit Universe of
discourse
y10 Total attrition: Total Dimensionless [1600, 3180]
number of employees
(male and female) laid
off in the organization

Table 2 Candidate networks

Network Hidden neurons Activation function (Hidden Activation function (Output
neuron) neuron)
Network 1 xn − 1 Logistic function Linear function
Network 2 xn Logistic function Linear function
Network 3 xn + 1 Logistic function Linear function
Network 4 xn − 1 Hyperbolic tangent function Linear function
Network 5 xn Hyperbolic tangent function Linear function
Network 6 xn + 1 Hyperbolic tangent function Linear function

3.2 Data Generation Process

The following processes were done for sample generation to train and validate the
models:
• By taking into consideration the universe of discourse, the random values were
generated for level 2 metrics using MS Excel.
• With the help of equations presented in the GRI model, the values of level 1
metrics were obtained by using level 2 metrics.
• Normalization of all the input and output values was done using Eq. 1.
As the random values generated between the given universe of discourse do not
include both extreme values, we have separately added a set of lower and a set of
upper bound values in each of the respective data sets.
For data collection, the annual GRI reports of Tata Motors from 2010 to 2018
(nine data sets) were analyzed, and metrics were extracted out and were arranged
in two levels. But nine data sets would not have been sufficient for ANNs as ANNs
require big data sets for training. So, by using MS excel, data for all the metrics was
generated within the range we defined by analyzing the GRI reports.
174 D. Singh et al.

3.3 Training and Validation Results

During the training and validation stage, the learning rate value was set to 0.03
initially according to Bilgehan (2011) and training epochs were set to 8000.
The values of MSE obtained during the training procedure of all the candidate
networks are presented in Table 3. During the training stage, the cause and effect
relationships get tuned among input and output metrics and after that validation takes
place. The MSE values generated during the validation stage of all the candidate
networks are also presented by Table 3. The candidate network with the lowest
MSE at the validation stage is considered as the network having the best prediction
accuracy for each model (Table 4). The best candidate networks are highlighted in
bold as shown in Table 3.
Pearson correlation coefficient (R) indicates the relationship degree among sample
values and output values at the validation stage. For selected candidate networks
(having least MSE), values of R were also obtained for evaluating prediction accuracy
in addition to MSE at the validation stage. R varies in [−1, 1] interval with −
1 representing a correlation which is perfectly negative, 1 representing a perfectly
positive correlation, and 0 representing correlation does not exist among the variables
(Montgomery & Runger, 2012). The pattern of association can be obtained among
the two variables as an equation. So in order to do that, linear regression tests were
also done using MATLAB. Results obtained for regression analysis and correlation
coefficient are shown in Fig. 5. The targeted performance values are referred by the
horizontal axis, and the ANN model estimated values are referred by the vertical
axis.

4 Conclusion

A prediction model of supply chain performance composed of MLP networks and

GRI metrics is proposed in this study. Sixty candidate networks, in total, were imple-
mented, and all the candidate networks were evaluated depending upon the validation
stage prediction accuracy. By analyzing the results, Fig. 5 and Tables 3 and 4, it can
be concluded that the performance of level 1 metrics can be estimated with high
accuracy by using MLP networks.
The validation stage MSE was recorded the lowest (8.0238E−07) for ANN model
10 and the highest (2.8760E−05) for ANN model 2. The R values (Fig. 5) indicate
that between predicted and expected values of the level 1 metrics, the correlation is
highly positive for every ANN model. The results so obtained show that use of GRI
metrics and MLP networks for prediction purpose of a sustainable supply chain‘s
performance is an effective method.
Through a prospective evaluation of sustainable supply chain efficiency, the
system proposed in the study enhances rational decision-making. The managers can
use it to acquire a prospective scenario of the performance of business by comparing
Predicting Sustainable Supply Chain Performance Based … 175

Table 3 MSE obtained in

ANN model Network MSE
training and validation stage
Training stage Validation stage
ANN model 1 Network 1 1.157E−04 1.148E−04
Network 2 1.797E−05 2.405E−05
Network 3 2.929E−05 3.741E−05
Network 4 1.892E−04 1.007E−04
Network 5 3.377E−05 1.078E−04
Network 6 4.453E−05 3.95E−05
ANN model 2 Network 7 7.225E−04 1.398E−03
Network 8 9.22E−05 1.029E−04
Network 9 7.101E−04 9.958E−04
Network 10 8.158E−04 1.969E−04
Network 11 9.379E−05 2.876E−05
Network 12 1.201E−04 1.028E−04
ANN model 3 Network 13 3.296E−05 2.7206E−05
Network 14 5.993E−04 7.9323E−04
Network 15 2.834E−04 2.0192E−04
Network 16 7.157E−05 3.589E−05
Network 17 4.973E−04 3.3838E−04
Network 18 3.891E−05 9.0673E−05
ANN model 4 Network 19 1.08E−04 3.653E−05
Network 20 1.589E−05 3.53E−04
Network 21 1.161E−05 1.186E−04
Network 22 1.66E−04 1.0469E−04
Network 23 3.986E−05 1.633E−05
Network 24 1.816E−05 1.692E−05
ANN model 5 Network 25 8.785E−05 7.909E−05
Network 26 2.468E−06 2.972E−06
Network 27 9.825E−06 9.742E−04
Network 28 1.05E−04 7.229E−05
Network 29 1.463E−05 9.055E−05
Network 30 1.989E−04 8.063E−05
ANN Model 6 Network 31 1.533E−04 7.6763E−03
Network 32 9.926E−06 1.549E−04
Network 33 1.264E−04 1.2889E−04
Network 34 7.695E−06 4.3953E−06
Network 35 1.949E−05 4.2195E−05
Network 36 2.673E−06 1.8737E−05
(continued)
176 D. Singh et al.

Table 3 (continued)
ANN model Network MSE
Training stage Validation stage
ANN model 7 Network 37 1.314E−04 1.0625E−04
Network 38 1.15E−05 4.2342E−05
Network 39 1.487E−04 8.237E−05
Network 40 1.768E−04 2.1278E−04
Network 41 1.856E−05 2.8435E−05
Network 42 2.028E−05 1.299E−05
ANN model 8 Network 43 1.308E−04 1.2793E−04
Network 44 1.357E−05 1.0609E−05
Network 45 1.203E−05 7.3038E−06
Network 46 1.621E−04 1.2607E−04
Network 47 1.983E−04 2.1786E−04
Network 48 1.458E−04 1.5091E−04
ANN model 9 Network 49 1.199E−04 7.9029E−05
Network 50 2.572E−05 1.7218E−05
Network 51 1.778E−06 1.1025E−06
Network 52 1.661E−04 6.5288E−05
Network 53 1.43E−04 1.0528E−04
Network 54 2.056E−04 2.4005E−04
ANN model 10 Network 55 1.131E−04 5.4813E−05
Network 56 1.113E−06 8.0238E−07
Network 57 7.477E−05 7.9082E−05
Network 58 1.591E−04 1.7954E−04
Network 59 3.21E−06 5.3957E−06
Network 60 1.764E−05 6.4184E−05

the predicted performance values with defined targets for all the level 1 metrics.
This comparison will enable managers to get a better insight of the ramifications of
underperformance or over performance in various operational facets. For example,
managers can assess environment dimension performance based on calculations of
the “air emissions” (level 1 metric) as a result of the model level 2 metrics (input
variables). In this way, the wide range of the GRI metrics leads to the identification
of critical areas of business in order to put preventive action plans at the input stage
for improvements.
Apart from all its advantages, this study lacks whenever a metric is removed or
added because in that case the whole model needs retraining. Another limitation is
that as ANNs are environment specific so as the work environment changes the input
values also change and hence the model will require retraining. ANNs produce highly
accurate results for complex data; however, the training time of models increases with
Predicting Sustainable Supply Chain Performance Based … 177

Table 4 Network candidate selected for every ANN model in the prediction systems
ANN Chosen MSE R (Validation Number Activation Learning
model network (Validation stage) of hidden function of the rate
stage) layer hidden layer
neuron
ANN Network 2.4050E−05 0.99988 4 Logistic Function 0.03
model 2
1
ANN Network 2.8760E−05 0.99981 2 Hyperbolic 0.03
model 11 tangent
2
ANN Network 2.7206E−05 0.99974 2 Logistic Function 0.03
model 13
3
ANN Network 1.6330E−05 0.99980 2 Hyperbolic 0.03
model 23 tangent
4
ANN Network 2.9720E−06 0.99997 3 Logistic function 0.03
model 26
5
ANN Network 4.3953E−06 0.99995 3 Hyperbolic 0.03
model 34 tangent
6
ANN Network 1.2990E−05 0.99993 3 Hyperbolic 0.03
model 42 tangent
7
ANN Network 7.3038E−06 0.99997 3 Logistic function 0.03
model 45
8
ANN Network 1.1025E−06 1 3 Logistic function 0.03
model 51
9
ANN Network 8.0238E−07 1 2 Logistic function 0.03
model 56
10

high complexity of the data. This model helps to original equipment manufacturer
(OEMs) to assess the sustainable performance metrics such as percentage renewable
energy used of total energy (GJ), total water abstraction (m3 ), GHG emissions (tCO2 ),
air emissions (Ton), waste generation (Ton) and waste disposed (Ton) effectively
which are useful to make internal and internal benchmarking. Further, it also helps
to initiate policy decisions on improvement of sustainability performance.
Further studies can use different artificial intelligencey techniques and algorithms
to develop comparative studies in this area and compare the advantages and disad-
vantages of the techniques. Also the idea of performance prediction model can be
implemented to agile, flexible or lean supply chain.
178 D. Singh et al.

Fig. 5 Regression analysis and correlation coefficient

Predicting Sustainable Supply Chain Performance Based … 179

Acknowledgements I would like to thank Francisco Rodrigues Lima Junior (Assistant Teacher at
Federal University of Technology, Brazil) for his assistance and support. I am grateful that the annual
GRI reports of Tata Motors for collection of data were available on Global Reporting Initiative.
Finally, I must express my very profound gratitude to my parents and to my friends for providing
me with unfailing support and continuous encouragement throughout the process of researching
and writing this paper.

References

Acquaye, A., Ibn-Mohammad, T., Genovese, A., et al. (2018). A quantitative model for environ-
mentally sustainable supply chain performance measurement. 49, 63–66
Agami, N., Saleh, M., & Rasmy, M. (2014). An innovative fuzzy logic based approach for supply
chain performance management. IEEE Systems Journal, 8, 336–342.
Bilgehan, M. (2011). Comparison of ANFIS and NN models—with a study in critical buckling load
estimation. Applied Soft Computing, 11, 3779–3791.
Buyukozkan, G., & Çifçi, G. (2012). A novel hybrid MCDM approach based on fuzzy DEMATEL,
fuzzy ANP and fuzzy TOPSIS to evaluate green suppliers. Expert Systems with Applications, 39,
3000–3011.
Chardine-Bauman, E., & Botta-Genoulaz, V. (2014). A framework for sustainable performance
assessment of supply chain management practices. Computers and Industrial Engineering, 76,
138–147.
Chen, L., Feldmann, A., & Tang, O. (2015). The relationship between disclosures of corporate
social performance and financial performance: Evidences from GRI reports in manufacturing
industry. International Journal of Production Economics, 170, 445–456.
Ding, G. K. (2005). Developing a multicriteria approach for the measurement of sustainable
performance. Building Research and Information, 33, 3–16.
Erol, I., Sencer, S., & Sari, R. (2011). A new fuzzy multi-criteria framework for measuring
sustainability performance of a supply chain. Ecological Economics, 70, 1088–1100.
Ganga, G. M. D., & Carpenitti, L. C. R. (2011). A fuzzy logic approach to supply chain performance
management. International Journal of Production Economics, 134, 177–187.
Global Reporting Initiative. (1997). https://www.globalreporting.org/Pages/default.aspx. Accessed
15 September 2019.
Gopal, P. R. C., & Thakkar, J. (2015). Development of composite sustainable supply chain perfor-
mance index for the automobile industry. International Journal of Sustainable Engineering, 8,
366–385.
Govindan, K., Khodaverdi, R., & Jafarian, A. (2013). A fuzzy multi criteria approach for measuring
sustainability performance of a supplier based on triple bottom line approach. Journal of Cleaner
Production, 47, 345–354.
Haykin, S. (2010). Neural networks and learning machines (3rd ed.). Pearson Education India.
Jakhar, S. K. (2015). Performance evaluation and a flow allocation decision model for a sustainable
supply chain of an apparel industry. Journal of Cleaner Production, 87, 391–413.
Elkington, J. (2004). https://www.johnelkington.com/archive/TBL-elkington-chapter.pdf. Enter the
triple bottom line.
Lenard, M. J., Alam, P., & Madey, G. R. (1995). The application of neural networks and a qualitative
response model to the auditor’s going concern uncertainty decision. Decision Sciences, 26, 209–
227.
Li, Y., & Mathiyazhagan, K. (2017). Application of DEMATEL approach to identify the influential
indicators towards sustainable supply chain adoption in the auto components manufacturing
sector. Journal of Cleaner Production, 172, 2931–2941.
180 D. Singh et al.

Lima-Junior, F. R., & Carpinetti, C. R. (2019). Predicting supply chain performance based on
SCOR® metrics and multilayer perceptron neural networks. International Journal of Production
Economics, 212, 19–38.
Montgomery, D. C., & Runger, G. C. (2012). Applied statistics and probability for engineers, 5th
ed. Willey
Morioka, S. N., & Carvalho, M. M. (2016). Measuring sustainability in practice: Exploring the
inclusion of sustainability into corporate performance systems in Brazilian case studies. Journal
of Cleaner Production, 136, 123–133.
Parker, C. (2000). Performance measurement. Work study.
SCC. (2012). Supply chain council. Supply Chain Operation Reference Model 12.0 (SCOR model).
SCC. (2012). Supply chain council. Supply Chain operation Reference Model. Supply Chain
Council 11.0.
Seuring, S., & Müller, M. (2008). From a literature review to a conceptual framework for sustainable
supply chain management. Journal of Cleaner Production, 16, 1699–1710.
Silva, I. N., Spati, D. H., Flauzino, R. A., Liboni, L. H. B., & Alves, S. F. R. (2017). Artificial neural
networks: A practical course. Springer.
Su, D., Horng, C., Tseng, M., et al. (2016) Improving sustainable supply chain management using
a novel hierarchical grey-DEMATEL approach. Journal of Cleaner Production, 134, 469–481.
Sweeney, E. (2011). Towards a unified definition of supply chain management: The four
fundamentals. International Journal of Applied Logistics, 2, 30–48.
The World Commission on Environment and Development. (1987). https://sustainabledevelopment.
un.org/content/documents/5987our-common-future.pdf. Report of the World Commission on
Environment and Development: Our Common Future.
Unahabhokha, C., Platts, K., & Tan, K. H. (2007). Predictive performance measurement system: A
fuzzy expert system approach. Benchmarking: An International Journal, 14(1), 77–91.
Uysal, F. (2012). An integrated model for sustainable performance measurement in supply chain.
Procedia-Social and Behavioral Sciences, 62, 689–694.
Green Supply Chain Management
in the Indian Petroleum Industry Using
AHP-VIKOR Approaches

Sourabh Kumar and Mukesh Kumar Barua

Abstract Petroleum industry operations have positively and negatively impacted

a range of areas covered by sustainable developments, including communities,
economies, and ecosystems. Therefore, the petroleum industry has redefined its busi-
ness development strategy for achieving the green sustainable goal. This study aims to
identify green supply chain management practices and prioritize these to know their
level of importance in petroleum industries. The literature review was scrutinized,
and the main criteria were identified, which were further filtered by the expert panel.
Two criteria, basic and advanced green practices and the seven subcriteria, were
evolved to assess green practices indices identification. This study employs multicri-
teria decision-making (MCDM) techniques AHP and VIKOR using R programming
to prioritize the green practice indices. The analysis shows that investing in and imple-
menting renewable energy, risk prevention systems to cover possible environmental
accidents, and preventing oil and methane leaks are the top three green practices.
The finding helps practitioners and policymakers of the petroleum companies to
implement effective green practices to sustain the environment.

Keywords Green practices · R- language · Oil and gas · Sustainability

1 Introduction

The supply and demand for crude oil and petroleum products are the key factors deter-
mining the world economy’s status (Hamilton, 2009; Krichene, 2002). Petroleum
companies provide more than 50% of global fuel consumption and are expected to
remain by 2035. On the one hand, global demands for fossil fuels are still rising. On
the other hand, companies face complex investment challenges due to the severe oper-
ating environment of exploration, production, and refining activities (Asif & Muneer,
2007). The 2030 Agenda for sustainable development also requires support from the

S. Kumar (B) · M. K. Barua

Indian Institute of Technology Roorkee, Haridwar, India
M. K. Barua
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 181
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_9
182 S. Kumar and M. K. Barua

oil and gas industries (Yusuf et al., 2013). The petroleum industry is a key pillar of
the global energy system and one of the drivers of economic and social development
(Omer, 2008). Environmental sustainability effects in greater amounts by the release
of waste in the environment. In the exploration and production segment, drilling
starts with an appraisal phase that explores hydrocarbon presence and identifies the
reservoir’s size and nature. The location of the drilling site depends on the charac-
teristics of geological formation and depends on surface constraints. It is because to
balance environmental sustainability with logistical constraints without affecting the
efficiency of drilling. Most of the oil reservoirs have already been developed, and
new reservoirs can harness environment conditions if proper green practices are not
implemented.
The petroleum industry projects are crucial because they directly enhance the
world economy and national GDPs. It plays a big economic impact, so it is also a
key player in politics and strategic decision making in government. The petroleum
industry often accuses of greenhouse gas emission in upstream, midstream, and
downstream segments. As per Shell oil company sustainability report, they fixed the
greenhouse gas intensity such as upstream segment GHD intensity ≤0.164, refining
GHG ≤1.05 tonnes, chemical plant GHG ≤0.97, and reduce flaring in upstream
business <5.2 tones (Shell, 2018). Previous literature shows that companies have to
focus on various aspects to achieve environmental sustainability, such as greenhouse
gas emissions, safety, technical standards, technology development, emission to air,
biodiversity, and ecosystem, and spill to the environment.

1.1 The Motivation and Proposed Contribution

The present study is motivated by the petroleum company’s faces accusations of

environmental degradation. The Indian petroleum industry is a core industry, and
it plays a significant role in influencing the decision making of the all-important
sector of the economy. Indian import 82% of its need, and it is the third largest in
the world. The issue of environmental sustainability is a major concern for govern-
ment and local players. The risk of environmental pollution is seemingly throughout
the petroleum supply chain. Indeed, many companies are committing significant to
advance sustainable environmental development as a matter of business practices.
However, while the current practices are already contributing to green sustainable
development, much more can and must be done. The current literature talks about
the green practices in other areas such as the food industry, automobile industry,
furniture industry, and health care, but no one has touched upon the green practices
adopted in the oil and gas industry (Aliour et al., 2019; Beske et al., 2014; Fernando
et al., 2016; Zhu et al., 2007).
Therefore, it is important to understand the green practices in the petroleum
industry. This leads to the following research questions:
(i) What are the important dimensions of green practices in petroleum industries?
Green Supply Chain Management in the Indian Petroleum Industry … 183

(ii) In which order these practices affect the petroleum industries?

(iii) How can environmental scientists/engineers assess and develop adequate
policies to achieve environmental sustainability?
To address research questions, the study sets the following objectives:
1. To determine the important dimensions of green practices in Indian petroleum
industries
2. To rank these practices in order to know their level of importance in the Indian
petroleum industries.
3. To present the managerial implications for improving environmental sustain-
ability in the Indian petroleum industry.
The list of green practice indices has been extracted from the literature review
and experts’ opinions. After that, multicriteria decision-making (MCDM) techniques
AHP and VIKOR have been applied to determine green practices’ priorities. The
proposed method AHP helps the decision-maker facing a complex problem with
multiple conflicting and subjective criteria. Simultaneously, the VIKOR method is
based on the particular measurement of “closeness” to the ideal solution. The study’s
advantage is that it uses both the methods and compares the result to give common
output. To achieve objective 1, the authors reviewed the green practices literature
and conducted semi-structured interviews with petroleum industry experts. Based
on the expert’s opinions, seven green practices shortlisted under two dimensions:
basic green practices and advanced green practices, as shown in Table 1. The second
objective is accomplished by analysis of data by AHP method and VIKOR method.
These two methods provide a ranking of green practices, and we compare both results

Table 1 Green practices indices in petroleum industries

Si. No Basic green practices References
1 Discloser of environmental awareness Shah and Rivera (2007), (Holditch, 2013),
and policies (DEA) George et al., (2016)
2 Employee training related to (Jung et al., 2001), Antão et al., (2016)
environmental performance (ETP)
3 Proper disposal of environmental (Obiajunwa et al., 2002), Johnston et al.,
pollutants (2016)
(PDP)
4 Risk prevention systems to cover possible (Dea & Flin, 2001), (Mckenzie et al.,
environmental accidents (RPS) 2012), Holditch (2013)
Advanced green practices
5 Regularly estimating cost and those Expert opinion
activities that protecting the environment
(REC)
6 Preventing oil and methane leaks (POM) Expert opinion
7 Investing in and implementing renewable Expert opinion
energy (IIE)
184 S. Kumar and M. K. Barua

and conclude the result. The last objective is accomplished by providing meaningful
managerial implications, so environmental scientist/engineer assesses and develops
adequate policies to achieve environmental sustainability.
We have divided this study into eight sections; after introducing and explaining
our motivation for the proposed contribution, the paper’s rest formation is arranged
as follows. Section 2 describes the related literature review on green practices in
industries. In Sect. 3, the proposed research methodologies AHP and VIKOR are
illustrated. In Sect. 4, a case of the Indian petroleum industry is illustrated. Section 5
explains the results of the study. Section 6 explains the implications of the study. In
Sect. 7, limitations and future work are explained. Finally, according to the finding
of the research, conclusion is presented in Sect. 8.

2 Literature Review

The concept of green sustainability development has entered into the petroleum
industry, and green practices, green economy, sustainable development, environ-
mental responsibility, and social responsibility have become common in the literature
(Ford et al., 2014; Golden et al., 2017; Matos & Hall, 2007). The petroleum industry
top managements are now in favor of digitalization solution implementations and
innovation in operational risk management such as environmental change, air quality,
contamination site management, supply chain management KPI (Kohli & Johson,
2011). However, there is always a difference between what companies declare about
sustainability and what they achieve (Ahmad et al., 2016). For this reason, most
of the studies have been conducted, and conclusive decision making that conveys
sustainability development has gained entrance into the decision-making levels of
petroleum companies, but in reality, the industry is one of the main causes of envi-
ronmental degradation and climate change (Hill, 2001). Organizations such as the
International Petroleum Industry Environment Conservation Association (IPIECA),
the American Petroleum Institute (API), International Association of oil and gas
producers (IOGP) are very active in shaping the structure and sustainable develop-
ment standards in petroleum companies (Schneider et al., 2013). Apart from these
organizations, also encourage their shareholders to provide sustainable development
strategies. The present study list of green practices indices is obtained through litera-
ture review and industry experts, shown in Table 1. The proposed research framework
for identifying green practices in the petroleum industry is shown in Fig. 1.

3 Research Methodology

In the present research, we used two multicriteria decision-making techniques AHP

and VIKOR. AHP is a powerful and simple method for the decision-making process.
AHP uses the hierarchical structure of a problem and uses multiple inputs from
Green Supply Chain Management in the Indian Petroleum Industry … 185

Fig. 1 A proposed research framework for identifying green practices in the petroleum industry

experts of the industry to consolidate outcomes. It is mostly used to find out the
highest priority variable among the set of the number of variables. On the other hand,
we also analyze our research with the VIKOR method as it provides an advantage
in a ranking procedure for positive and negative attributes. In the VIKOR method, it
is assumed that experts are not biased toward one of the alternatives in the decision-
making process. Our research used both AHP and VIKOR methods separately and
compared both methods’ results to obtain clarity of the outcomes.
(A) AHP method
The AHP method is developed by Thomas L. Saaty (1980) in which problem
structure is generally a multilevel hierarchy (Narasimhan, 1983). The method is a
186 S. Kumar and M. K. Barua

multicriteria decision-making method that solves complex problems by breaking

them into levels of evaluation criteria (Jensen, 1984). The method top layer repre-
sents the goal of the problem and middle level, namely the tactic layer, representing
intermediate links. The bottom layer represents the criteria and the subcriteria
that used to evaluate the problems. AHP commonly consists of five main steps
explained below:
1. Formulate the hierarchic tree: In this step, a hierarchy consists of three different
steps (goal, criteria, and alternative) is constructed.
2. Construct pairwise comparison matrix: A pairwise comparison matrix
constructed in which elements in each level are compared in pair w.r.t. their impor-
tance to an element in the next higher level. Starting at the top of the hierarchy
and working down, the pairwise comparisons at a given level can be reduced to a
number of square matrices A = [aij ]n×n as in the following:
⎛ ⎞
a11 a12  a1n
⎜ a21 a22  a2n ⎟
⎜ ⎟
⎝M M O M⎠
an1 a2n  ann

A pairwise comparison matrix is filled by the industry expert using a nine-point

scale, as shown in Table 2. Where 1 is referred to as “equally important” while 9
referred to as “extremely important.”
3. Find out relative weights of elements in the hierarchy: Once the pairwise
comparison matrix is constructed, the vector of weight is calculated on the basis
of the eigenvalue procedure. It consists of two steps—one is comparisons matrix
A = [aij ]n×n normalization another is weight calculation.
Normalization
ai j
ai∗j = n
i=1 ai j

For all j = 1, 2, …., n.

Weight calculation

Table 2 Saaty’s 1–9 scale for

Numerical rating The verbal judgment of preferences
pairwise
1 Equal important
3 Moderately important
5 Strongly important
7 Very strongly important
9 Extremely important
2, 4, 6, 8 Intermediate rate
Green Supply Chain Management in the Indian Petroleum Industry … 187

Table 3 Randomly inconsistency indices

N 1 2 3 4 5 6 7 8 9 10
RI 0.00 0.00 0.58 0.90 1.12 1.24 1.32 1.41 1.46 1.49

n
j=1 ai∗j
wi =
n
For all i = 1, 2, …, n.
The relationship between pairwise comparisons matrix and vector weight as:

Aw = λmax w

4. Consistency check: Once experts’ judgment has been entered, it is necessary to

check that they are consistent. Consistency ratio (CR), also known as an index on
inconsistency, is used to represent the probability that the matrix was randomly
generated.

CR = CI/RI

where RI is the average of the resulting consistency index, and it depends on the
order of the matrix. Randomly inconsistency indices (RI) for N = 10 are shown in
Table 3.
CI represents the consistency index, and it has a range of 0–1. It can be given as:

CI = (λmax − n)/(n − 1)

where λmax is the largest eigenvalue of the matrix and n is the order of the matrix. If
the CR < 0.1, then the comparisons are acceptable.
5. Calculate overall priority ranking: In this step, the overall priority of the
different decision alternatives is determined. In this average criterion weight with
the average priority weight for the different decision, alternative combine with
respect to the different criteria.
(B) VIKOR method
The VIKOR method is an MCDM method used to solve discrete decision problems
that consist of non-commensurable and conflicting (competing) criteria (Opri-
cović, 1998). The method focuses on the ranking of alternatives and selection of
alternatives. The VIKOR method consists of five steps explained below as:
1. Established a matrix of criteria and different alternatives: The first step is to
formulate the decision matrix of criteria and the alternative as following:
188 S. Kumar and M. K. Barua

⎛ ⎞
I11 I12  I1n
⎜ I21 I21  I2n ⎟
X =⎜
⎝M
⎟
M O M⎠
In1 In2  Inn

2. Normalize the decision matrix: In this step, normalize the decision matrix as:
j
Ii
fi j = , i = 1, 2, . . . , m; j = 1, 2, . . . , n
m j
2
i=1 Ii

3. Determine the ideal solutions ( f i∗ ) and Nadir Value ( f i− )

If the ith function represents the benefit, then

f i∗ = max f i j , ..., f i− = min f i j

j j

If the ith function represent the cost, then

f i∗ = min f i j , ..., f i− = max f i j

j j

4. Compute the S j ,Rj and Qi values


f i∗ − f i j
n
Sj = wi 
i=1
f i∗ − f i−
 
f i∗ − f i j
R j = max wi 
i f i∗ − f i−
 
S j − S∗ R j − R∗
Qj = υ − + (1 − υ) −
(S − S ∗ ) (R − R ∗ )

where

S ∗ = min S J , ..., S − = max S j

J j

R ∗ = min R J , ..., R − = max R j

J j

5. Rank the alternatives: By sorting the values S, R, and Q, in decreasing order

ranking can be done of the alternatives.
Green Supply Chain Management in the Indian Petroleum Industry … 189

4 A Case of the Indian Petroleum Industry and Result

and Discussion

4.1 Case Study

The oil and gas industry is one of the largest industries in the world. This industry
still accounts for the majority of the world’s energy generation despite increasing
the viability and acceptability of alternative sources of fuels. The petroleum industry
is divided into three parts upstream, midstream, and downstream. In the upstream
exploration and production comes, in midstream crude oil transportation from the
production to the storage take and storage of crude oil comes and in the downstream
fuel transportation from the storage take to the refinery, marketing, and distribu-
tion come. Despite doing a lot to achieve sustainability oil and gas industry is often
accused of the environmental harness. In our research, we are exploring environ-
mental sustainability through various green practices. The green practices are clas-
sified into two groups—one is basic green practices and advanced green practices.
Basic green practices are scrutinizing through literature review, and advanced green
practices are acquired by Indian petroleum industry experts. Our study experts panel
are from Indian Oil Corporation Limited, Bharat Petroleum Corporation Limited,
Oil and Natural Gas Corporation, and Reliance Industries Limited. A seven-member
team was constituted for filling the pairwise matrix. Team members have a different
type of perspective to identify important green practices. Different types of team
composition become an important decision of quality of green indices identifica-
tion. General Manager, logistic expert, drilling expert, tow marketing experts, R&D
expert, and customer service expert should be involved in the decision making the
process of green Indices identifications.

4.2 Implementation of Proposed Methodologies

To show the quantitative data analysis of the determination of priorities of factors of

green practices, comparisons are conducted between the AHP and VIKOR methods.
All the analysis of both the method done through R statistical software. R-codes of
AHP and the VIKOR method are written in Appendix-1. Here, in section contains
some brief data analysis. The first method is AHP is explained as: The first step
is to construct the pairwise matrix and acquired data from the decision experts.
Comparisons of criteria are given in Table 4.

Table 4 Pairwise
Basic practices Advanced practices
comparison—criteria
comparison Basic practices 1 1/7
Advanced practices 7 1
190 S. Kumar and M. K. Barua

Here, alternatives are divided in two criteria as basic green practice and advanced
green practices. Two pairwise comparisons matrixes were constructed according to
the criteria. These comparison matrixes were filled by the experts as shown in Tables 5
and 6.
The next step of the AHP is to check the consistency of the data. The consistency
requirement is consistency ration (CR) should be less than 0.1. In case of both criteria,
basic green practices consistency ratio (CR = 0.398) and advanced green practices
consistency ration (CR = 0.3938) are satisfied with consistency requirement. The
upcoming data analysis provides the importance of green practices as shown in Table
7.

Table 5 Pairwise comparison—basic green practices

DEA ETP PDP RPS REC POM IIE
DEA 1 1/4 1/3 1/6 1/4 2 1/7
ETP 4 1 4 1/2 4 1/5 1/6
PDP 3 1/4 1 3 2 1 2
RPS 6 2 1/3 1 3 1/2 4
REC 4 1/4 1/2 1/3 1 1/4 1/3
POM 1/2 5 1 2 4 1 1/3
OOE 7 6 1/2 1/4 3 3 1

Table 6 Pairwise comparison—advanced green practices

DEA ETP PDP RPS REC POM IIE
DEA 1 1/5 1/6 1/4 1/2 1/7 1/6
ETP 5 1 3 1/5 7 1/6 1/7
PDP 6 1/3 1 1/4 4 1 4
RPS 4 5 4 1 5 1/5 4
REC 2 1/7 1/4 1/5 1 1/2 1/5
POM 7 6 1 5 2 1 1/2
OOE 6 7 1/4 1/4 5 2 1

Table 7 Ranking of the

Green practices Weights Ranking
green practice using AHP
method DEA 0.0257 7
ETP 0.1082 5
PDP 0.1539 4
RPS 0.2380 2
REC 0.0375 6
POM 0.2476 1
IIE 0.1889 3
Green Supply Chain Management in the Indian Petroleum Industry … 191

Table 8 Scale value 1–5

Intensity of important Definition
1 Equal important
2 Moderate important
3 Strong important
4 Very important
5 Extreme important

By using the same set of criteria of identification, green indices is analyzed by

VIKOR method. Decision-maker uses a scale to fill the questionnaire matrix by
using a 1–5 scale as shown in Table 8. Scale rating of decision-makers to basic green
practices and advanced green practices shown in Tables 9 and 10.
Criteria weight is given by the decision-maker as basic green practices (0.4) and
advanced green practices (0.6). Both the criteria fall into the category of beneficial
function. After done analysis using R statistical software ranking of alternative are
given in Table 11.

Table 9 Basic green practices rating by decision experts

DEA ETP PDP RPS REC POM IIE
DEA 2 1 2 4 2 2 3
ETP 2 2 2 3 2 4 3
PDP 6 3 3 3 4 2 4
RPS 4 4 2 3 3 3 3
REC 2 5 3 4 3 4 2
POM 3 4 4 5 2 4 4
OOE 3 4 4 5 5 3 3

Table 10 Advanced green practices rating by decision experts

DEA ETP PDP RPS REC POM IIE
DEA 3 4 2 3 2 4 5
ETP 2 2 4 4 4 2 3
PDP 2 3 3 5 3 2 2
RPS 2 3 4 5 5 3 4
REC 4 5 5 2 2 2 4
POM 3 4 2 3 2 4 3
IIE 3 4 5 5 2 4 3
192 S. Kumar and M. K. Barua

Table 11 Ranking of green practices

S (Group utility) R (Individual regret) Q (Aggregating index) Ranking
DEA 0.6964 0.4000 0.8097 5
ETP 0.7310 0.4023 0.8353 6
PDP 0.6739 0.6000 0.9609 7
RPS 0.1834 0.1834 0.2783 2
REC 0.3428 0.1976 0.3992 3
POM 0.5393 0.5011 0.7865 4
IIE 0.0000 0.0000 0.0000 1

Table 12 Common (AHP

Method Main four green practices Common practices
and VIKOR) main green
indices AHP POM, RPS, IIE, PDP RPS, IIE, POM
VIKOR IIE, RPS, REC, POM

5 Results

The most important four green practices in the AHP method, namely POM
(preventing oil and methane leaks), RPS (risk prevention systems to cover possible
environment accidents), IIE (investing into and implementing renewable energy),
and PDP (proper disposal of environmental pollutants). However, in case of VIKOR
method, the most four green practices are IIE (investing into and implementing renew-
able energy), RPS (risk prevention systems to cover possible environment accidents),
REC (regularly estimating cost and those activities that protecting the environment),
and POM (preventing oil and methane leaks).
There are three green practice indices common (out of the main four) in both the
methods shown in Table 12.
So, therefore, we can conclude that the most important green practice indices in
petroleum companies are IIE (investing into and implementing renewable energy),
RPS (risk prevention systems to cover possible environment accidents), and POM
(preventing oil and methane leaks).

6 The Implication of the Study

In this study, we considering the green practices indices associated with the petroleum
industry. Our study’s findings have significant insights, which have several impli-
cations for managers, society, and academicians. Based on the insights, top-level
managers can take specific measures to prioritize, implement, and improve the
identified prime green practice indices.
Green Supply Chain Management in the Indian Petroleum Industry … 193

Based on the theoretical perspective, the proposed study develops the top three
most significant green practices identification method by considering AHP and
VIKOR multicriteria decision-making methods. The proposed study helps iden-
tify the common indices by considering both the method for better obtaining the
result. In the present study, the main green practices are investing in and imple-
menting renewable energy: renewable energy is a highly popular market nowadays,
petroleum companies are taking notice. Investing in the renewable market doesn’t
diminish the investments; it is the future development and growing market in the long
term. Secondly, the main indices are risk prevention systems to cover possible envi-
ronmental accidents: mostly oil and gas produced offshore, often in harsh geograph-
ical geological conditions. The industry workers face the risk of fire and explosion
due to the ignition of flammable vapors or gases. So there is a great need to protect
the accident and save the human being, environment, and huge investment. Thirdly,
the main indices prevent oil and methane leaks: Methane is the main source of air
pollution (sulfur dioxide, carbon monoxide, ammonia, and other toxic compounds)
and climate change. By using the new technology with the help of digitalization,
petroleum companies can minimize waste full oil spill, breakdown, and unwanted
methane leaks.

7 Limitation and Future Work

This study comprises a number of limitations, which have been mentioned here.
Firstly, the study considers only seven green practices, whereas more practices can
be taken into consideration. Secondly, in our study, we consider seven decision
experts. At the same time, more decision experts could give their views regarding
environmental sustainability. The proposed methods only consider only the ranking
of green practices; therefore, one future work can be extended to determine inter-
relationship among important green practices. To follow through this, one can use
DEMATEL.

8 Conclusion

Petroleum companies have been accused of creating disorder in global environ-

mental issues and face the trustable issue by policymakers and customers. Therefore,
the present explores environmental sustainability in the petroleum industry. A case
study of the Indian petroleum industry has been taken to examine the petroleum
industry’s prime green practices. This study has taken two criteria: basic green prac-
tices and advanced green practices and seven green practices’ attributes. Out of seven
attributes, three attributes were proposed for the first time in the literature. This study
uses two separate MCDM methods, i.e., AHP and VIKOR, for identifying the prime
green practices in the petroleum industry. These MCDM methods are implemented in
194 S. Kumar and M. K. Barua

two phases; firstly, the AHP uses to find out the prime green performance indices, and
secondly, the VIKOR method is used. AHP has been accepted as a robust and flexible
MCDM tool for dealing with complex decision problems. VIKOR is a compromise
ranking method that determines the compromise ranking list. It focuses on ranking
and selecting from a set of alternatives in the presence of conflicting criteria. After
finding the prime green practices by both the methods, common green practices are
identified. The proposed study’s methodological novelty is the use of two different
MCDM methods in group decision making to find out the common green practices.
These prime green practices invest in and implement renewable energy, risk preven-
tion systems to cover possible environmental accidents, preventing oil and methane
leaks.

Acknowledgements The authors would like to thank the anonymous reviews for their generous,
insightful, and constructive comments on earlier versions, which helped to improve the paper
drastically.

Appendix I

R-Codes of AHP Methodology

Green Supply Chain Management in the Indian Petroleum Industry … 195

## Goal: Ranking of Green practices in petroleum industry

## Criteria (2): "BasicGreenPractices", "AdvancedGreenPractices"

## Alternatives (7): "DEA", "ETP", "PDP","RPS","REC","POM","IIE"

## Alternatives for Basic green practices criterion.

BasicGreenPractices=t(matrix(c(1, 1/4, 1/3, 1/6, 1/4, 2, 1/7,

4, 1, 4, 1/2, 4, 1/5, 1/6,

3, 1/4, 1, 3, 2, 1, 2,

6, 2, 1/3, 1, 3, 1/2, 4,

4, 1/4, 1/2, 1/3, 1, 1/4, 1/3,

1/2, 5, 1, 2, 4, 1, 1/3,

7, 6, 1/2, 1/4, 3, 3, 1),

nrow = 7, ncol = 7)); BasicGreenPractices

colnames(BasicGreenPractices)= c("DEA", "ETP",

"PDP","RPS","REC","POM","IIE" )

rownames(BasicGreenPractices)= c("DEA", "ETP",

"PDP","RPS","REC","POM","IIE"); BasicGreenPractices

AdvancedGreenPractices= t(matrix(c(1, 1/5,1/6, 1/4, 1/2, 1/7, 1/6,

5, 1, 3, 1/5, 7, 1/6, 1/7,

6, 1/3, 1, 1/4, 4, 1, 4,

4, 5, 4, 1, 5, 1/5, 4,

2, 1/7, 1/4, 1/5, 1, 1/2, 1/5,

196 S. Kumar and M. K. Barua

7, 6, 1, 5, 2, 1, 1/2,

6, 7, 1/4, 1/4, 5, 2, 1),

nrow = 7, ncol = 7)); AdvancedGreenPractices

colnames(AdvancedGreenPractices)= c("DEA", "ETP",

"PDP","RPS","REC","POM","IIE")

rownames(AdvancedGreenPractices)= c("DEA", "ETP",

"PDP","RPS","REC","POM","IIE"); AdvancedGreenPractices
## Alternative for all criterion.
Green Supply Chain Management in the Indian Petroleum Industry … 197

alternativesPC= list(BasicGreenPractices=BasicGreenPractices, Ad-

vancedGreenPractices=AdvancedGreenPractices); alternativesPC

## Criteria with respect to goal.

criteriaPC= t(matrix(c(1,1/7,

7, 1),

nrow=2, ncol=2 )); criteriaPC

colnames(criteriaPC)= c("BasicGreenPractices", "AdvancedGreenPractic-

es")

rownames(criteriaPC)= c("BasicGreenPractices", "AdvancedGreenPractic-

es"); criteriaPC

## Consistency Check.

library(MCDA)

pairwiseConsistencyMeasures(criteriaPC)$CR

pairwiseConsistencyMeasures(BasicGreenPractices)$CR

pairwiseConsistencyMeasures(AdvancedGreenPractices)$CR

## AHP Score for each alternative.

overall= AHP(criteriaPC, alternativesPC); overall

# Rank the scores of the alternatives.

rank(-overall)

#######

R-Codes of VIKOR Methodology

198 S. Kumar and M. K. Barua

## Goal: Ranking of Green practices in petroleum industry.

## Criteria (2): "BasicGreenPractices", "AdvancedGreenPractices".

## Alternatives (7): "DEA", "ETP", "PDP","RPS","REC","POM","IIE".

## Matrix with Performance scores.

performanceTable= matrix(c(2.29, 3.29,

2.50, 3.14,

3.57, 2.83,

3.14, 3.71,

3.29, 3.43,

3.71, 3.00,

3.86, 3.71),

nrow=7, ncol = 2,

byrow = TRUE); performanceTable

rownames(performanceTable)= c("DEA", "ETP",

"PDP","RPS","REC","POM","IIE" )

colnames(performanceTable)= c("BasicGreenPractices", "Advanced-

GreenPractices"); performanceTable

## criteria weigths

weights = c(0.40, 0.60)

names(weights)= colnames(performanceTable); weights

Green Supply Chain Management in the Indian Petroleum Industry … 199

##perferance direction for each criterion

criteriaMinMax = c("max", "max")

names(criteriaMinMax)= colnames(performanceTable); criteriaMinMax

## Vikormodeling

library(MCDM)

##Define acceptable voting level

v=0.5

mod=VIKOR(performanceTable,weights, criteriaMinMax, v); mod

######

References

Ahmad, W. N. K. W., de Brito, M. P., & Tavasszy, L. A. (2016). Sustainable supply chain manage-
ment in the oil and gas industry: A review of corporate sustainability reporting practices.
Benchmarking: An International Journal, 23(6), 1423–1444.
Alipour, N., Sangari, M. S., & Nazari-Shirkouhi, S. (2019, January). Investigating green human
resource practices in the healthcare sector: A joint application of balanced scorecard and SIR
method. In 2019 15th Iran International Industrial Engineering Conference (IIIEC) (pp. 283–
288). IEEE.
Antão, P., Calderón, M., Puig, M., Michail, A., Wooldridge, C., & Darbra, R. M. (2016). Identifica-
tion of occupational health, safety, security (OHSS) and environmental performance indicators
in port areas. Safety Science, 85, 266–275.
Asif, M., & Muneer, T. (2007). Energy supply, its demand and security issues for developed and
emerging economies. Renewable and Sustainable Energy Reviews, 11(7), 1388–1413.
Beske, P., Land, A., & Seuring, S. (2014). Sustainable supply chain management practices and
dynamic capabilities in the food industry: A critical analysis of the literature. International Journal
of Production Economics, 152, 131–143.
Fernando, Y., Shaharudin, M. S., & Wahid, N. A. (2016). Eco-innovation practices: A case study
of green furniture manufacturers in Indonesia. International Journal of Services and Operations
Management, 23(1), 43–58.
Ford, J. A., Steen, J., & Verreynne, M. L. (2014). How environmental regulations affect innovation
in the Australian oil and gas industry: Going beyond the porter hypothesis. Journal of Cleaner
Production, 84, 204–213.
George, R. A., Siti-Nabiha, A. K., Jalaludin, D., & Abdalla, Y. A. (2016). Barriers to and enablers
of sustainability integration in the performance management systems of an oil and gas company.
Journal of Cleaner Production, 136, 197–212.
200 S. Kumar and M. K. Barua

Golden, J. S., Virdin, J., Nowacek, D., Halpin, P., Bennear, L., & Patil, P. G. (2017). Making sure
the blue economy is green. Nature Ecology & Evolution, 1, 0017.
Hamilton, J. D. (2009). Causes and consequences of the oil shock of 2007–08 (No. w15002).
National Bureau of Economic Research.
Hill, M. R. (2001). Sustainability, greenhouse gas emissions and international operations manage-
ment. International Journal of Operations & Production Management, 21(12), 1503–1520.
Holditch, S. A. (2013). Unconventional oil and gas resource development–Let’s do it right. Journal
of Unconventional Oil and Gas Resources, 1, 2–8.
Jensen, R. E. (1984). An alternative scaling method for priorities in hierarchical structures. Journal
of Mathematical Psychology, 28(3), 317–332.
Johnston, J. E., Werder, E., & Sebastian, D. (2016). Wastewater disposal wells, fracking, and
environmental injustice in southern Texas. American Journal of Public Health, 106(3), 550–556.
Jung, E. J., Kim, J. S., & Rhee, S. K. (2001). The measurement of corporate environmental
performance and its application to the analysis of efficiency in oil industry. Journal of Cleaner
Production, 9(6), 551–563.
Kohli, R., & Johnson, S. (2011). Digital transformation in latecomer industries: CIO and CEO
leadership lessons from Encana Oil and Gas (USA) Inc. MIS Quarterly Executive, 10(4).
Krichene, N. (2002). World crude oil and natural gas: A demand and supply model. Energy
Economics, 24(6), 557–576.
Matos, S., & Hall, J. (2007). Integrating sustainable development in the supply chain: The case
of life cycle assessment in oil and gas and agricultural biotechnology. Journal of Operations
Management, 25(6), 1083–1102.
McKenzie, L. M., Witter, R. Z., Newman, L. S., & Adgate, J. L. (2012). Human health risk assess-
ment of air emissions from development of unconventional natural gas resources. Science of the
Total Environment, 424, 79–87.
Narasimhan, R. (1983). An analytical approach to supplier selection. Journal of Purchasing and
Materials Management, 19(4), 27–32.
Obiajunwa, E. I., Pelemo, D. A., Owolabi, S. A., Fasasi, M. K., & Johnson-Fatokun, F. O. (2002).
Characterisation of heavy metal pollutants of soils and sediments around a crude-oil production
terminal using EDXRF. Nuclear Instruments and Methods in Physics Research Section b: Beam
Interactions with Materials and Atoms, 194(1), 61–64.
O’Dea, A., & Flin, R. (2001). Site managers and safety leadership in the offshore oil and gas
industry. Safety Science, 37(1), 39–57.
Omer, A. M. (2008). Energy, environment and sustainable development. Renewable and Sustainable
Energy Reviews, 12(9), 2265–2300.
Opricović, S. (1998). VIKOR method. Multicriteria optimization of civil engineering systems
(pp. 142–175). University of Belgrade-Faculty of Civil Engineering.
Schneider, J., Ghettas, S., Merdaci, N., Brown, M., Martyniuk, J., Alshehri, W., & Trojan, A.
(2013). Towards sustainability in the oil and gas sector: Benchmarking of environmental, health,
and safety efforts. Journal of Environmental Sustainability, 3(3), 6.
Shah, K. U., & Rivera, J. E. (2007). Export processing zones and corporate environmental perfor-
mance in emerging economies: The case of the oil, gas, and chemical sectors of Trinidad and
Tobago. Policy Sciences, 40(4), 265–285.
Shell. (2018). Sustainability report of shell in 2018 (pp. 1–86). https://reports.shell.com/sustainab
ility-report/2018/
Yusuf, Y. Y., Gunasekaran, A., Musa, A., El-Berishy, N. M., Abubakar, T., & Ambursa, H. M.
(2013). The UK oil and gas supply chains: An empirical analysis of adoption of sustainable
measures and performance outcomes. International Journal of Production Economics, 146(2),
501–514.
Zhu, Q., Sarkis, J., & Lai, K. H. (2007). Green supply chain management: Pressures, practices and
performance within the Chinese automobile industry. Journal of Cleaner Production, 15(11–12),
1041–1052.
Staff Allocation for Projects in IT Service
Industries: A Simulation-Based
Approach

V. V. Rajarajan and M. S. Gajanand

Abstract Finding and retaining the right team is the most important aspect of
successful resource management. Something that is equivalently important, espe-
cially in the context of Information Technology (IT)-based service industries, is
the effective allocation of the staff to the available projects. However, managers
often allocate resources based on the current situation without adopting a carefully
thought-out process. This chapter highlights the various challenges in allocating staff
for projects in service industries and presents a possible simulation-based approach
for better allocation of resources, considering various aspects like the project type,
frequency, capability, employee levels and the typical time spent by them on projects.
An illustration of the proposed approach is presented using a case study, followed
by the practical implications.

Keywords Service industries · Resource allocation · Simulation · IT services

1 Introduction

Information Technology (IT) professional services include (but is not limited to)
project-oriented services, market research and analysis, IT support and training
services, enterprise cloud computing services and also the integration of these
services into the existing business operations of organizations. Enterprises, irre-
spective of the industry vertical, are opting for IT professional services owing to its
manifold benefits. Thanks to this, the number of IT professional service projects that
are being outsourced and the IT service output of several countries has significantly
increased over the years. In 2017, the global IT professional services market was
valued at $562.06 billion, and by 2025, it is expected to reach $1,070.28 billion

V. V. Rajarajan (B)
Manager - Business Analytics, ZoomRx Healthcare Solutions Pvt Ltd, Chennai, Tamil Nadu, India
e-mail: [email protected]
M. S. Gajanand
Faculty of Operations Management and Decision Sciences, Indian Institute of Management,
Tiruchirappalli, Tamil Nadu, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 201
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_10
202 V. V. Rajarajan and M. S. Gajanand

Fig. 1 IT service output of four economies, 2005–2015

(Grand View Research, 2019). India has become an important hub to offer IT profes-
sional services to organizations around the world. Figure 1 shows the IT service
output of the European Union, USA, India and China (Henry-Nickie et al., 2019).
Organizations which provide IT professional services, while having operational
efficiency as a priority, are actively seeking different ways to improve their service
quality and profitability. They have emphasized the optimization of operational
processes to deliver value and to meet customer expectations. This has also become
inevitable due to the increasing pressure of globalization, increasingly competitive
markets and volatile market dynamics (Yee et al., 2008). The effective allocation
of resources to the projects is very essential to achieve operational efficiency, meet
customer expectations, improve service delivery and profitability.
According to a study by McKinsey, about 83% of executives highlighted resource
allocation as the most important management lever responsible for the growth of an
organization. Resource allocation refers to the diligent deployment of resources to
do work on different projects, considering various factors including their skill set
and timelines. Poor resource management has seen an increasing trend in several
organizations during the recent years. Project and operations managers often fail to
use allocation of resources to their advantage. They often allocate resources based
on the current situation without adopting a carefully thought-out process.
While allocating resources to projects, organizations should not think of it as
delegating assignments. Resource allocation may significantly affect business perfor-
mance as it impacts the functioning of any organization and plays an important role
in project and operations management. It affects the operating margin and profit
of the organization. Adopting a process that promotes healthy allocation of staff to
projects can also lead to efficient workforce planning and improve the overall morale
of the team members. Some of the methods that can take you along this path include
optimizing the utilization levels of all resources involved. This chapter highlights the
various challenges in allocating staff for projects in service industries and presents
Staff Allocation for Projects in IT Service Industries … 203

a possible simulation-based approach for better allocation of resources, considering

various aspects like the project type, frequency, capability, employee levels and the
typical time spent by them on projects.

2 Related Work

Various mathematical models for workforce scheduling have been successfully

adapted and applied to different scenarios in the past. Baker (1976) presented a survey
on the basic mathematical models for workforce scheduling with cyclic demand
for staff and reviewed several problem areas where these models were success-
fully applied. Segelod (2002) described and identified the different factors which
influenced the resource allocation process in a few industries. They highlighted
the perceived development direction in which resource allocation was headed and
showed how five key factors affected the resource allocation systems based on their
study in six industries. Ernst et al. (2004) reviewed the applications of rostering
problems in different areas and the corresponding models and algorithms that were
used to obtain the solution. They highlighted that each industry sector will have
their own set of issues and so they must be viewed carefully with due attention
to sector-specific details. They concluded that organizations will require decision
support systems in order to get the various benefits obtained by optimizing the staff
schedules. These decision support systems must be carefully implemented to satisfy
customer demands at a lower cost while meeting requirements like staff preferences,
shift equity, flexible workplace agreements, etc.
Chen et al. (2016) integrated medical staff allocation and staff scheduling in the
context of an uncertain environment. They developed a two-stage goal programming
approach to identify the smallest possible number of medical staff required and
to determine the best schedule for them. This research presented an approach to
determine the right staff size required based on the objective function to reduce the
number of medical staff. Kekkonen et al. (2018), in the healthcare context, studied the
influence of rational task allocation on the increase in the overall positive outcome of
the system. They studied the ongoing relationship between the in-house logistics and
material supply services, and the nursing staff of an organization. They also presented
various practical ideas for improving the cooperation between the stakeholders.
Namasivayam and Denizci (2006) studied the impact of value-creating processes
on human capital valuation in service industries. They presented a tool for identifying
the human capital value drivers across various industries. The value matrix proposed
by them can be used to group industries and make comparisons, both across and
within industries, based on the value characteristics. Yee et al. (2008) used structural
equation modelling and investigated the impact of employee satisfaction on opera-
tional performance in service industries. They showed that the employee satisfaction
significantly impacts service quality and customer satisfaction, and this in turn influ-
ences the profitability of the firm, leading to a “satisfaction–quality–profit cycle”.
204 V. V. Rajarajan and M. S. Gajanand

As a result, operations managers should prioritize employee satisfaction in order to

boost service quality and customer satisfaction.
Murakami et al. (2010) studied the human resource allocation problem (hRAP)
for optimal allocation of human resources to tasks while trying to minimize the
total human resource costs and the human resource usage. In addition to the other
operational constraints, they also considered precedence and skill constraints. They
designed it as a network problem and proposed a multistage genetic algorithm to
determine the solution. Ağralı et al. (2017) considered an employee scheduling
problem with flexible demand and flexible employee availability. They formulated
the problem as a mixed-integer programming model, capable of solving practical
scenarios within a reasonable time. They generated an assignment of employees
to service requirements such that conditions such as maximum weighted number of
service requirements, government and labour union regulations, individual employee
work hours and a fair balance between employee work assignments are satisfied.
Sasanfar et al. (2020), with a goal to improve the workflow of the patients and staff,
presented an approach to improve the allocation of resources in various wards of an
emergency department.
Momeni and Martinsuo (2018) conducted a qualitative comparative case study in
two project-based service firms and highlighted prioritization and adapting to change
and delay to be the major issues which managers will face while allocating resources
to projects in dynamic environments. They identified the various challenges and
practices adopted for resource allocation in service units which carry out both project
and non-project activities. They showed that project resource allocation mechanisms
must be replaced or supplemented with more flexible context-dependent practices,
rather than having a top-down approach. Some of the learnings from the literature on
service industries can be applied to the IT-based service industries as well. However,
the resource allocation in the case of IT service industries require a slightly different
approach, owing to the unique nature of the sector.

3 Problem Description

3.1 Background

We highlight the generic process flow that happens in most IT service projects in
Fig. 2. The companies which require part of their work to be outsourced would send
out an intimation with all their requirements for a project to all the potential service
providers seeking a request for proposals. Based on the requirements, the service
providers interested to take up the project would submit their proposal highlighting
all their capabilities. The company and the service provider would confirm the part-
nership upon mutual agreement of terms can conditions and sign the statement of
work (SOW). Post the confirmation, based on the scope agreed upon and the need
of the project, the staffing for the project would be done.
Staff Allocation for Projects in IT Service Industries … 205

Request for Proposal Project Identify

Proposal (RFP) Submission Confirmation Staff

Data Collec- Top-line Complete Project Clo-

tion and Analysis Report Report sure / Renewal

Fig. 2 Generic process flow of IT service projects

Based on the business requirements, the data collection process would be started,
and the project will be executed by the identified set of resources. An initial top-line
report will be prepared to give a summary and quick insights on the tasks completed.
A complete report will provide the exhaustive information and details about all the
activities carried out as a part of the project. Depending on the requirements of the
company, the project with the service provider may either be renewed or closed. The
completed projects will be earmarked as delivered, and in certain cases, the service
provider may also provide support for an extended period.
Identifying the resources (staff) required to carry out the project is a crucial
stage in the life cycle of a project. While allocating resources to projects, service
providers often tend to ignore factors like priorities, resource availability, dependen-
cies, estimations, etc. and allocate by virtue of situations. Currently, the projects are
staffed manually by considering the existing project load purely based on the total
number of existing projects. Any information about the intensity and scope of the
existing projects are not utilized to determine new allocations. This creates unnec-
essary burden and overload of work on a few resources. Also, it often leads to poor
distribution of workload among the available people and inefficient utilization of
resources.

3.2 Problem Statement

Most organizations have an existing staffing model for IT service projects based
on manual intervention and it is purely based on the free time available with the
employees. Upon the arrival of a new project into the system, the project gets allocated
to the team members based on their availability at that moment. This may not be the
effective way to manage the existing load and the new load. The need of the hour is
an effective staffing approach, which considers various aspects that characterize and
influence both existing and incoming projects, so that proper utilization of resources
can be achieved.
206 V. V. Rajarajan and M. S. Gajanand

4 Proposed Solution Approach

Service providers are actively seeking ways to measure the effectiveness of their
existing staffing model and to determine the required data points that would help
them in achieving an effective mechanism for staffing their projects. They should
have a clear understanding of the important parameters that must be considered
while making the staffing decisions as these parameters may vary depending on
the nature and type of the service industry. The existing workload allocated to the
people across the various projects needs to be clearly identified and recorded. An
assessment of whether the Business Analysis team is currently working under ideal
efficiency conditions is essential. Based on these requirements, we present an alter-
native approach for identifying and allocating staff as shown by the shaded elements
in Fig. 3.

4.1 Evaluate Hours Required

The details of all the projects that are currently carried out by the service provider is an
important information that is required and will be available with most organizations.
We propose to collect and use the data of the time spent by each resource on various
projects in the past. Using this data collected across different projects, we can arrive
at the time taken for each project by all the resources allocated to a particular project.
This will help to understand the time available with each resource, based on the
current set of projects assigned, and also help to estimate the amount of effort required
for a new project depending on the nature and type of the project.
As with any user reported data, the biggest challenge is to ensure data consistency
and data quality. Since the information is reported by the employees, it may be
different depending on the circumstances under which the work was completed, as

Request for Proposal Project Evaluate

Proposal (RFP) Submission Confirmation Hours Required

Simulate Determine Optimal Al- Data Collec-

Hours Available location of Personnel Ɵon and Analysis

Top-line Complete Project Clo-

Report Report sure / Renewal

Fig. 3 Flowchart of the proposed approach

Staff Allocation for Projects in IT Service Industries … 207

well as the capabilities of the employee. In some cases, there are also chances for
over-reporting and under-reporting. In order to ensure that the data collected is closer
to reality, it is necessary to have checks and processes in place. Management must
make sure that the employees understand the importance of the data collected, since
it is used to staff the employees according to the time available. This in turn helps
the employees as well. Also, it will be ideal if there is an effective way to automate
the whole data collection process and reduce manual intervention.

4.2 Simulate Hours Available

Using the evaluation of the hours required based on the past data, we can simulate the
expected number of hours that would be taken up by the resources for the existing
projects in the upcoming week. As and when the amount of data increases with
every week, the accuracy of the simulation will also increase. The efficiency and
capabilities of an employee may not affect the overall measure if this is simulated at
a member level. This simulated data will help us in determining the actual work hours
available for new projects across each of the designations in the Business Analysis
team.

4.3 Determine Optimal allocation

We propose to adopt a simulation-based optimization approach using the simulated

hours available for the employees. Even though this is the unbiased way to allocate the
staff to projects, we might face challenges in the future where there might be a need
to allocate staff based on the priority of the project as well. Allocation of personnel
to projects always comes with an issue associated with the efficiency and capacity
of the personnel. Any firm would want the best resources to work on high-value
projects. Additional restrictions, if any, on the composition of the members allocated
to a project may be incorporated in the simulation model in order to determine the
optimal allocation of personnel to the available projects.
We illustrate the proposed approach using a case study in the subsequent section.

5 Case Study

5.1 Company Background

We consider a pharmaceutical market research company, which is now at a stage

where it is slowly transitioning off from being a start-up into the next phase. They
208 V. V. Rajarajan and M. S. Gajanand

help other life sciences companies make better decisions by doing market research
differently. They believe that numbers and graphs alone cannot measure a brand’s
performance. Their research goes beyond the traditional awareness, trial and usage
statistics. They provide key insights on brand performance and suggestions for
improvement by collecting qualitative information and connecting it with secondary
data. Brands achieve great things when Marketing and Sales teams work in tandem.
All the market research studies that they do for their clients involve collection of
primary data through surveys.
Every market research study should at least have a minimum of two members
across the four designations in the company. Few of the projects might have more
than four members if the scope of the project is big. The four designations within
the business analysis team are: Associate, Associate Consultant, Consultant and
Manager. Depending on the requirement, the company also employs interns to offset
some of the relatively less-complex work done by an Associate and the Principal
Consultant might also work on certain projects occasionally. The Head of Operations
wanted to revisit the methodology using which they have been staffing their projects.
The business analysis team highlighted the drawbacks of the existing process and
proposed the need for an effective mechanism that considers various influencing
factors. They decided to adopt a structured data-based approach, which will help
them analyse the staffing decisions.

5.2 Data Collection

An internal tool was developed to track the time taken by an individual working
on a certain project. At the end of the work week, every team member enters the
total hours that they clocked in that week working across different projects. They
split the total hours across each of the project. There could be scenarios where some
team members would be staffed on a project, but would not have spent any hours
working on it. Some of them might have spent significant time on internal non-client
facing projects and might even have out of office (OOO) hours. All these details are
accurately captured through this mechanism. Based on the historical data collected
for each team member, we observed that the time taken by an individual varies across
the projects and followed a triangular distribution. We determine the maximum hours
spent over a week, minimum hours spent over a week and also the most likely time
spent over a week. The general spread of the time spent by an individual belonging
to a particular designation is given in Table 1.
The details about the type and the number of the projects that are available, their
time span and the estimated number of hours required are presented in Table 2.
Staff Allocation for Projects in IT Service Industries … 209

Table 1 Pattern of the typical time spent by an individual

Designation Number of people Minimum Most likely Maximum
Associate 25 37 45 60
Associate consultant 29 36 48 80
Consultant 10 36 52 77
Manager 5 37 52 71

Table 2 Nature and type of the projects available

Project type Time span Estimated total Hours required Number of Hours per
hours (Weekly) projects week per
project
ATU Yearly 27 1 2 1
Half-yearly 1613 85 4 21
Quarterly 8135 428 13 33
Monthly 779 41 3 14
One off 2739 144 8 18
ATU / CA Half-yearly 62 3 1 3
CA Half-yearly 319 17 1 17
Quarterly 2003 105 2 53
Monthly 607 32 1 32
Fortnightly 604 32 1 32
One off 611 32 1 32
Internal Monthly 1857 98 6 16
PET Half-yearly 357 19 3 6
Quarterly 6550 345 15 23
Bi-monthly 907 48 2 24
Monthly 6650 350 7 50
One off 1442 76 4 19
PET–Lite Quarterly 5361 282 8 35
Bi-monthly 543 29 1 29
Monthly 1314 69 1 69
One off 34 2 1 2
PET–TP1 Lite Half-yearly 243 13 1 13
Quarterly 344 18 2 9
Qual One off 67 4 1 4
RT Quarterly 531 28 2 14
One off 92 5 1 5
210 V. V. Rajarajan and M. S. Gajanand

5.3 Analysis and Results

In the existing system, the members (staff or resources) were allocated to projects
purely based on the current load, defined by the quantity (number) of projects that
they handled at any point in time. As seen in Table 1, the time taken by an individual
to work on a project is a random variable, and this needs to be accounted appropri-
ately while allocating the members to the projects. Across the various live projects
that are currently available in the organization, we needed to determine a better
staffing composition among the four designations (Associate, Associate Consultant,
Consultant and Manager).
We consider the number of members allocated to a project from each designation
as a decision variable. All these decision variables are discrete in nature with a lower
limit of 0 and an upper limit of 2. This is to ensure that not more than two members of
the same designation are staffed in the same project. In order to keep in line with the
company policy, the total number of Consultants and Managers across any project
is not preferred to exceed two. This also guarantees that the teams are not heavy
and are optimally staffed, and that there is appropriate role clarity for the business
analysis team members. Finally, to ensure that there is a proper communication
channel between the Associates and the Consultant / Manager group, at least one
Associate Consultant was required to be staffed across all the live projects. These
restrictions were incorporated in the simulation model.
We used Microsoft Excel and the OptQuest feature of Oracle Crystal Ball to build
and run the simulation-based optimization model for the analysis. Oracle Crystal
Ball is a spreadsheet-based application for predictive modelling, forecasting, simu-
lation and optimization. It gives insights into the various critical factors affecting the
simulation model. Table 3 shows the output of the model with the optimal staffing
plan for the projects with the resources from four designations based on 10,000 trials.
To understand the effectiveness of the staffing model, we compare the results
from the simulation model with the existing staffing model. Resources allocated to
the new project would form a team to work on the respective project. Also, at any
given point in time they would be part of multiple projects. Across these multiple
projects, they might have few overlapping team members or entirely new set of
members based on the resource allocation and availability. Table 4 shows a summary
of the analysis based on the total number of projects staffed and the average number
of projects across the designations. This is a high-level indication of how the project
load is evenly distributed across associates and associate consultants. Earlier, the role
played by an associate consultant as a bridge between associates and consultants was
not very evident across the project types. As associate consultants, they should be part

Table 3 Allocation of projects among the four designations of resources

Designation Associate Associate consultant Consultant Manager
Number of projects staffed 64 81 64 17
Average number of projects 3 3 6 3
Staff Allocation for Projects in IT Service Industries … 211

Table 4 Comparison of the existing staffing process with the proposed approach
Designation Number of projects staffed Average number of projects
Existing Proposed Difference Existing Proposed Difference
process approach process approach
Associate 64 64 0 3 3 0
Associate 71 81 −10 2 3 −1
consultant
Consultant 63 64 −1 6 6 0
Manager 63 17 46 13 3 10

of more projects as they play a crucial role for better functioning of the project and
this should not be left unfilled. The simulation model with the necessary constraints
ensures that this requirement is adhered to always.
The average number of projects across the designations is also an important indi-
cator of the improvement in terms of sharing of the workload. Based on the existing
staffing model, Managers were on an average staffed on 13 projects and the load was
not uniformly distributed across the roles.
After the allocation using the proposed methodology, the load is evenly distributed
across the various roles. Also, all team members across the designations (except the
consultants) are staffed on an average of three projects. This frees up the time of the
managers and releases the existing bottle neck, due to the sheer number of additional
projects assigned to a manager. This indeed opens the time and scope for business
development as well as focus on internal quality improvement to make the overall
process efficient.

5.4 Recommendations

Based on the results obtained from the simulation study, we can see that the existing
practice of staffing the projects purely based on the current load (number of projects
assigned) is not very efficient and should be avoided. Most of the Managers across
the floor are currently over-burdened with more work load than they can handle. In
some cases, certain Managers were part of almost 63 projects which is not essentially
required. Removing a Manager from certain projects will also help other members
in the team to step up and handle the next level of work in those projects. This way,
the organization can indirectly improve personnel development as well as constantly
reinforce the workforce with a fresh set of project managers.
Associate Consultants may be given more projects if they have the scope to take
up an additional project. If a Manager can be removed from the project, an Associate
Consultant can be staffed, so that the Consultant will have leverage to take care of
the tasks related to the execution of the project. By doing this, the organization will
212 V. V. Rajarajan and M. S. Gajanand

be relieving the work load of the Managers, and at the same time ensure maximum
utilization of other team members.

6 Practical Implications and Scope for Future Work

Earlier, managers used to spend a lot of time on the project execution. Based on the
proposed approach, the load on the managers have been distributed evenly across
the next level of employees and so they can execute their role as a Manager effec-
tively. They can also concentrate on business development initiatives. The increased
project handling responsibilities, as a result of the divided load due to the new staffing
method, enables the nurturing of Consultants for the next role simultaneously. Asso-
ciate consultants also get the opportunity to explore and contribute to insights, as
a result of the proposed staff allocation process. We can observe the desired and
required improvement across the three levels.
Staffing of the projects is an integral part of the processes involved in delivering
a project and so it should be given at most importance and the decisions should be
made purely based on data to avoid manual intervention. Considering various factors
such as the existing project load of an individual, nature of the project, frequency,
type, scope and other parameters is very essential for effective allocation of the staff
to projects. For a growing organization like the one under study, it is very essential
that they have a streamlined way to identify and allocate team members for each
project. By doing this, they can improve the quality and complete the work at the
right time by distributing the resources across different projects.
A database of the time taken by various individuals to complete different projects
is the key aspect required for analysing any scope for improvement and for adopting a
structured data-based approach for staff allocation. This approach will also eliminate
the inherent biases that are present in the manual allocation of staff to projects. All
of this is not only aimed at changing the staffing methodology but also ensures
constant personal and group-level development across the designations. Managers
having additional time at their disposal can concentrate on Business development
initiatives, which in turn will bring in more projects into the system at a higher rate
than what it is currently.
In this context, the scope of the project depends on the project type. For example,
a quarterly project goes through a scope change every quarter and we will be sending
a report for each quarter. The project type clearly captures the velocity of the project.
Technically, if the project scope remains the same or if the project has one objective
throughout the statement of work, then the velocity of the project would vary in the
beginning and reach a steady state. The type of the probability distribution used for
the activity times may vary depending on the type of the IT professional services
provided by the Organization. The model can be extended by considering the velocity
of the project as a separate random variable and also by using different probability
distributions for the time taken to complete a task by different individuals.
Staff Allocation for Projects in IT Service Industries … 213

7 Summary

Effective resource allocation involves much more than just matching a person to a
task. Effective resource allocation involves much more than just matching a person
to a task. There is sufficient literature evidence that the employee satisfaction signif-
icantly impacts service quality and customer satisfaction, and this in turn influ-
ences the profitability of the firm. Also, each industry sector has its own set of
issues and must be viewed carefully with due attention to sector-specific details.
Context-dependent practices are essential for efficient allocation of resources to
projects. Effective use of customized decision support systems may be required for
organizations to get the complete benefits from optimizing the staff schedules.
Owing to its unique nature, the IT service industries require a slightly different
approach for allocation of staff to projects. This chapter presented a simulation-based
approach for better allocation of staff to projects, considering various aspects like
the project type, frequency, capability, employee levels and the time spent by them
on projects. The results from the case study indicated that the existing practice of
staffing the projects manually is not very efficient and should be avoided. While the
proposed approach is designed specifically for service-based firms, the idea and the
concepts used are pretty much applicable to any industry.

Acknowledgements We thank Professor C. Rajendran and Dr. Vipin (the editors of this book)
for providing us the opportunity to contribute to this edited book. We also thank the anonymous
reviewers for their valuable feedback and suggestions to improve the quality of this work.

References

Ağralı, S., Taşkın, Z. C., & Ünal, A. T. (2017). Employee scheduling in service industries with
flexible employee availability and demand. Omega, 66(Part A), 159–169.
Baker, K. R. (1976). Workforce allocation in cyclical scheduling problems: A survey. Journal of
the Operational Research Society, 27(1), 155–167.
Chen, P. S., Lin, Y. J., & Peng, N. C. (2016). A two-stage method to determine the allocation and
scheduling of medical staff in uncertain environments. Computers and Industrial Engineering,
99, 174–188.
Ernst, A. T., Jiang, H., Krishnamoorthy, M., & Sier, D. (2004). Staff scheduling and rostering: A
review of applications methods and models. European Journal of Operational Research, 153(1),
3–27.
Grand View Research. (2019). IT professional services market size share and trends analysis report
by type by deployment (On-premise, Cloud) by end use by region and segment forecasts, 2019–
2025. https://www.grandviewresearch.com/industry-analysis/it-professional-services-market
Henry-Nickie, M., Frimpong, K., & Sun, H. (2019). Trends in the Information Technology sector.
https://www.brookings.edu/research/trends-in-the-information-technology-sector/
Kekkonen, P., Pohjosenperä, T., Kantola, H., & Väyrynen, S. (2018). Rational and participative
task allocation between the nursing staff and the logistics support service provider in healthcare.
Human Factors and Ergonomics in Manufacturing and Service Industries, 28(3), 117–129.
214 V. V. Rajarajan and M. S. Gajanand

Momeni, K., & Martinsuo, M. M. (2018). Allocating human resources to projects and services in
dynamic project environments. International Journal of Managing Projects in Business, 11(2),
486–506.
Murakami, K., Gen, M., Tasan, S. O., & Oyabu, T. (2010, July). A solution of human resource
allocation problem in a case of hotel management. In The 40th International Conference on
Computers and Industrial Engineering (pp. 1–6). IEEE.
Namasivayam, K., & Denizci, B. (2006). Human capital in service organizations: Identifying value
drivers. Journal of Intellectual Capital, 7(3), 381–393.
Sasanfar, S., Bagherpour, M., & Moatari-Kazerouni, A. (2020). Improving emergency departments:
Simulation-based optimization of patients waiting time and staff allocation in an Iranian hospital.
International Journal of Healthcare Management, 1–8.https://doi.org/10.1080/20479700.2020.
1765121
Segelod, E. (2002). Resource allocation in a few industries: Determinants and trends. International
Journal of Production Economics, 77(1), 63–70.
Yee, R. W., Yeung, A. C., & Cheng, T. E. (2008). The impact of employee satisfaction on quality
and profitability in high-contact service industries. Journal of Operations Management, 26(5),
651–668.
 Literature Review and Conceptual
Framework on Operations and Supply
Chain Management Aspects
Supply Chain Management Practice
and Supply Chain Performance:
A Conceptual Systematization
of Terminology and Proposed
Framework for SCMP Implementation
in Indian Manufacturing SMEs

Rohit Kumar and Manish Gupta

Abstract Supply chain management practice has been the most crucial competing
strategy to improve organizational performance and profitability. Over the last
decade, the concept of Supply chain management practice has grown and has become
an imperative component in supply chain operation. Therefore, a set of Supply chain
management practices have been implemented worldwide without covering its real
impact on supply chain performance and competitive advantage. A significant set of
recent research has attempted to find the impact of supply chain management prac-
tices on supply chain performance and ultimately on operational performance world-
wide. However, these attempts have been obstructed by the diversity and standard-
ization of the used nomenclature that characterize the supply chain management
practice and supply chain performance. In this context, firstly, this chapter presents a
conceptual systematization of the used nomenclatures that characterize supply chain
management practices and supply chain performance worldwide. Secondly, and more
importantly, this chapter aims to develop a framework for the successful implemen-
tation of supply chain managment practices in manufacturing SMEs. Moreover, this
chapter assists practitioners, policymakers, CEO, and managers in developing an in-
depth understanding of the terminology that describe the supply chain management
practices, supply chain performance, and critical supply chain management practices
that positively affects the supply chain performance and competitive advantage.

Keywords Supply chain management · Supply chain management practice

implementation · Supply chain performance · Digital technology · SMEs

1 Introduction

As the competition intensified after the unveiling of information and communica-

tion technology (ICT) in the supply chain (SC) field, manufacturing organizations

R. Kumar (B) · M. Gupta

Department of Mechanical Engineering, Motilal Nehru National, Institute of Technology
Allahabad, Prayagraj 211004, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 217
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_11
218 R. Kumar and M. Gupta

have been forced to implement supply chain management practice (SCMP), infor-
mation sharing practice(IS), and ICT to stay competitive in the global race. Orga-
nizations cannot sustain themselves without an efficient, effective, and agile SC in
today’s competitive global market. For an efficient and effective SC, the organization
will have to bring customers, suppliers, logistics activities, financial flow activities,
interorganizational systems, and intraorganizational systems together on a single
platform (Li et al., 2006; Koh et al., 2007; Oswald & Kleinemeier, 2017; Novais et al.,
2019). SCMP such as lean practice and JIT supply practice has guided organizations
to achieve excellence in their business operations (Bayraktar et al., 2009). Effec-
tive coordination and collaboration of trading partners (supplier, customer, distrib-
utor, and IT partners) have improved the organizations’ organizational performance
(OP). Information system such as Enterprise Resources Planning (ERP), Material
Requirement Planning (MRP), and Manufacturing Resources Planning (MRPII) has
attempted to successfully integrate many business processes for fast, accurate, real-
time data and online access to data (Bayraktar et al., 2009). With the increasing use of
integrated information systems and enabling technologies, it has become possible to
create a seamless supply chain linking suppliers to customers to reduce the supplier’s
inefficient performance, unpredictable customers’ demands, and uncertain business
environment (Novais et al., 2019; Oswald & Kleinemeier, 2017; Yipeng, 2011).
Therefore, the knowledge of the SCMP, IS practices, and ICT has become an essen-
tial requirement for staying competitive in the global race (Li et al., 2006; Koh et al.,
2007; Bayraktar et al., 2009; Gawankar et al., 2017; Novais et al., 2019). Successful
and effective implementation of the SCM concept can be found in large manufactures
industries, and it is found that the SCM practices have a significant positive impact
on SCP, OP, and ultimately on competitive advantage (Li et al., 2006; Gharakhani
et al., 2012; Kumar and Nambirajan, 2013; Saber et al., 2014).
There are several definitions provided for SCMP. SCMP includes a set of activi-
ties for effective and successful integration of suppliers, manufacturers, distributors,
and customers for improving the SCP and OP in a cohesive and high-performance
business model (Li et al., 2006; Jabbour et al., 2011). SCMP is also defined as a
set of practices that seek to enhance competitive performance by close integration
of the internal functions within an organization and efficiently linking them with
the external operations of suppliers, customers, and other trading partners (Li et al.,
2006). Successful Implementation of SCMP is crucial for the effective integration of
internal functions within organizations and the effective integration of supply chain
traders. Some organizations have realized the importance of implementing SCMP.
Still, they usually do not know exactly what to implement due to a shallow under-
standing of what organizes complete SCM practices set (Bayraktar et al., 2009;
Li et al., 2005; Oswald & Kleinemeier, 2017). Therefore need arises to manage
comprehensive SCM practices set and propose a framework for successful imple-
mentation of SCMP. All in all, previous works (Li et al., 2005, 2006; Thakkar, 2008a,
2009; Cuthbertson & Piotrowicz, 2008; Bayraktar et al., 2009) have shown a clear
interest in examining the impact of SCMP, IS practice, and ICT on OP and organi-
zational performance. Many authors have used various nomenclatures that describe
SCMP and SCP worldwide. It results in a confused understanding of the fundamental
Supply Chain Management Practice and Supply Chain Performance … 219

concepts of the area. In this chapter, we present a conceptual systematization of the

nomenclature used by authors for SCMP. In this context, this chapter presents the
conceptual systematization deeming the whole supply chain grouping 45 practices
found in supply chain literature with different nomenclature in a set of 14 significant
concepts/practices and further proposes a framework for SCMP implementation in
Indian manufacturing SMEs successfully.

2 Review Methodology

An extensive literature review has been used to identify and categorize the nomen-
clature used for describing SCMP. Related publications are obtained through an
extensive online search, which aims to collect, organize, and integrate existing
SCMP terminologies and their impact on SCP and organizational performance.
Collected publications also identify SCMP implementation issues in Indian SMEs.
Collected papers reveal the relationship between SCMP, operational and organiza-
tional performance, and the crucial factors responsible for SCMP implementation
within SMEs. The explained research in this chapter has been mostly carried out via
mobilizing various types of literature review. Considering the following electronic
database: Elsevier(Science direct), Emerald Insight, Scopus and Springer over the
2006–2018 time frame period, including scientific papers, journals, articles, govern-
ment reports, business reports from companies which allow the access to works that
were published in the most important international periodicals.

3 Background: SCMP and SCP Measures

The literature is available on the dimensions of SCM practices from a variety of

perspectives. Li (2006) have developed six empirically validated and reliable dimen-
sions that comprise strategic supplier partnership, customer relationship, information
sharing, information quality, internal lean practices, and postponement (Li et al.,
2006). Gawankar (2017) investigated the relationship between SCMP (strategic
supplier partnership, customer relationship practices, information sharing, informa-
tion quality, postponement practice, lean retailing strategies), and SCP measures.
The authors have used different nomenclature for lean retailing strategies with the
same meaning. Therefore need arises to categorize all the different nomenclature into
corresponding practices. Gawankar (2017) concluded from the study that effective
management of supplier partnership, information sharing (IS), and quality of infor-
mation sharing can positively impact SCP and organization accomplishment toward
sustainability. The author also reveals that internal lean practice is not effectively
implemented across the manufacturing sector (Gawankar et al., 2017). Srinivasan
(2011) also investigated that there is a positive relationship between partnership
quality and SCP. The author also reveals that partnership quality’s positive effect
220 R. Kumar and M. Gupta

weakens when there is a greater level of uncertainty and supply risk. On the other
hand, the positive effect of partnership quality on SCP is more potent when firms face
greater demand-side risk. Saber (2014) investigated that strategic supplier partner-
ship is not effectively implemented in Iran industrial area. The impact of SCMP on
strategic supplier partnership is 0.087, which is very low. Still, SCMP with strategic
supplier partnership, customer relationship, level of information sharing, quality of
information sharing, and procrastination positively impact competitive advantage
measures price, product’s quality, innovation, and marketing time.
Various SCMPs have been implemented to improve SCP and gain a compet-
itive advantage in various industries and countries. Successful implementation of
SCMP varies from organization to organization. Hence, SCMP constructs should be
validated before implementing it in any country. On the other hand, Al-Tit (2017)
revealed that the impact of organizational culture is more significant than SCMP on
organizational performance. So the organization’s managers should focus on orga-
nizational culture rather than focusing on SCMP. Both managers and researchers
should pay more attention to any individual’s beliefs, values, and assumptions within
organizations. Jabbour et al. (2011) revealed by presenting four constructs from prin-
cipal component analysis. In this study, the author had proposed six constructs of
SCMP (supply chain integration, information sharing, customer service manage-
ment, customer relationship, supplier relationship, and postponement) and gets four
principal factors extracted from principal component analysis. These four principal
factors are SC integration for production planning and control support, information
sharing about products and targeting strategies, strategic relationship with customer
and supplier, and support customer order. Supply chain integration for production
planning and control consist of supply chain integration constructs. Information
sharing about products and targeting strategies consists of customer support manage-
ment, customer relationship, and supplier relationship. Strategic relationship with
customer and supplier combines constructs of customer and supplier relationships
and supports customer order combining constructs of supply chain integration, post-
ponement, and information sharing. Whereas Koh (2007) have developed two major
constructs extracted from principal component analysis. In this study, Koh have
proposed twelve constructs of SCMP (close partnership with customers, close part-
nership with suppliers, JIT Supply, few suppliers, holding safety stock, strategic
planning, supply chain benchmarking, many suppliers, subcontracting, outsourcing,
3PL and E-procurement) and get two principal factors extracted from principal factor
analysis. These two factors are strategic collaboration and lean practices (SCLP) and
outsourcing and multisuppliers (OMS). Koh also investigate that OMS and SCLP
have a direct positive and significant impact on SMEs’ operational performance. In
contrast, OMS and SCLP do not have a direct positive and significant impact on
the organizational performance of SMEs. Jabbour et al. (2011) Came up with the
effect of company size, company position, and bargaining power on SCMP adop-
tion, and the author revealed that the company size, company position, and company
bargaining power influences the adoption of SCMP within the organization. Despite
the significant and positive influence of the SCMP over operational and organiza-
tional performance, it is possible to get a great diversity of practices proposed by the
Supply Chain Management Practice and Supply Chain Performance … 221

researchers with different nomenclature but with considerable conceptual similarity.

It can result in a confused understanding of the constructs and concepts. So conceptual
systemization of different terminology is a requisite step for a better understanding
of concepts. In this chapter, we aim to provide standardization of the terminology
and classify them into significant concepts. We have grouped 45 practices found
in the supply chain literature into 14 significant concepts/practices validated by the
practitioners and extensive literature. These fourteen constructs are crucial for an
effective, efficient, and agile supply chain and cover all SC aspects. A conceptual
systemization of terminologies that characterize SCMP is explained in Table 1.
Organizations cannot sustain themselves without an efficient, effective, and agile
supply chain in today’s competitive global market. For this purpose, all these fourteen
practices should be successfully implemented to enhance OP and gain a competi-
tive advantage. Thakkar (2008b, 2009) suggested that organizations should improve
the organizational culture and view of the employee for the efficient application of
the SCMP. Thakkar further states that strong commitment from top management is
crucial to know where SCMP has the most significant impact on SMEs. For the effec-
tive and efficient application of SCMP, organization readiness is very crucial (Oswald
& Kleinemeier, 2017). Organization readiness comprises organization infrastructure,
organization culture, and availability of human resources with the IT system. Many
studies (Varma, 2006; Thakkar, 2008b, 2009) reveal the causes for inefficient, slow,
and invisible supply chain activities are inadequate SCM applications in SMEs, lack
of supply chain integration, lack of collaborative environment, immature buyer–
supplier relationships, inadequate information sharing between SC partners, inade-
quate supply chain framework, lack of IT tools usage, and poor strategic vision. Due
to this, it has become very challenging for SMEs to compete in the global world,
and it is also necessary to propose a framework for the successful implementation
of SCMP in Indian SMEs.

4 Framework for SCMP Implementation in Indian

Manufacturing SMEs

The framework for the successful implementation is classified into five parts: features
of SCMP applications in SMEs and selection of appropriate SCMP, self-assessment
of SMEs and strategic planning, development of an appropriate strategy, implemen-
tation issue, and development of the business model. The proposed framework is
shown in Fig. 1.
222 R. Kumar and M. Gupta

Table 1 A conceptual systemization of terminologies that characterize SCMP

Classification of major concept Other terminologies
Customer–supplier relationship 1. Strategic supplier partnership (Gawankar et al., 2017)
2. Customer relationship practices (Gawankar et al., 2017)
3. Partnership quality (Srinivasan et al., 2011)
4. Strategic supplier partnership (Saber et al., 2014)
5. Customer relationship (Al-tit, 2017)
6. Supplier relationship (Al-tit, 2017)
7. Supply chain integration (Jabbour et al., 2011)
8. Strategic relationship with customer and supplier (Jabbour
et al., 2011)
9. Close partnership with suppliers and close partnership
with customers (Koh et al., 2007)
10. Strategic supplier partnership and customer relationship
(Li et al., 2005)
11. Partner relationship (Jabbour et al., 2011)
12. Long-term relationship (Paulraj et al., 2006)
13. Strategic collaboration and lean practices (Koh et al.,
2007)
Level of information sharing (IS) 1. Information sharing (Gawankar et al., 2017)
2. Level of information sharing (Saber et al., 2014)
3. Information sharing about products and strategies
(Jabbour et al., 2011)
4. Information sharing (Li et al., 2005)
5. Information sharing (Chong et al., 2011)
6. Information technology (Ali et al., 2020)
7. Information integration (Paulraj et al., 2006)
8. Interorganization systems (Paulraj et al., 2006)
Quality of information sharing 1. Information quality (Gawankar et al., 2017)
(QOIS) 2. Internal operation (Chong et al., 2011)
3. Quality of information sharing (Saber et al., 2014)
4. Information quality (Li et al., 2005)
6. Adaption of information technology (Ali et al., 2020)
7. Two-way communication (Paulraj et al., 2006)
Postponement (PP) 1. Postponement practices (Gawankar et al., 2017)
2. Procrastination (Saber et al., 2014)
3. Postponement (Jabbour et al., 2011)
4. Postponement (Li et al., 2005)
Internal lean practices (ILP) 1. Lean retailing strategies (Gawankar et al., 2017)
2. Internal lean practices (Li et al., 2005)
E-procurement 1. Strategic purchasing (Paulraj et al., 2006)
2. E-procurement (Bayraktar et al., 2009)
3. E-procurement (Koh et al., 2007)
Safety stock 1. Safety stock (Koh et al., 2007; Gawankar et al., 2017)
Outsourcing 1. Strategic outsourcing (Koh et al., 2007; Gawankar et al.,
2017)
2. Outsourcing and multisuppliers (Koh et al., 2007)
(continued)
Supply Chain Management Practice and Supply Chain Performance … 223

Table 1 (continued)
Classification of major concept Other terminologies
3PL 1. E-business (Koh et al., 2007; Gawankar et al., 2017)
Strategic planning 1. Strategic planning (Koh et al., 2007; Gawankar et al.,
2017)
Subcontracting 1. Subcontracting or E-business (Koh et al., 2007; Gawankar
et al., 2017)
Many suppliers/Few suppliers 1. Supplier selection (Li et al., 2006; Koh et al., 2007;
Gawankar et al., 2017)
TQM 1. TQM (Koh et al., 2007; Gawankar et al., 2017)
JIT supply 1. JIT supply system (Bayraktar et al., 2009)

Fig. 1 Proposed framework for SCMP implementation

4.1 Features of SCMP Applications in SMEs and Selection

of Appropriate SCMP

SMEs have a flat structure rather than a hierarchical structure (Thakkar, 2008b).
Implementation and management change are more straight forward in SMEs than
large organizations. The owner has the full authority to make the decisions and make
the long-term partnership with suppliers and customers. Essential success factors
and features of the SCMP implementation in SMEs are discussed below in Table 2.
Other characteristics of SCMP with definitions are provided in Table 3.
224 R. Kumar and M. Gupta

Table 2 Features of SCMP application in SMEs

Flat structure Decisions can be changed at any moment by
the owner. It ensures that top management
takes part in every activity, and it is one of the
requirements while implementing SCMP in
SMEs
Ability to innovate SCMP will render the ability to innovate and
develop new products at a faster rate
Difference between actual and forecast demand A collaborative working environment can
assure better information sharing, trust, and
transparency and reduce the difference
between the forecast and actual demand
Transparency Long-term relationship and a collaborative
working environment assure high transparency
within SMEs
Agility (speed) The use of IT with timely information sharing
makes SMEs agile
Collaborative Long-term partnership with customer and
environment supplier provides a collaborative environment
Accuracy Quality of information sharing can provide
accurate and real-time data

Li et al. (2006) have developed six empirically validated and reliable dimen-
sions that comprise strategic supplier partnership, customer relationship, informa-
tion sharing, information quality, internal lean practices, and postponement. These
validated and reliable dimensions are discussed below with their definition, and
benefits of implementation and supporting literature are defined in Table 4. Many
other SCMPs (E-procurement, strategic planning, JIT supply system, and TQM)
are developed to enable coordination and collaboration with suppliers. Very often,
an E-procurement tool also interfaces with ERP to automate many purchasing and
payment tasks services.
Outsourcing and subcontracting are associated with a type of service of activi-
ties by an external party to accomplish related functions desired to be rendered or
managed by the purchasing organizations (Hong et al., 2018). Third-party logistics
(3PL) is a type of service of multiple activities by an external party to accomplish
related functions that are desired to be rendered or managed by the purchasing organi-
zations (Koh et al., 2007; Byraktar et al., 2009). Hence, 14 validated and appropriate
SCMPs are explained above and have a positive impact on nine SCP. SCP is referred
to as the performance of operations that indicate SCM’s overall efficiency and effec-
tiveness. Nine major dimensions of SCP are proposed based on the studies presented
in this chapter. Traditional measure: SC flexibility, SC integration, responsiveness to
customer, efficiency, quality, product innovation, and market performance and rela-
tionship measure: relationship quality and supplier performance (output measures)
(Li et al., 2006; Kumar and Nambirajan, 2013; Gawankar et al., 2017; Hong et al.,
2018).
Supply Chain Management Practice and Supply Chain Performance … 225

Table 3 Benefits of SCM practice

SCMP Definition Benefits
Strategic supplier–customer Strategic supplier–customer Collaborative environment
Relationship partnership represents the Visibility from supplier to
long-term relationship with customer
customer and supplier Working on low inventory
Reduced lead time in
production
Information sharing Information sharing is defined as Forecasting accuracy
the extent to which critical and Better understanding of
proprietary information is customer need
communicated to trading partners Faster response to market
changes
Information quality It relates to the correctness, Saving of time
timeliness, adequacy, reliability, Better operational
and completeness of information efficiency
exchanged between trading Better resources planning
partners Provide agility to the
system
Postponement It is defined as the practice of Cost saving
moving forward one or more Assure product verities
operations or activities Customize product
Internal lean practices It is defined as the practice of Cost saving
eliminating waste (cost, time, Time saving
etc.) in a manufacturing system
characterized by reduced set-up
times, small lot sizes, and pull
production

Table 4 Literature review on SCM practice

SCM practice Supporting literature
Customer–supplier relationship (Li et al., 2006; Jabbour et al., 2011; Li et al., 2005;
Okoth, 2016)
Information sharing (IS) (Gawankar et al., 2017; Saber et al., 2014; Chong
et al., 2011)
Quality of information sharing (QOIS) (Jabbour et al., 2011; Li et al., 2005; Okoth, 2016;
Chong et al., 2011)
Postponement (PP) (Gawankar et al., 2017; Li et al., 2006; Saber et al.,
2014)
Internal lean practices (ILP) (Gawankar et al., 2017; Li et al., 2006)

4.2 Self-assessment of SMEs and Strategic Planning

In this step, the organization’s owner reviews the current organizational culture and
structure and makes some valuable decisions (Srinivasan et al., 2011). The owner
226 R. Kumar and M. Gupta

chooses whether the organization will integrate all selected SCMP with the conven-
tional process or not. For successful implementation of SCMP, the owner addresses
a list of the requirements of the SCMP based on organization culture and structure
requirements and seeks to transform the current organizational culture and structure
into SCMP-based organization culture and structure. Developing a strategy for SCM
also becomes a complicated task in itself. The primary issue for top management is
developing an appropriate strategy for the successful implementation of SCMP.

4.3 Development of an Appropriate Strategy

Before developing an appropriate strategy, top management should assess their

organization and supply chain capabilities. This step is a prerequisite because a
firm’s SCM strategy would depend on its organization and supply chain capabili-
ties (Thakkar et al., 2009). A self-assessment model based on organization structure
and governance, organization culture, IS practice, ICT, IT usage, transportation, and
warehousing system can be used (Thakkar et al., 2012; Oswald & Kleinemeier, 2017).
The strategy must focus on four key elements: people (employees, supplier, customer,
IT team, and 3PL) (Varma, 2006), organization (culture, leadership, structure, and
governance, top management support) (Oswald & Kleinemeier, 2017), systems and
technological processes (IS practice, ICT, digital technology (Bayraktar et al., 2009).
Type of product, product life cycle, type of industry, and dynamic environment mainly
affect the formulation process of SCM strategy (Singh & Garg, 2008; Jabbour et al.,
2011). Top management should have these four factors in mind while assessing the
supply chain capabilities and formulating an appropriate strategy.

4.4 Implementation Issue of SCMP

Several factors found in the literature review mainly affect SCMP and performance
measures. Some of these factors are discussed below, and critical factors for SCMP
implementation in SMEs are shown in Fig. 2. The extensive literature review found
that company size influences the adoption of supply chain management practice.
That is, the larger the size of the company, the higher degree of adoption of SCMP.
The degree of adoption of SCMP in large companies is higher than the small- and
medium-sized enterprises (Jabbour et al., 2011). Hence, the appropriate SCM prac-
tices proposed for large organizations should be implemented within SMEs. Jabbour
et al., (2011) reveal that the company position and bargaining power also influ-
ence the adoption of SCMP within organizations. The adoption of SCMP in the
organization that assembles components differs from the organization that supplies
components. The organization that assembles components or supplies raw material
has different degrees of adoption of SCMP in comparison to the organizations that
are distributors or retailers, or component suppliers. Hence, the SCMP proposed for
Supply Chain Management Practice and Supply Chain Performance … 227

Fig. 2 Key factors for SCMP

organizations that supply raw materials should not be implemented within organi-
zations that are distributors or retailers, or component suppliers (Thakkar, 2008b;
Jabbour et al., 2011). Al-Tit (2017) revealed that the impact of organizational culture
is more significant than SCM practice on organizational performance. So the orga-
nization’s managers should focus on organizational culture rather than focusing on
SCMP. Both managers and researchers should pay more attention to the beliefs,
values, and assumption held by any individual within organizations.
Bayraktar (2009) investigated that IS practices have a positive impact on SCMP
and OP. For effective implementation of SCMP within SMEs, information practices
(EDI, ERP, and MRP, MRPII) should be successfully implemented, as shown in
Fig. 2. There are some inhibiting factors in the adoption of IS practice, as shown
in Fig. 2. Ali (2020) examined that ICT offers valuable new ways of engaging and
collaborating with trading partners of SMEs (Novais et al., 2019). Ali also revealed
that a higher digitization level influences firm performance in terms of cash flow
and material flow. The implementation of SCMP in SMEs needs investment, and
suppliers to SMEs are generally small fragmented players that carry less motivation
for investment (Thakkar, 2008a, 2008b), as shown in Fig. 2. Supply chain partners
should be prepared to adopt ICT and IS practices for better and real-time information
exchange on inventory status and delivery schedules. Real-time and accurate data
228 R. Kumar and M. Gupta

can help top management to visualize plans and strategy on more realistic ground
(Novais et al., 2019).
The government can improve ICT awareness by conducting conferences,
publishing articles, and workshop on information and communication technology.
The adoption of ICT can change the way organizations design, compete, and engage
with trading partners. Adopting digital technology has a positive impact on logistics
activity, product development, financial flow, and marketing activity (Novais et al.,
2019). Adoption of ICT within an organization influences the implementation of
SCMP and its impact on ultimately operational and organizational performance. By
using ICT, supply chain partners can access information about product flow, produc-
tion system schedule, and financial flow at any time from any place (Bayraktar et al.,
2009; Kamble et al., 2019; Lian et al., 2014. Therefore, ICT has a positive impact on
SCMP such as information sharing and the quality of information sharing. Hence,
IS practice and ICT can play a vital role in the SC operations field.
Before developing a business model for SMEs, it should be in mind that company
size, position, bargaining power, structure, governance, and culture, with effective
digitalization, and successful implementation of IS practice can significantly influ-
ence SCMP and SCP. Successful implementation of IS practice and ICT is essential
for effective implementation of SCM practice within SMEs.

5 Discussion and Conclusion

As we know, SCM is essential for the smooth and effective integration of trading
partners and effective management of logistics activities. Therefore, a set of SCM
practices have been implemented worldwide without measuring its real impact on
SCP and organizational performance. Thus, a significant set of recent research has
attempted to find the impact of SCMP on SCP and organizational performance world-
wide. However, these attempts have been hampered by the diversity and lack of
standardization of the used nomenclature. The systemization of terminology and
constructs used for SCM is very necessary for improving the measurement of
constructs. This chapter tried to present a conceptual systematization of the used
nomenclature that characterizes SCMP and SCP worldwide and secondly proposed
a framework for successful SCMP implementation.
Strategic supplier partnership and customer relationship practices can be grouped
as customer–supplier relationship. Many authors (Al-tit, 2017; Gawankar et al., 2017;
Jabbour et al., 2011; Li et al., 2006; Saber et al., 2014; Srinivasan et al., 2011)
have used various terminologies for customer–supplier relationships. The conceptual
systemization of terminology provides 14 SCMP proposed in various studies, organi-
zations, and countries. These 14 constructs cover mostly all the practices developed
in the field of SCM. These 14 proposed SCM practices are grouped as a signifi-
cant concept: customer–supplier relationship, information sharing, quality of infor-
mation sharing, postponement, internal lean practices, E-procurement, safety stock,
outsourcing, subcontracting, third-party logistics(3PL), many supplier/few suppliers,
Supply Chain Management Practice and Supply Chain Performance … 229

strategic planning, total quality management (TQM) and Just-in-Time supply system.
Many researchers have evaluated the direct impact of these proposed SCM practices
on SCP and ultimately on organizational performance. Some of them have evaluated
the direct effect of SCM practice on organizational performance through compet-
itive advantage. It is found that successful and effective implementation of SCMP
within an organization enhance OP and, ultimately, organizational performance. Five
proposed SCM practices successfully implemented within the organizations in many
countries are customer–supplier relationship, information sharing, quality of infor-
mation sharing, postponement, and internal lean practices. These SCM practices
significantly influence SCP, referred to as supplier performance, supply chain inte-
gration, responsiveness to the customer, supply chain flexibility, innovation, market
performance, partnership quality, and operations efficiency. These five constructs
should be successfully implemented within the SMEs for the effective, efficient, and
agile supply chain.
The proposed framework assists the policymakers and managers in realizing the
feature of SCMP application within SMEs. The policymakers and managers should
be knowledgeable of fourteen validated constructs and select appropriate constructs
that have the most significant impact on SCP. The policymakers and managers
should be knowledgeable of the operation area they want to improve. According
to the selected SCMP, policymakers and managers can develop a strategy that must
focus on four key elements: people (employees, supplier, customer, IT team, and
3PL), organization (culture, leadership, structure, and governance, top management
support), systems and technological processes (IS practice, ICT, digital technology.
There are factors such as organization size, organization position, and bargaining
power of organizations that also affect the successful and effective implementation
of SCMP and SCP measures. The larger the size of the company, the higher degree
of adoption of SCMP. The degree of adoption of SCMP in large companies is higher
than the small—and medium-sized enterprises (SMEs). Successful implementation
of IS practice and ICT is essential for effective implementation of SCMP within
SMEs while focusing on four key elements: people (employees, supplier, customer,
IT team, and 3PL), organization (culture, leadership, structure, and governance, top
management support), systems and technological processes (IS practice, ICT, digital
technology.
A framework for the successful implementation of SCMP is proposed in this
chapter. Successful implementation of SCMP with IS practice and ICT will provide
an efficient, effective, flexible, and agile supply chain. Hence, this chapter assists
practitioners, policymakers, CEO, and managers to use the proposed framework for
the effective, efficient, and agile supply chain for Indian SMEs.

Acknowledgements The reviewers’ comments on the earlier format of this chapter are much
appreciated. Without their valuable comments, the paper should not have reached this shape. Their
comments increased our thinking and understanding capability as we started the project from scratch.
230 R. Kumar and M. Gupta

References

Al-Tit, A. A. (2017). Factors affecting the organizational performance of manufacturing firms.

International Journal of Engineering Business Management, 9, 1–9.
Ali, Z., Gongbing, B., & Mehreen, A. (2020). Does supply chain finance improve SMEs perfor-
mance? The moderating role of trade digitization. Business Process Management Journal,
26(1), 150–167.
Bayraktar, E., Demirbag, M., Koh, S. L., Tatoglu, E., & Zaim, H. (2009). A causal analysis of the
impact of information systems and supply chain management practices on operational perfor-
mance: Evidence from manufacturing SMEs in Turkey. International Journal of Production
Economics, 122(1), 133–149.
Chong, A. Y. L., Chan, F. T. S., Ooi, K. B., & Sim, J. J. (2011). Can Malaysian firms improve
organizational/innovation performance via SCM? Industrial Management Data Systems, 111(3),
410–431.
Cuthbertson, R., & Piotrowicz, W. (2008). Supply chain best practices–Identification and cate-
gorisation of measures and benefits. International Journal of Productivity and Performance
Management, 57(5), 389–404.
de Sousa Jabbour, A. B. L., Alves Filho, A. G., Viana, A. B. N., & Jabbour, C. J. C. (2011).
Factors affecting the adoption of supply chain management practices: Evidence from the Brazilian
electro-electronic sector. IIMB Management Review, 23(4), 208–222.
de Sousa, L., Jabbour, A. B., Viana, A. B. N., & Jabbour, C. J. C. (2011). Measuring supply chain
management practices. Measuring Business Excellence, 15(2), 18–31.
Gawankar, S. A., Kamble, S., & Raut, R. (2017). An investigation of the relationship between
supply chain management practices (SCMP) on supply chain performance measurement (SCPM)
of Indian retail chain using SEM. Benchmarking: An International Journal, 24(1), 257–295
Gharakhani, D., Mavi, R. K., & Hamidi, N. (2012). Impact of supply chain management practices
on innovation and organizational performance in Iranian companies. African Journal of Business
Management, 6(19), 5939–5949.
Hong, J., Zhang, Y., & Ding, M. (2018). Sustainable supply chain management practices supply
chain dynamic capabilities and enterprise performance. Journal of Cleaner Production, 172,
3508–3519. https://doi.org/10.1016/j.jclepro.2017.06.093
Jabbour, A. B. L D. S., Filho A. G. A., Viana, A. B. N., & Jabbour, C. J. C. (2011). Factors
affecting the adoption of supply chain management practices: Evidence from the Brazilian electro-
electronic sector. IIMB Management Review 23(4), 208–222.
Jabbour, A. B. L. D. S., Filho, A. G. A., Viana, A. B. N., & Jabbour, C. J. C. (2011). Measuring
supply chain management practices. Measuring Business Excellence 15(2), 18–31.
Kamble, S. S., Gunasekaran, A., Parekh, H., & Joshi, S. (2019). Modeling the internet of things
adoption barriers in food retail supply chains. Journal of Retailing and Consumer Services, 48(C),
154–168
Koh, S. L., Demirbag, M., Bayraktar, E., Tatoglu, E., & Zaim, S. (2007). The impact of supply
chain management practices on performance of SMEs. Industrial Management and Data Systems,
107(1), 103–124. https://doi.org/10.1108/02635570710719089
Kumar, C. G., & Nambirajan, T. (2013). The impact of supply chain management components
on supply chain performance and organizational performance: Measuring moderating effects of
supply chain practices clusters. Journal of Contemporary Research in Management, 8(3), 55–68.
Li, S., Rao, S. S., Ragu-Nathan, T. S., & Ragu-Nathan, B. (2005). Development and validation of a
measurement instrument for studying supply chain management practices. Journal of Operations
Management, 23(6), 618–641.
Li, S., Ragu-Nathan, B., Ragu-Nathan, T. S., & Rao, S. S. (2006). The impact of supply chain
management practices on competitive advantage and organizational performance. Omega, 34(2),
107–124.
Supply Chain Management Practice and Supply Chain Performance … 231

Lian, J. W., Yen, D. C., & Wang, Y. T. (2014). An exploratory study to understand the critical
factors affecting the decision to adopt cloud computing in Taiwan hospital. International Journal
of Information Management, 34(1), 28–36. https://doi.org/10.1016/j.ijinfomgt.2013.09.004
Lopes de Sousa Jabbour, A. B., Gomes Alves Filho, A., Backx Noronha Viana, A., & José Chiappetta
Jabbour, C. (2011). “Relationships between company size production system and supply chain:
Evidence from electro-electronics sector in Brazil. Journal of Advances in Management Research,
8(1), 30–52.https://doi.org/10.1108/09727981111129291
Novais, L., Maqueira, J. M., & Ortiz-Bas, Á. (2019). A systematic literature review of cloud
computing use in supply chain integration. Computers and Industrial Engineering, 129, 296–314.
Okoth, O. T., & Ochieng, O. S. (2016). Supply chain quality management practices and performance
of pharmaceutical distributors and wholesalers in Mombasa Kenya. International Journal of
Managerial Studies and Research, 4(8), 94–102.
Oswald, G., & Kleinemeier, M. (2017). In Shaping the digital enterprise.
Paulraj, A., Chen, I. J., & Flynn, J. (2006). Levels of strategic purchasing: Impact on supply
integration and performance. Journal of Purchasing and Supply Management, 12(3), 107–122.
Saber, Z., Bahraami, H. R., & Haery, F. A. (2014). Analysis of the impact of supply chain manage-
ment techniques: A competitive advantage in the market. International Journal of Academic
Research in Economics and Management Sciences, 3(1), 75.
Singh, R. K., Garg, S. K., & Deshmukh, S. G. (2008). Strategy development by SMEs for compet-
itiveness: a review. Benchmarking: An International Journal, 15(5), 525–547. https://doi.org/10.
1108/14635770810903132
Srinivasan, M., Mukherjee, D., & Gaur, A. S. (2011). Buyer—supplier partnership quality and
supply chain performance : Moderating role of risks, and environmental uncertainty q. European
Management Journal Els, 29(4), 260–271. https://doi.org/10.1016/j.emj.2011.02.004
Thakkar, J., Kanda, A., & Deshmukh, S. G. (2008). Interpretive structural modeling (ISM) of
IT-enablers for Indian manufacturing SMEs. Information Management and Computer Security,
16(2), 113–136. https://doi.org/10.1108/09685220810879609
Thakkar, J., Kanda, A., & Deshmukh, S. G. (2008). Supply chain management in SMEs: Devel-
opment of constructs and propositions. Asia Pacific Journal of Marketing and Logistics, 20(1),
97–131. https://doi.org/10.1108/13555850810844896
Thakkar, J., Kanda, A., & Deshmukh, S. G. (2009). Supply chain performance measurement frame-
work for small and medium scale enterprises. Benchmarking: An International Journal, 16(5),
702–723.
Thakkar, J., Kanda, A., & Deshmukh, S. G. (2012). Supply chain issues in Indian manufacturing
SMEs: Insights from six case studies. Journal of Manufacturing Technology Management, 23(5),
634–664.
Varma, S., Wadhwa, S., & Deshmukh, S. G. (2006). Implementing supply chain management in
a firm: Issues and remedies. Asia Pacific Journal of Marketing and Logistics, 18(3), 223–243.
https://doi.org/10.1108/13555850610675670
Yipeng, L. (2011) ‘The impact of “cloud computing” based information sharing on supply chain. In
2011 Fifth International Conference on Management of e-Commerce and e-Government (pp. 173–
175). IEEE. https://doi.org/10.1109/ICMeCG.2011.39
Behavioural Operations Management:
Trends and Insights

S. Yamini

Abstract Human factor plays an important role in impacting the operating system
and the way in which the processes are being performed. Research in operations
management (OM) has of-late shown importance in human decision-making and its
relationships with OM theory along with its practices. The significance of behavioural
operations management (BOM) could be well understood only if the researchers
realize that almost all contexts studied within OM contain people. A rising amount
of research studies in OM have investigated the behavioural underpinnings of opera-
tions and supply chain management. This chapter highlights the significance of role
played by humans in operational outcomes. It also indicates the consequences of
including behavioural factors into various other operational techniques. Economic
models assume people behave rationally in their choices and preferences. However,
the decisions that they make are influenced by their emotions, feelings, social prefer-
ences, cultural influences and so on. This chapter gives a brief review of behavioural
operations management field that is enriched by adopting various behavioural princi-
ples. It also discusses about various research methodologies that are used in the field
of behavioural operations management. Further, a general discussion and blueprint
for future research predicting the trends in future research is also given. The objective
of this chapter is not providing an extensive review to all the fields that are related
to behavioural operations, instead highlight on the recent research in this field that
can used by the OM scholars.

S. Yamini (B)
Faculty of Operations and Analytics, National Institute of Technology Tiruchirappalli,
Tiruchirappalli 620015, Tamil Nadu, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 233
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_12
234 S. Yamini

1 Introduction

1.1 Evolution of OM to BOM

The term “Operation Research” or “Operations Management” has earned value

with changing technology and rising complexities in business world. Operation
research is a decision-making technique that tries to find optimal solutions to manage-
rial problems mathematically. Initially, operations management and organisational
behaviour/human behaviour were two developing fields of study independently over
a period, even though both fields have a lot of commonalties between each other. We
can observe and relate that most of the OM strategies are carried out only by people
and many OB policies have to be followed while doing any operational activities.
Around the turn of the century, human aspects and other behaviour related to them
were studied by OM researchers. Behavioural operations management (BOM) works
under the postulation that decision-makers are self-interested, and they perform in a
completely aware/thoughtful or non-emotional manner (Croson et al., 2013). BOM
postulates that the decisions are encouraged to be expressed in the form of monetary
terms, where decision-makers behave optimally using a definite objective function.
It examines the individual’s behavior at a micro-level.

1.2 Behavioural Operations Management

Behavioural operations is an evolving research area within the field of operations

management, and it has grabbed the attention of many OM researchers. The study of
behavioural operations is developing in the field of research as well, which gathers
references from multiple disciplines. BOM is a multidisciplinary outlet of OM that
considers the effects of human behaviour in the field of operations management.
Various research studies have given different definitions and useful introduction for
the term behavioural operations management. For instance, Bearden and Rapoport
(2005) focus on sequential decision-making problems in operations management
(OM) and examine how people make these types of judgments compared with norma-
tive theory. Loch and Wu (2007) include a wider variety of operations problems and
give direction on how to assimilate more practical behavioural traits into analytical
models, including the behavioural issues such as individual decision-making, social
preferences and cultural differences. Gino and Pisano (2008) give a general overview
and a plan for future research in the field, pondering on happenings in recent study
and conceptual links between various areas of knowledge. Bendoly et al. (2010)
discusses about multiple areas of behavioural research and its contributions to OM.
Croson et al. (2013) defines that BOM ranges from understanding an individual’s
behaviour to eventually supporting in the designing and improving of processes and
supply chains. It studies the behaviour of non-hyper-rational actors and highlights
the research findings that impact the design, management and improvement of the
Behavioural Operations Management: Trends and Insights 235

operational systems. Most research in the field of behavioural operations manage-

ment considers the effects of cognitive biases and heuristics, individual and social
preferences, personal beliefs, perceptions and expectations and social and cultural
standards on decision-making in operations management.
In the last decade, behavioural operations research has captivated many OM
scholars. There were many special issues in top-tier journals (Bendoly et al., 2006;
Gans & Croson, 2008; Eckerd & Bendoly, 2011). There were also many literature
reviews written in the area of BOM. Only few articles follow a systematic approach,
and most of them focuses on the research progress in the broader context of supply
chain management, with the primary focus on methods such as laboratory exper-
iments (Knemeyer & Naylor, 2011) and analytical models (Boudreau et al., 2003;
Loch & Wu, 2007). There was no consideration given to the interrelationship between
these methods. The techniques such as the case study approach, survey-based method
and using secondary data for analysis were also scarce in this field (Fahimnia et al.,
2019). Efforts to widen the scope for this field have been made by many researchers
to include the advancements in this domain in few books and handbooks as well
(Bendoly et al., 2015).

1.3 Methodology Followed for Classifying the Literature

A systematic review procedure is adopted, starting with an elaborate search in the

Google Scholar database to find research studies related to behavioural operations
management using the text and keywords such as behavioural anomalies related to
operations, behavioural studies in supply chain management, human factors affecting
health care industry, agricultural supply chain, project management, cognitive biases
affecting logistics, contracting and buyer–supplier relationship in a supply chain.
The search results are analysed, and the screened papers, depending on relevance to
the topic, are divided based on the subdivisions discussed in this chapter (Table 1).

1.4 Organisation of the Chapter

This chapter is organized as follows. It starts with the introduction to the field of
behavioural operations management highlighting on the evolution of OM to BOM
and the research related to this area. The second section discusses about the impact
of different stakeholders’ behaviour in the supply chain management (SCM). It
covers various topics related to SCM that discusses about the behavioural aspects
of contracting, buyer–supplier relationship, logistics management, inventory sharing
and inventory management. The next section examines the service supply chain and
the behavioural factors impacting it. Section 4 gives an overview of the behavioural
research carried out in various arenas such as health care, agri-food supply chain and
236 S. Yamini

Table 1 List of journals referred in this chapter that publishes research in BOM
Name of the Journal Number of papers published
Production and operations management 12
Management science 11
Journal of supply chain management 9
Manufacturing and service operations management 7
Journal of operations management 6
Journal of business logistics 4
Decision sciences 3
supply chain management: an international journal 2
Journal of business and industrial marketing 2
International journal of production economics 1
International journal of physical distribution and logistics 1
management
Journal of risk assessment and management 1
American economic review 1
International journal services industry management 1
International journal of disaster risk reduction 1
Modern supply chain research and applications 1
International journal of logistics management 1
European journal of operations research 1
Annals of operations research 1
Financial services review 1
International transactions in operational research 1
California management review 1
Chinese journal of management science 1

project management. Finally, the chapter is concluded with the emerging studies in
BOM and some future directions.

2 Supply Chain Management

Behavioural studies in the area of supply chain management cover topics from many
subdivisions in this area. Table 2 shows the topics that are discussed in this section.
Behavioural Operations Management: Trends and Insights 237

Table 2 Summary of behavioural topics discussed in Supply chain management

S. Topic Authors Major Contribution Methodology
No.
1 Supply chain Davis (2015), Impact of risk perspectives Analytical
Contracting Katok and Pavlov of stakeholders, loss modelling
(2013), Ho et al. aversion behaviour and and
(2014), Yi et al. fairness concerns behavioural
(2018), Fehr and experiments
Schmidt (1999),
Bolton and
Ockenfels (2000)
2 Buyer–supplier Ireland and Webb Effect of power and trust Survey and
relationship (2007), Terpend level between buyers and behavioural
and Ashenbaum suppliers, usage of experiments
(2012), Pulles behavioural theories
et al. (2014),
Brito and Miguel
(2017), Aral et al.
(2018),
Kaufmann et al.
(2018)
3 Logistics Management Mentzer and Flint Examining the Quantitative
(1997), assumptions of traditional modelling
Wallenburg et al. inventory models and the
(2011) influence of trust,
trustworthiness and
fairness in logistics related
activities
4 Information Sharing Bolton et al. Quality of information Game
(2004), shared, moral hazard and Theory,
Ebrahim-Khanjari analysing the demand Behavioural
et al. (2012), forecasts experiments
Spiliotopoulou
et al. (2016),
Cantor and
Macdonald,
(2009); Inderfurth
et al. (2013),
(continued)

2.1 Supply Chain Contracting

The behavioural issues focused while discussing supply chain contracting studies
are risk attitudes of decision-makers (Chen et al., 2014) and the findings from
prospect theory (Kahneman and Tversky, 2013). Davis (2015) analysed the impact
of risk aversion and loss aversion behaviour on the decisions taken in supply chain
contracts. Other behavioural issues discussed related to supply chain include the
fairness concerns (Katok & Pavlov, 2013; Ho et al., 2014; Yi et al., 2018), which
238 S. Yamini

Table 2 (continued)
S. Topic Authors Major Contribution Methodology
No.
5 Inventory Management Schweitzer and To understand the Quantitative
Cachon (2000), behavioural anomalies in Modelling,
Agrawal and inventory decision-making, Game
Seshadri (2000), impact of various cognitive Theory,
(Kremer et al., biases in ordering Behavioural
2014), Ho et al., behaviour experiments
(2010), Moritz
et al., (2014),
(Ren & Croson,
2013), (Chen
et al., 2013),
(Long & Nasiry,
2015), Bolton
et al. (2012);
Croson et al.
(2013); (Tokar
Asaad, 2015)

uses inequality aversion models for investigating the related topics (Fehr & Schmidt,
1999; Bolton & Ockenfels, 2000). The most widely accepted research methods in
this area of research are analytical modelling and behavioural experiments.

2.2 Buyer–Supplier Relationships

Behavioural lessons on buyer–supplier relations are an important component in the

area of supply chain management. Some of the emerging topics include power equi-
librium and trust level between buyers and suppliers (Ireland & Webb, 2007; Terpend
& Ashenbaum, 2012; Pulles et al., 2014; Brito & Miguel, 2017; Aral et al., 2018;
Kaufmann et al., 2018). Other established areas include integrating supply chain
members (Enz & Lambert, 2015; Thornton et al., 2016), supplier development and
knowledge enrichment (Kim et al., 2015; Preston et al., 2017). There are many
behavioural theories such as social network theory, social exchange theory, social
capital theory and transaction cost economics that discusses about the buyer–supplier
relationship. Most of these studies make use of survey and behavioural experiments
as a research method.
Behavioural Operations Management: Trends and Insights 239

2.3 Logistics Management

Logistics can generally be recognized by its importance on optimizing business

actions related to the efficient flow and storage of inventory, goods and services
throughout the supply chain. While logistics management is a well-explored area in
the operations management literature, it has often disregarded the impact of human
behaviour. The mathematical models established in logistics and SCM typically make
very obvious as well as implied assumptions about human behaviour. The accuracy
of these assumptions is often indefinite and sometimes fairly uncertain. Only few of
these models have been empirically verified which would offer a signal about the
predictive power of these models or show when their forecasts do not hold.
The everyday need of logistics and the enormous amount of human interface in
SCM suggests that the empirical methods should be at the front pitch of logistics
research. Therefore, there is great possibility for realizing some important benefits
to exercise and building rich theory from such efforts. Laboratory research is not a
frequently used research method in logistics, (Mentzer & Flint, 1997), and hence, the
occurrence of behavioural experiments is also very rare. The results of behavioural
experiments, surveys and observations help us in understanding the dealings between
suppliers and customers. They also provide a comprehensive and precise information
about the behavioural issues that allows managers to transfer product and service
agreements efficiently with supply chain partners. If it gets input on how decision-
makers in partner firms respond to a particular situation, the company could plan and
act strategically to reach the favourable outcome.
Few research studies have analysed some topics from behavioural logistics
management area (Tokar, 2010). Some of them include the influence of trust, trust-
worthiness and fairness in logistics outsourcing (Wallenburg et al., 2011), reckless
decision-making in humanitarian logistics operations (Tomasini & Van Wassenhove,
2009) and behavioural studies related to congestions and traffic systems.

2.4 Information Sharing

The research insights from behavioural supply chain management also cover how
information sharing among different stages of supply chain can impact the prof-
itability of a supply chain. Some of the behavioural investigations discuss about the
role of trust and trustworthiness in the quality of information shared (Özer et al.,
2011). There are also few studies that analyse the demand forecasts that is common
among the supply chain players (Bolton et al., 2004; Ebrahim-Khanjari et al., 2012;
Spiliotopoulou et al., 2016). Further, there are many studies to explore the behavioural
impact of information sharing on overall supply chain performance and coordination
(Cantor and Macdonald, 2009; Inderfurth et al., 2013) and supply chain partnerships
and associations with external participants. Game theory is an extensively adopted
240 S. Yamini

tool in the analytical modelling category, and there are also few behavioural exper-
iments conducted to understand the problems related to information sharing in a
supply chain.

2.5 Inventory Management

Majority of the articles published in this specialization are related to newsvendor

model, wherein the ordering behaviour of retailers is investigated. The number of
research articles published in this field is increasing in the recent past (Yamini,
2020). Figure 1 shows the chronological arrangement of the experimental work
carried out in the newsvendor problem. One of the classical articles in this field of
research is given by Schweitzer and Cachon (2000), who considered a number of
behavioural anomalies in ordering decisions such as the risk attitudes of players, loss

Fig. 1 Chronological
classification of
experimental studies of
newsvendor problem
Behavioural Operations Management: Trends and Insights 241

aversion behaviour of stakeholders involved, anchoring and insufficient modifica-

tions in different profit settings of newsvendor problem. Other newsvendor-related
studies examined the behavioural representations such as expected utility maximiza-
tion (Agrawal & Seshadri, 2000; Kremer, et al. 2014), bounded rationality (Dan
Ariely, 2008), pull-to-centre effect (Ho et al., 2010), cognitive reflection (Moritz
et al., 2014), overconfidence (Ren, and Croson, 2013), mental accounting (Chen
et al., 2013) and prospect theory (Long & Nasiry, 2015) to describe the ordering
biases exhibited by decision-makers in many settings.
The individual heterogeneity of the decision-maker is a major determinant of the
ordering behaviour. It has been tested in various circumstances, where the behaviour
of managers is compared with the behaviour of students to see if both the samples
order in a similar pattern (Bolton et al., 2012). Next the influence of gender differ-
ences on the ordering behaviour is observed where the difference between ordering
pattern of female versus male is noted. The cultural differences between Chinese
and Americans are also observed to understand its impact on the ordering decisions.
Another class of literature looks at the behavioural roots of the bullwhip effect in
the supply chain to mitigate and try to resolve the behavioural risks involved in it
(Croson et al., 2013; Tokar Asaad, 2015).

3 Service Supply Chain

Behavioural economics is an important modification made to the traditional

economics which helps to make the academic research better serve practice (Dan
Ariely, 2008). As service heavily relies upon human participation (Boshoff & Leong,
1998; Sengupta et al., 2006), it is essential to study the impact of human’s behaviour
on the conventional service settings (Bendoly et al., 2010). Service supply chain
(SCC) is defined as a network of suppliers, consumers, service providers who trans-
form resources into core services and then deliver these services to customers (Balta-
cioglu et al., 2007). Conventional research in operations management (OM) focused
on providing tools and techniques to help agents who make tactical and operational
decisions and disregard the results which lead to behavioural biases. Due to the
characteristics of service industry, such as being intangible and heterogeneous, the
research in service supply chain has varied decision-making circumstances. Baindur
and Macário (2013) discusses about the case of lunch box delivery system organized
by the Mumbai dabbawalas. They have shown that an informal service division was
competent enough to develop an urban logistics system that can be afforded by the
middle-class people of Mumbai. Further in the recent years, many researchers focus
on cross-disciplinary research from a behavioural operations perspective.
242 S. Yamini

3.1 Classifications of Service Supply Chain

Service supply chain consists of service provider (SP) and service integrator (SI) who
usually have strong control and outsource the functional services to SPs in order to
maintain competitive advantages (Choy et al., 2007). SI also integrates these func-
tional facilities into integrated service solutions for end customers. Many researchers
have explored the behavioural aspects from the perspective of service supply chain
structure and enhanced the research findings in this area. The behavioural findings
from a variety of industries including advertising, consulting, call-centre and profes-
sional services are also well explored (Liu et al., 2019). From the product perspec-
tive, Wang et al. (2015) has divided the service supply chain broadly into service
and product. In case of service, the merchandise is only the service. For example,
health check-up in healthcare industry. In case of product, the merchandise is the
combination of a physical product and intangible service. The mainstream of service
supply chain literature focuses on the product.

3.2 Behavioural Research in Service Supply Chain

Service supply chain focuses on behavioural factors such as risk behaviour; equality
concern, forecast bias, reciprocal, altruism and strategic behaviour. There are also
few research articles focusing on the behaviours in the service supply chain, such
as understanding the relationship, competition and cognition. While studying these
behaviours, they have explored them in various functional areas including, demand
management, order process management, capacity and resources management,
customer and supplier relationship management, service capability, delivery and
performance management, market management, knowledge and information-flow
management, cash-flow management and risk management (Baltacioglu et al., 2007).
Partnership and cooperation among partners are an important element in the
service supply chain than in the traditional manufacturing supply chain, and give-and-
take behaviour among members certainly impacts the overall performance (Zhang
et al., 2017). Chu et al. (2012) have showed that reciprocal behaviour has an encour-
aging effect on supplier flexibility. Wang and Dai (2014) have explored the impact of
altruistic behaviour on the supply chain which works under optimal decision-making.
Research methods such as empirical investigation, case study, field experiment to
examine the presence of reciprocal and altruistic behaviour based on reflections
from industrial practice will provide more valuable recommendations to practi-
tioners. Given the significant role of customers in the service supply chain, much of
the strategic behaviour literature considers customer’s impact—including customer
preference, satisfaction and loyalty (Song et al., 2016; Dan et al., 2012).
Behavioural Operations Management: Trends and Insights 243

4 Behavioural Research in Other Contexts of Operations

Management

4.1 Humanitarian Operations

Disasters can be of two types—natural occurrence such as earthquake, flood

or and man-made one such as conflict, civil unrest. They can also be cyclical
(floods), gradual, or unexpected while arrival. Disasters have demonstrated the link
between economic underdevelopment, resource overutilization, political instability
and increased urbanization. For a process to be considered humanitarian, the notions
of humanity, neutrality and impartiality must be present (Tomasini & Van Wassen-
hove, 2009). Humanitarian organizations should combine every phase of disaster life
cycle, namely mitigation, preparedness, response and rehabilitation while designing
the relief programs and also design it in such a way that it is easier for the orga-
nizations to respond rapidly and appropriately to minimize the impact of a disaster
(Goldschmidt & Kumar, 2016).

4.2 Health Care Management

The importance of health care processes depends on employees’ participation in

behaviours, such as making and suggesting changes and contributing in their execu-
tion at work. Such an association is particularly noticeable only in a specialized
delivery structure where health care employees have substantive independence in
decision-making and control of operations. Knowledge sharing among workers and
teams is getting due consideration in the field of operations management (OM)
(Siemsen et al., 2009) as it can activate revolution in the operations and allow higher
productivity and performance. To be specific, front-line service providers such as
call-centre workers, airline squads, fire troops, police officers, teachers, healthcare
specialists and repair and maintenance operators. They are someone who addresses
the daily issues and faces the challenge of providing value to the customers under
time pressure, volatile workload, being primary contact of patients and still being
dependent on others for information and goods (Tucker & Edmondson, 2003). In this
kind of dynamic work setting, workers’ knowledge plays a major role in improving
the practices (Hertog, 2000).
Even though the significance of knowledge sharing is well known, the creativity
and initialization to encourage knowledge sharing mostly will not work well due to
employees’ aversion (Siemsen et al., 2009). Hence, there are research studies that
employ psychological and behavioural models to know about the knowledge sharing
behaviour of workers (Siemsen et al., 2009). Experimental and survey evidences
are treasured resources to realize the knowledge assets that intensify employees’
involvement in knowledge sharing context and in the modernization of current
operations.
244 S. Yamini

4.3 Agri-Food Supply Chain

Collaboration and cooperation are the most important factors for farmers and people
with low social-economic status to improve their profitability. Farmers being an
important stakeholder in agri-supply chain face a lot of limitations related to busi-
ness skills, ambitions and system thinking. Effective communications between the
farmers and traders have led to improvements in economic, environmental and social
standards (Hamprecht et al., 2005). In fact we can mitigate the misinterpretation and
resolve conflicts by collaborating the supply chain partnership. To achieve collabo-
ration, sharing of information, risk, reward, knowledge plays a major role. Further,
display of trust, reciprocation increases reliability, mutual understanding, coordina-
tion and commitment. This can help us in joint problem-solving and will help to
evaluate the quality outline of a supply chain collaboration.

4.4 Project Management

Problems related to behavioural aspects of project management have been acknowl-

edged by most researchers in this area, and it has also recently received a greater
attention in the operations and project management literature. Topics such as whistle-
blowing (Park & Keil, 2009), task design, interdependence, difficulty (Katok &
Siemsen, 2011; Schoenherr et al., 2017) and innovation (Hutchison-Krupat & Chao,
2014) have been examined by many researchers. Behavioural theories such as group-
think, expectancy theory and goal-setting theory (Kanfer, 1990) are applied in these
works. Surveys are the generally prevalent research method in this context.

5 Emerging Studies in Behavioural Operations

Management and Future Directions

Integrating the learnings from human behaviour is an interesting path of research in

the field of operations management. In the sustainable supply chain, the emphasis
on behavioural anomalies of supply chain members is an evolving area of research
(Kirchoff et al., 2016). Further, understanding the meaningful partnership among
various stakeholders is essential for achieving sustainability in supply chains. The
important factors required for effective collaboration are joint efforts, sharing activi-
ties, collaboration value, adaptation, trust, commitment, power, continuous improve-
ment, coordination and stability. The interplay among various behavioural factors
affecting the supply chain can be explored using empirical studies.
Other developing and potential area of research in the behavioural supply chain
is to study the occurrence of disruptions (Gurnani et al., 2014). The study of supply
chain risk and how agents respond to different types of risks is also a possible area
Behavioural Operations Management: Trends and Insights 245

of research in the behavioural supply chain (Tazelaar & Snijders, 2013; Reimann
& Ketchen, 2017). The perception of supply chain about fairness and the impact of
positive or negative reciprocation is also an important field of study to be consid-
ered in the future to have a complete understanding of behavioural supply chains
(Soundararajan & Brammer, 2018).
A variety of biases affect the behaviour of people in various facets of oper-
ations management including product development, R&D, supply chains, nego-
tiations, resource allocation, forecasting, inventory management and IT services.
The common problem to be addressed across these domains is the procurement,
managing and analysis of information from various sources. For instance, biases
such as anchoring and insufficient adjustment, availability bias, conformation bias,
overconfidence errors are very common across these domains. It is very difficult for
the organizations to train the employees to behave rationally. Hence, it is important
for them to understand the common behavioural biases exhibited and the general
thumb rules used by various stakeholders involved in these organizations. Other
potential area of research is to understand the common data interpretation errors that
can happen while forecasting or making any other data-related predictions.

6 Conclusions

Traditionally, quantitative models in operations management assume that the

decision-makers are rational, wherein the behavioural biases exhibited by people
are considered as just noise. The systematic way of people’s judgement, manage-
ment and improvement of operating systems is not clearly understood by many
stakeholders who are involved in the process of decision-making. Considering the
behavioural perspective to OM leads to better understanding of primary drivers of
operating systems. Hence, our efforts as researcher in the field of BOM should focus
on understanding and finding the hidden factors influencing the human behaviour.
Predicting the behavioural pattern can help in knowing the cognitive psychology of
decision-makers. The major objective of this systematic literature review is to under-
stand the recent trends in the field of BOM and to discuss about the relevant insights
acquired from this research. Similarly, the study clearly reveals that the appreciation
of human behaviour in future OM work can take multiple forms which will reap
many benefits. A variety of mixed methodologies, such as laboratory experiments,
field-based research, analytical modelling and empirical analysis, were used and
will be used to examine the important BOM problems in future too, where the major
objective will be to find the new behavioural and cognitive aspects that arise from
OM contexts.

Acknowledgements I sincerely thank the efforts of the editor and the reviewers for their
suggestions to improve the quality of the chapter.
246 S. Yamini

References

Agrawal, V., & Seshadri, S. (2000). Impact of uncertainty and risk aversion on price and order
quantity in the newsvendor problem. Manufacturing & Service Operations Management, 2(4),
410–423.
Aral, S., Bakos, Y., & Brynjolfsson, E. (2018). Information technology, repeated contracts, and the
number of suppliers. Management Science, 64(2), 592–612.
Baindur, D., & Macário, R. M. (2013). Mumbai lunch box delivery system: A transferable
benchmark in urban logistics? Research in transportation economics, 38(1), 110–121.
Baltacioglu, T., Ada, E., Kap, B. B., Lan, M. D., Yurt And, O., & Cem Kaplan, Y. (2007). A new
framework for service supply chains. The Service Industries Journal, 27(2), 105–124.
Bearden, J. N., & Rapoport, A. (2005). Operations research in experimental psychology. In Emerging
Theory, Methods, and Applications (pp. 213–236). INFORMS.
Bendoly, E., Croson, R., Goncalves, P., & Schultz, K. (2010). Bodies of knowledge for research in
behavioral operations. Production and Operations Management, 19(4), 434–452.
Bendoly, E., van Wezel, W., & Bachrach, D. G. (Eds.). (2015). The handbook of behavioral opera-
tions management: Social and psychological dynamics in production and service settings. Oxford
University Press.
Bolton, G. E., & Ockenfels, A. (2000). ERC: A theory of equity, reciprocity, and competition.
American Economic Review, 90(1), 166–193.
Bolton, G. E., Katok, E., & Ockenfels, A. (2004). How effective are electronic reputation
mechanisms? An experimental investigation. Management science, 50(11), 1587–1602.
Bolton, G. E., Ockenfels, A., & Thonemann, U. W. (2012). Managers and students as newsvendors.
Management Science, 58(12), 2225–2233.
Boshoff, C., & Leong, J. (1998). Empowerment, attribution and apologising as dimensions of service
recovery. International Journal of Service Industry Management, 9(1), 24–47.
Boudreau, J., Hopp, W., McClain, J. O., & Thomas, L. J. (2003). On the interface between operations
and human resources management. Manufacturing & Service Operations Management, 5(3),
179–202.
Brito, R. P., & Miguel, P. L. (2017). Power, governance, and value in collaboration: Differences
between buyer and supplier perspectives. Journal of Supply Chain Management, 53(2), 61–87.
Cantor, D. E., & Macdonald, J. R. (2009). Decision-making in the supply chain: examining problem
solving approaches and information availability. Journal of Operations Management, 27(3), 220–
232.
Chen, L., Kök, A. G., & Tong, J. D. (2013). The effect of payment schemes on inventory decisions:
The role of mental accounting. Management Science, 59(2), 436–451.
Chen, X., Shum, S., & Simchi-Levi, D. (2014). Stable and coordinating contracts for a supply chain
with multiple risk-averse suppliers. Production and Operations Management, 23(3), 379–392.
Choy, K. L., Li, C. L., So, S. C., Lau, H., Kwok, S. K., & Leung, D. (2007). Managing uncertainty in
logistics service supply chain. International Journal of Risk Assessment and Management, 7(1),
19–43.
Chu, P. Y., Chang, K. H., & Huang, H. F. (2012). How to increase supplier flexibility through
social mechanisms and influence strategies? Journal of Business & Industrial Marketing, 27(2),
115–131.
Croson, R., Schultz, K., Siemsen, E., & Yeo, M. L. (2013). Behavioral operations: the state of the
field. Journal of Operations Management, 31(1–2), 1–5.
Dan, A. (2008). Predictably Irrational: The Hidden Forces that Shape Our Decisions, Harper Collins
Publishers.
Dan, B., Xu, G., & Liu, C. (2012). Pricing policies in a dual-channel supply chain with retail
services. International Journal of Production Economics, 139(1), 312–320.
Davis, A. M. (2015). An experimental investigation of pull contracts in supply chains. Production
and Operations Management, 24(2), 325–340.
Behavioural Operations Management: Trends and Insights 247

Ebrahim-Khanjari, N., Hopp, W., & Iravani, S. M. (2012). Trust and information sharing in supply
chains. Production and Operations Management, 21(3), 444–464.
Eckerd, S., & Bendoly, E. (2011). Introduction to the discussion forum on using experiments in
supply chain management research. Journal of Supply Chain Management, 47(3), 3–4.
Enz, M. G., & Lambert, D. M. (2015). Measuring the financial benefits of cross-functional
integration influences management’s behavior. Journal of Business Logistics, 36(1), 25–48.
Fahimnia, B., Pournader, M., Siemsen, E., Bendoly, E., & Wang, C. (2019). Behavioral operations
and supply chain management—A review and literature mapping. Decision Sciences, 50(6),
1127–1183.
Fehr, E., & Schmidt, K. M. (1999). A theory of fairness, competition, and cooperation. The Quarterly
Journal of Economics, 114(3), 817–868.
Gans, N., & Croson, R. (2008). Introduction to the special issue on behavioral operations.
Manufacturing & Service Operations Management, 10(4), 563–565.
Gino, F., & Pisano, G. (2008). Toward a theory of behavioral operations. Manufacturing & Service
Operations Management, 10(4), 676–691.
Gurnani, H., Ramachandran, K., Ray, S., & Xia, Y. (2014). Ordering behavior under supply risk: An
experimental investigation. Manufacturing & Service Operations Management, 16(1), 61–75.
Goldschmidt, K. H., & Kumar, S. (2016). Humanitarian operations and crisis/disaster management:
A retrospective review of the literature and framework for development. International Journal of
Disaster Risk Reduction, 20, 1–13.
Hamprecht, J., Corsten, D., Noll, M., & Meier, E. (2005). Controlling the sustainability of food
supply chains. Supply Chain Management: an International Journal, 10(1), 7–10.
Hertog, P. D. (2000). Knowledge-intensive business services as co-producers of innovation.
International Journal of Innovation Management, 4(04), 491–528.
Ho, T. H., Lim, N., & Cui, T. H. (2010). Reference dependence in multilocation newsvendor models:
A structural analysis. Management Science, 56(11), 1891–1910.
Ho, T. H., Su, X., & Wu, Y. (2014). Distributional and peer-induced fairness in supply chain contract
design. Production and Operations Management, 23(2), 161–175.
Hutchison-Krupat, J., & Chao, R. O. (2014). Tolerance for failure and incentives for collaborative
innovation. Production and Operations Management, 23(8), 1265–1285.
Inderfurth, K., Sadrieh, A., & Voigt, G. (2013). The impact of information sharing on supply chain
performance under asymmetric information. Production and Operations Management, 22(2),
410–425.
Ireland, R. D., & Webb, J. W. (2007). A multi-theoretic perspective on trust and power in strategic
supply chains. Journal of Operations management, 25(2), 482–497.
Kahneman, D., & Tversky, A. (2013). Prospect theory: An analysis of decision under risk.
In Handbook of the fundamentals of financial decision making: Part I, 99–127.
Kanfer, R. (1990). Motivation theory and industrial and organizational psychology. Handbook of
Industrial and Organizational Psychology, 1(2), 75–130.
Katok, E., & Pavlov, V. (2013). Fairness in supply chain contracts: A laboratory study. Journal of
Operations Management, 31(3), 129–137.
Katok, E., & Siemsen, E. (2011). Why genius leads to adversity: Experimental evidence on the
reputational effects of task difficulty choices. Management Science, 57(6), 1042–1054.
Kaufmann, L., Rottenburger, J., Carter, C. R., & Schlereth, C. (2018). Bluffs, lies, and conse-
quences: A reconceptualization of bluffing in buyer–supplier negotiations. Journal of Supply
Chain Management, 54(2), 49–70.
Kim, H., Hur, D., & Schoenherr, T. (2015). When buyer-driven knowledge transfer activities really
work: A motivation–opportunity–ability perspective. Journal of Supply Chain Management,
51(3), 33–60.
Kirchoff, J. F., Tate, W. L., & Mollenkopf, D. A. (2016). The impact of strategic organizational
orientations on green supply chain management and firm performance. International Journal of
Physical Distribution & Logistics Management, 46(3), 269–292.
248 S. Yamini

Knemeyer, A. M., & Naylor, R. W. (2011). Using behavioral experiments to expand our horizons
and deepen our understanding of logistics and supply chain decision making. Journal of Business
Logistics, 32(4), 296–302.
Kremer, M., Minner, S., & Van Wassenhove, L. N. (2014). On the preference to avoid ex post
inventory errors. Production and Operations Management, 23(5), 773–787.
Liu, W., Wang, D., Long, S., Shen, X., & Shi, V. (2019). Service supply chain management: a
behavioural operations perspective. Modern Supply Chain Research and Applications, 1(1), 28–
53.
Loch, C. H., & Wu, Y. (2007). Behavioral operations management. Now Publishers Inc.
Long, X., & Nasiry, J. (2015). Prospect theory explains newsvendor behavior: The role of reference
points. Management Science, 61(12), 3009–3012.
Mentzer, J. T., & Flint, D. J. (1997). Validity in logistics research. Journal of business logistics,
18(1), 199.
Moritz, B., Siemsen, E., & Kremer, M. (2014). Judgmental forecasting: Cognitive reflection and
decision speed. Production and Operations Management, 23(7), 1146–1160.
Özer, Ö., Zheng, Y., & Chen, K. Y. (2011). Trust in forecast information sharing. Management
Science, 57(6), 1111–1137.
Park, C., & Keil, M. (2009). Organizational silence and whistle-blowing on IT projects: An
integrated model. Decision Sciences, 40(4), 901–918.
Preston, D. S., Chen, D. Q., Swink, M., & Meade, L. (2017). Generating supplier benefits through
buyer-enabled knowledge enrichment: a social capital perspective. Decision Sciences, 48(2),
248–287.
Pulles, N. J., Veldman, J., Schiele, H., & Sierksma, H. (2014). Pressure or pamper? The effects of
power and trust dimensions on supplier resource allocation. Journal of supply chain management,
50(3), 16–36.
Reimann, F., & Ketchen, D. J., Jr. (2017). Power in supply chain management. Journal of Supply
Chain Management, 53(2), 3–9.
Ren, Y., & Croson, R. (2013). Overconfidence in newsvendor orders: An experimental study.
Management Science, 59(11), 2502–2517.
Schoenherr, T., Bendoly, E., Bachrach, D. G., & Hood, A. C. (2017). Task interdependence impacts
on reciprocity in IT implementation teams: Bringing out the worst in us, or driving responsibility?
Production and Operations Management, 26(4), 667–685.
Schweitzer, M. E., & Cachon, G. P. (2000). Decision bias in the newsvendor problem with a known
demand distribution: Experimental evidence. Management Science, 46(3), 404–420.
Sengupta, K., Heiser, D. R., & Cook, L. S. (2006). Manufacturing and service supply chain
performance: a comparative analysis. Journal of Supply Chain Management, 42(4), 4–15.
Siemsen, E., Roth, A. V., Balasubramanian, S., & Anand, G. (2009). The influence of psychological
safety and confidence in knowledge on employee knowledge sharing. Manufacturing & Service
Operations Management, 11(3), 429–447.
Song, H., Yu, K., Chatterjee, S. R., & Jia, J. (2016). Service supply chain: strategic interaction and
relationship value. Journal of Business & Industrial Marketing, 31(5), 611–624.
Soundararajan, V., & Brammer, S. (2018). Developing country sub-supplier responses to social
sustainability requirements of intermediaries: Exploring the influence of framing on fairness
perceptions and reciprocity. Journal of Operations Management, 58, 42–58.
Spiliotopoulou, E., Donohue, K., & Gürbüz, M. Ç. (2016). Information reliability in supply chains:
The case of multiple retailers. Production and Operations Management, 25(3), 548–567.
Tazelaar, F., & Snijders, C. (2013). Operational risk assessments by supply chain professionals:
Process and performance. Journal of Operations Management, 31(1–2), 37–51.
Terpend, R., & Ashenbaum, B. (2012). The intersection of power, trust and supplier network size:
Implications for supplier performance. Journal of Supply Chain Management, 48(3), 52–77.
Thornton, L. M., Esper, T. L., & Autry, C. W. (2016). Leader or lobbyist? How organizational
politics and top supply chain manager political skill impacts supply chain orientation and internal
integration. Journal of Supply Chain Management, 52(4), 42–62.
Behavioural Operations Management: Trends and Insights 249

Tokar Asaad, C. (2015). Financial literacy and financial behavior: Assessing knowledge and
confidence. Financial Services Review, 24(2), 101–117.
Tokar, T. (2010). Behavioural research in logistics and supply chain management. The International
Journal of Logistics Management, 21(1), 89–103.
Tomasini, R. M., & Van Wassenhove, L. N. (2009). From preparedness to partnerships: case study
research on humanitarian logistics. International Transactions in operational research, 16(5),
549–559.
Tucker, A. L., & Edmondson, A. C. (2003). Why hospitals don’t learn from failures: Organizational
and psychological dynamics that inhibit system change. California Management Review, 45(2),
55–72.
Wallenburg, C. M., Cahill, D. L., Michael Knemeyer, A., & Goldsby, T. J. (2011). Commitment
and trust as drivers of loyalty in logistics outsourcing relationships: cultural differences between
the United States and Germany. Journal of Business Logistics, 32(1), 83–98.
Wang, L., & Dai, G. X. (2014). Altruism concern, service and the supply chain stackelberg game
research. Chinese Journal of Management Science, 22, 473–478.
Wang, Y., Wallace, S. W., Shen, B., & Choi, T. M. (2015). Service supply chain management: A
review of operational models. European Journal of Operational Research, 247(3), 685–698.
Yamini, S. (2020). Behavioral perspective of newsvendor ordering decisions: review, analysis and
insights. Management Decision, https://doi.org/10.1108/MD-07-2019-0975.
Yi, Z., Wang, Y., Liu, Y., & Chen, Y. J. (2018). The impact of consumer fairness seeking on distribu-
tion channel selection: Direct selling vs. agent selling. Production and Operations Management,
27(6), 1148–1167.
Zhang, Y., Li, J., & Gou, Q. (2017). An allocation game model with reciprocal behavior and its
applications in supply chain pricing decisions. Annals of Operations Research, 258(2), 347–368.
Investigating the Vaccine Supply Chain:
A Review

Dheeraj Chandra and B. Vipin

Abstract This paper presents a review of the existing research studies on vaccine
supply chain (VSC) management. We analyze the current VSC system in India and
investigate the research on VSC management specific to India. We focus on the coor-
dination aspect and network design in the literature review. The paper summarizes the
challenges in managing the VSC and provides future research directions to improve
their performance.

1 Introduction

The availability of vaccines at the health centers and their timely administration are
essential for the national immunization program. Intending to vaccinate children
and pregnant women with life-saving vaccines, the Government of India (GoI) in
1985 started a national vaccination program as a universal immunization program
(UIP), one of the largest immunization program globally. At present, UIP provides
12 vaccines free of cost to 27 million newborns, 29 million pregnant women and 100
million children aged 1–5 years annually (Universal Immunization Program, 2020).
Timely vaccination in prescribed dose requires a high service level by UIP. To a
large extent, the success of UIP depends on how the respective vaccine supply chain
(VSC) operates. An efficient VSC forms the backbone of immunization programs so
that vaccines can be delivered to the desired locations at the right time in the desired
quantity. However, due to such large-scale operations, UIP is struggling to properly
manage their VSC, which results in inefficiency in VSC (Chandra & Kumar, 2018).
For example, by the end of 2019, UIP India could vaccinate only 65% of children
below one year (Mission Indradhanush, 2020).

D. Chandra (B) · B. Vipin

Department of Industrial and Management Engineering, Indian Institute of Technology Kanpur,
Kanpur 208016, Uttar Pradesh, India
e-mail: [email protected]
B. Vipin
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 251
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3_13
252 D. Chandra and B. Vipin

Factors contributing to VSC inefficiency include the shortage of skilled health

workers, vaccine shortage, poor funding from the government to immunization pro-
gram and poor infrastructure. These factors have a significant effect on VSC that in
turn affects immunization program performance. Researchers of developing coun-
tries such as Nigeria and Benin have already started to support their immunization
program by redesigning the VSC, which has helped to reduce stockout and other
VSC issues. Despite the fact that India has demonstrated enhancements in some key
areas of VSC such as introducing an electronic vaccine intelligence network (eVIN)
to monitor vaccine stocks, temperature, real-time tracking, in any case, there are still
various zones which must be investigated to improve overall system performance.
So, we arrive at the following research questions (RQ):
RQ1. How is the existing VSC studied by the government, researchers, academi-
cians, etc., to revamp the effectiveness of the child immunization programs?
RQ2. How can communication, coordination and distribution, which are some of
the key issues in the VSC of India be improved through effective VSC contracts and
network design?
Based on the above discussion, we present a literature-based study in the context
of VSC in India. The motivation of this paper is twofold: first, through exhaustive
literature review, we investigate the current scenario of VSC with a special focus on
India, which can provide directions to the researchers, academician to focus on the
relatively untouched VSC aspects, and secondly, we analyze literature on the sup-
ply chain contract for vaccines and the vaccine distribution network in the context
of developing economies. Focus on the VSC contracts and distribution may help to
improve coordination and availability of vaccines at the health centers to optimize the
entire supply chain performance. Finally, we provide directions for future research
in the area of VSC in the context of India.
The remainder of the paper is organized as follows: Section 2 provides an overview
of VSC in India, and Sect. 3 discusses the literature review process and the sum-
mary. Section 4 reviews the supply chain coordination models, and Sec. 5 explores
the distribution network of VSC. Section 6 presents observations and insights from
the study, while Sect. 7 discusses the implication of the study and Sect. 8 concludes
the paper with listing the scope for future study.

2 Overview of VSC in India

To provide vaccines to the public, several players including government hospitals,

clinics and private organizations are involved. In this study, the focus is on the vac-
cination of children (including infants) and pregnant women through government
national immunization program and government hospitals. A generic map of the
VSC procurement and delivery mechanism in India is shown in Fig. 1. The central
drugs and standards control organization, which is a national regulatory authority, is
responsible for the vaccine regulatory system in India by approving vaccine intro-
duction, awarding permission to lead the clinical preliminaries and controlling the
Investigating the Vaccine Supply Chain: A Review 253

Fig. 1 VSC generic map in India

quality of vaccines. The VSC in India and many developing countries almost take a
similar form starting from the central-level procurement of vaccines for distribution
to regional vaccine stores.
Using the annual rate contracts of the general financing rules, UIP vaccines are
procured in India. Parallel contracts are granted for most vaccines due to the limited
production capacity of the suppliers to cover enormous vaccines demand (National
Vaccine Policy, 2020). After the central-level procurement, the vaccines move to
various regional vaccine stores and then to state vaccine stores. From the state vaccine
stores, the vaccines move to various divisional stores, which covers district vaccine
stores or immunization health centers (IHCs). The IHC supplies vaccines to each
town/unit/sub-immunization health center (SIHC) of their district. Finally, at each
SIHC, the health workers vaccinate the children and pregnant women, or the health
workers such as Anganwadi workers, ASHA workers and auxiliary nurse midwife
move to each house to vaccinate the targeted population.

3 Review Procedure and Analysis

A detailed and systematic literature review is performed to identify relevant docu-

ments in the VSC domain.
254 D. Chandra and B. Vipin

3.1 Literature Collection Procedure

The method of literature collection used in the current study involved three impor-
tant phases, which are discussed below. In the first phase. the period of study cov-
ered is set, which is from the year 1965 to 2020 (till May). In the second phase,
we search electronic databases ScienceDirect, Scopus, Web of Science, Taylor &
Francis Online, Springer, Emerald Insight and National Center for Biotechnology
Information (NCBI). We employ keyword vaccine supply chain on the databases
to identify appropriate journals. In addition, we perform Google search with the
same keyword to identify white papers, annual reports of healthcare organizations
such as WHO, UNICEF and GAVI. The search criteria disclose materials related to
WHO’s EVM Tool, The Bulletin of the World Health Organization, UNICEF, PATH,
Project Optimize, Ministry of Health & Family Welfare (Government of India), etc.
In the final phase, we select the documents (research papers, white papers and annual
reports). For the assessment and assortment of the research papers from the identi-
fied journals, the following inclusion criteria are defined: (i) English language for the
communication and appearance in peer-reviewed journals (ii) focus of the paper on
VSC of India and other countries. Next to search papers, we use various keywords,
for example, vaccine, immunisation/immunization, vaccination, supply and chain in
journals/libraries and websites identified in Phase 2. Following the two inclusion
criteria and the keyword search criteria, we collect the relevant papers. Then, we
finalize the papers for the study using the following three rounds. In the first round,
the papers are identified based on the title, abstract and keywords. Next, the identi-
fied papers are reviewed to recognize and remove duplication arising from references
determined in multiple databases. In the final round, the successful papers from the
first two rounds are reviewed comprehensively. Finally, following the three-phase
procedure, we identify 263 documents on VSC indexed in Science Citation Index
Expanded (SCIE), Social Sciences Citation Index (SSCI), Arts & Humanities Cita-
tion Index (A&HCI) and Emerging Sources Citation Index (ESCI) from 1965–2020
(until March 2020). Of the identified 263 papers, 18 papers focus on the VSC of
India.

3.2 Bibliometric Analysis

Bibliometric analysis is done in the field of VSC utilizing VOSviewer (v.1.6.13).

It is a software that generates bibliometric systems and maps dependent on the
paper abstract and title fields to help with recognizing well-known and important
terms/keywords/fields used by the researchers in their research (Chandra & Kumar,
2021). For using VOSviewer, the identified 263 documents are submitted to the
software, and the results are depicted in Fig. 2 and Table 1. Figure 2 comprises three
clusters, and the popular terms/fields of all the three clusters are shown in Table 1. The
bibliometric analysis result for VSC shows that research terms such as model, dose,
Investigating the Vaccine Supply Chain: A Review 255

Fig. 2 Bibliography mapping of VSC research (263 documents: 1965–2020)

Table 1 Top ten popular term/keyword in VSC research (263 documents: 1965–2020).
Term Occurrence
Model 62
Dose 52
Disease 49
Impact 49
Vaccine supply chain 48
Storage 42
Development 41
Benefit 35
Research 32
Case 32

disease and impact are the terms appearing in the maximum number of documents
and are popular fields of research in the VSC domain.
Then, we use SciMAT (v-1.1.0.4) to identify the distribution of research publica-
tions and popular journals in the VSC domain. The same 263 papers are submitted
to SciMAT software, and the results are shown in Figs. 3 and 4 and Table 2. The
results revealed that the research in VSC in the last two decades has shown improve-
ments, especially from 2012–2017. The sudden increase in 2017 is due to a special
issue on VSC by the journal Vaccine. Nevertheless, from the last two years, there is
not much growth in the publications, which implies that the VSC domain is still in
infancy and needs more contributions from researchers, academicians, government
organizations, healthcare industries, etc.
256 D. Chandra and B. Vipin

Fig. 3 Distribution of articles on VSC research (2000–2020).

Fig. 4 Journals with most number of articles on VSC research (1965–2020)

We also report research on VSC in India, and the distribution of literature is

presented in Fig. 5 and Table 3. Although UIP India started in 1985, the progress in
VSC research is slow. The significant contribution of researchers can be seen from
the year 2000.

4 Supply Chain Coordination

Supply chain coordination aims to improve the system performance by aligning the
objectives of supply chain partners with that of the system. In this regard, contracts
Investigating the Vaccine Supply Chain: A Review 257

Table 2 Distribution of literature based on journals

Year
Journal Till 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
Vaccine 2 1 1 2 2 4 6 6 12 10 30 6 10 3
BOWHO 3 – – – – – 1 1 – 2 1 – – –
HV&I – – – – – – – 2 1 – – 1 – 2
EROV 1 1 – – – 1 – 1 – – – 1 1 –
BMCPH – – – – 2 – – 1 – – – – 1 1
PO – – – – 1 – 2 – 1 – – 1 – –
EJOR – – – – – 1 – – – – – 2 – 1
JOID – – – – 1 – – – – – 2 1 – –
C&IE – – – – – – – – – – – 1 1 1
HA – – – – 2 – – – – 1 – – – –
Other 29 3 1 3 6 6 3 8 7 18 12 13 12 3
Total 35 5 2 5 14 12 12 19 21 31 45 26 25 11
list of abbreviations; BOWHO—Bulletin of the World Health Organization, HV&I—Human Vaccines & Immunothera-
peutics, EROV—Expert Review of Vaccines, BMCPH—BMC Public Health, PO—Plos One, EJOR—European Journal
of Operational Research, JOID—Journal of Infectious Diseases, C&IE—Computers & Industrial Engineering, HA—
Health Affairs

Fig. 5 Distribution of documents in VSC research in India (18 documents: 1965–2020).

play a major role in establishing channel coordination. In the VSC, accomplish-

ing channel coordination is consistently a troublesome undertaking for the VSC
decision-makers due to uncertainties present in the supply and demand side. Vaccine
is a complex biological product, and it requires years of effort and state-of-the-
art manufacturing facilities to finally produce a quality vaccine. On the supply side,
uncertainties such as equipment failures, Vivo testing, less number of vaccine produc-
ers with WHO prequalification, dynamic government regulations, etc., often result
258 D. Chandra and B. Vipin

Table 3 Contribution of researchers in VSC in India

Author Contribution
Islam et al. (2020) Study the implementation of cholera vaccine in
national immunization program of India and
Bangladesh
Chandra et al. (2019) Prioritize the 25 key issues in VSC of child
immunization of India and other developing
countries
Lee et al. (2019) Study the effect of coping in VSC in Bihar,
India
Dasgupta et al. (2019) Validate the removal and disposal of tOPV
vaccine in various health centers in India
Wedlock et al. (2019) Study the effect of introducing microneedle
patch vaccines into VSC
Chandra et al. (2019) Identify the key performance indicators to
improve VSC performance using two-way
assessment
Wedlock et al. (2018) Study the effect of introducing the
dual-chamber injection device for
measles-rubella vaccine in Benin, Bihar and
Mozambique
Chandra et al. (2018) Identify the driving and dependence power of
the 25 key issues in VSC to improve decision
making
Lee et al. (2017) Examine the economic impact of thermostable
vaccine in Benin, Bihar and Niger
Hanson et al. (2017) Study the issue of freezing in vaccine cold
chain of India and other nations
Gilbert et al. (2017) Analyze the performance of digital VSC in
Uttar Pradesh, India
Ashok et al. (2017) Identify the challenges and solutions to
improve cold chain systems in 10 countries
Kristensen et al. (2016) Improve the vaccination services by taking
experts opinions of immunization stakeholders
of every stage of supply chain
Chatterjee et al. (2016) Study the financial needs of UIP, India
Lim et al. (2016) Identify the optimal outreach vaccination
location for improving vaccination coverage
Harsh et al. (2016) Estimate vaccine wastage rate and wastage
factor in 92 immunization health centers of
Gujarat
Panda et al. (2015) Assess the effect of supportive supervision in
immunization program of Odisha
Murhekar et al. (2013) Monitor the vaccine cold chain temperature in
10 states of India
Investigating the Vaccine Supply Chain: A Review 259

in a random yield or disruptions for the supplier. Therefore, suppliers are always at
a high risk of losses in the vaccination industries. On the other side, child immu-
nization programs across the developing nations have a high pressure to improve
vaccination rates. With poor financial support from the government and tremendous
pressure, the immunization programs do not pay much attention to VSC key factors
such as better forecasting tools and techniques. Today, the child immunization pro-
grams still rely on the old and outdated methods of forecasting, which can result in
the mismatch between supply and demand. Due to such supply and demand issues,
it has been observed that immunization programs and suppliers often blame each
other that other side has not ordered or produced correctly. In fact, the problem lies
in the missing coordination between the VSC players with an incorrect way of incen-
tivizing each other. Thus, designing appropriate VSC contracts can be a very crucial
element for the decision-makers that can help to correctly incentivize each player in
order to improve the system performance. So, in this section, we explore research
studies on achieving channel coordination in VSC through contracts.
Chick et al. (2008) formulate a cost-sharing contract between the supplier and gov-
ernment to improve the supply of the influenza vaccine. In the influenza vaccine
production, strains are required to develop a vaccine. To do so, egg-based technol-
ogy is used to produce these strains. Thus, the number of eggs used in the pro-
duction cycle is an important factor for the manufacturer. Also, due to the com-
plex production stages, the actual production quantity may be different from the
planned production quantity, and it may vary in a random manner, which is labeled
as the yield uncertainty or random yield (a common process in vaccination, agri-
culture industries). The objective function to the government problem is defined by
G F = E[s max{P − ξ BA , 0} + Pa W + Pr Z + Pe n e ]. The model comprises a social
cost s that is a function of the total number of infected people in the population, vac-
cine administration cost Pa , procurement cost Pr and production cost per egg C.
First, the authors design various contracts with commonly used contracts such as
wholesale price, buyback and observe that these contracts could not help in achiev-
ing coordination. For instance, in the buyback, the production level is below the
optimal production level, which is not feasible for coordination. Hence, the authors
design a cost-sharing contract in which the government pays Pe per egg for a produc-
tion volume of n E to the supplier. The optimal production level n ∗E for the supplier
 f N B/n ∗E
is 0 yg(y)dy = C−P
Pr
e
, where g(y) is the probability density function of the
yield. The authors propose that if Pe is chosen in a manner that C−P Pr
e
= ξ b/dC−Pa ,
then it can help to coordinate the supply chain as it provides supplier an incentive
to produce more and become risk-free for extra production and also supports their
efforts through incentive Pe . Therefore, such cost-sharing contracts coordinate the
supply chain by coordinating the incentives.
Chick et al. (2017) in their second paper consider information asymmetry while
designing the contracts. Information asymmetry arises when there is an imbalance
in the flow of information that affects the decision process. For example, the sup-
plier may hide the production cost and may charge as rent (information rent) to
the government to reveal the information. Thus, the authors explore an influenza
260 D. Chandra and B. Vipin

vaccine supply contract between the government whose objective is to minimize

the expected social costs and the supplier who has production yield uncertainty,
private information of its productivity and unverifiable production effort. Authors
design a set of contracts in which the government can minimize its expected cost
by providing incentives to the supplier to exert an appropriate level of effort and
information rent from the government. For example, in one of the scenarios, the
authors consider a wholesale price contract with productivity factor θ = θi known
to both government and supplier. If vaccines are procured for fraction f from
the total susceptible population N , yield uncertainty has CDF as Y , and n is the
production effort of the supplier, then the supplier objective function is given by
M F = E[cn + D( f N /(θ Y ) − n)+ − Pr f N ], whereas government objective func-
tion is G F = E[sT ( f ) + Ca ( f N + δ( f N − nθ Y )+ ) + Pr f N ]. The model com-
prises a variable production cost c, social cost per infected person s, cost per unit
effort D to cover a shortage, vaccine administration cost Ca , procurement cost Pr ,
δ as additional administration cost to government due to late delivery, total social
cost for infected population sT ( f ). The authors propose that if θ is known to both
parties and there is no additional charge for vaccine administration due to late deliv-
ery (δ = 0) , then a traditional wholesale price can coordinate the supply chain by
coordinating the incentives. However, if δ > 0, then a traditional wholesale price
is not able to coordinate the supply chain because the government orders less, and
 f N /θ n ∗
the supplier produces less than the optimum level 0 dG(y) = Dc . In addition,
for δ > 0, if the government charges penalty Cs to the supplier due to late deliv-
ery, then it helps to coordinate the supply chain. Thus, the wholesale price contract
with penalty cost due to shortage coordinates the supply chain and new objective
functions become M F = E[cn + D(( f N /Y ) − n)+ − Pr f N + Cs ( f N − nY )+ ]
and government objective function G F = E[sT ( f ) + Ca ( f N + δ( f N − nY )+ ) +
Pr f N − Cs ( f N − nY )+ ].
Dai et al. (2016) consider demand uncertainty and propose a contract between the
supplier and retailer to improve on-time delivery performance of influenza vaccine.
First, they design the delivery time-dependent quantity flexibility (D-QF) contract
and late rebate (LR) contract and observe that these contracts are not able to provide
incentives and motivation to the supplier who bears the risk of the early production
of influenza vaccine to improve on-time delivery. Then, the authors improve their
model with buyback-and-late-rebate (BLR) contract, which assist to remove double
marginalization. Suppose (1 − α) is the probability of late delivery, β is the probabil-
ity of influenza vaccine strain similar to the composition approved by the authority,
cr and ce are the unit production cost in regular and early production mode, ca is
administrative cost, b is buyback credit, w is wholesale price, p is the selling price
per unit to retailer, and ρ is the proportion of wholesale price that supplier rebates
for late delivery, then the authors propose that BLR coordinates the supply chain if,
w−ce ) ( p−w−ca )(ce −β cr )
b∗ = β(p(β
p−ca )−ce
and ρ ∗ = w(1−α){β ( p−ca )−ce }
. This contract integrates the properties
of buyback and late rebate contracts to provide credit for leftover inventory and rebate
for late delivery.
Investigating the Vaccine Supply Chain: A Review 261

Arifoglu and Tang (2020) propose a two-sided incentive scheme to coordinate the
influenza VSC. The model considers yield uncertainty from the supply side and posi-
tive and negative externalities from the demand side. Positive externalities are defined
as factors that have a positive effect on society, for instance, an individual getting vac-
cination also has a benefit for the society by protecting them against the infection due
to herd immunity. Likewise, negative externalities have a cost to society, for instance,
any individual getting vaccinated reduces the chance of others getting vaccination
and increases chances of infection due to the usage of a vaccine dose. In general, two
sets of incentive programs are designed to coordinate the supplier and customer’s
incentives with a social optimum level: (i) coordinating vaccination incentive r (Q r )
and (ii) coordinating transfer payment T (Q r ). It is observe that when vaccine supply
is high, the customer’s demand in the decentralized supply chain is relatively low
than the social optimum due to positive externalities, therefore the model works in
a way that it generates a positive incentive (r (Q r ) > 0) to customers so that more
customers can come for vaccination. When supply is low, the demand is more than
the social optimum level due to negative externalities; therefore, to restrain the high
vaccine demand and to ensure that the customer’s having high infection costs gets
a vaccination, the model generates a negative vaccination incentive r (Q r ) < 0. To
coordinate the transfer payment, a transfer amount of T (Q r ) is given to the supplier
so that supplier produces an optimal production quantity Q c , which can help to coor-
dinate the marginal benefits of the supplier and the society.
Apart from the above-mentioned studies, other researchers also contribute to design-
ing effective VSC contracts. For example, Yan et al. (2017) consider a health author-
ity, vaccine supplier and population to propose a set of contracts to optimize the
influenza VSC. First, the authors design a payment scheme contract in which the
payment is made on the order quantity Q from the health authority. Such payment
scheme contracts improve the efforts decisions of suppliers and health authorities.
Next, they configure contracts in which extra installment is an element of the number
of infections. Such contracts help healthcare organizations to minimize costs due to
the payments based on infections, deaths, etc. Shamsi et al. (2018) develop an option
contract for vaccines required in emergencies from two suppliers. The model plans
to limit the acquirement and social costs utilizing the susceptible-infected-recovered
(SIR) model. Mamani et al. (2012) consider an oligopoly market and design a sub-
sidy contract for vaccine market coordination and socially optimal vaccine coverage.
Mamani et al. (2013) study the the inefficiencies in allocation of influenza vaccines
and design a game theoretical model to reduce inefficiencies. Yarmand et al. (2014)
consider an infection outbreak and propose a two-phase vaccine allocation model to
different locations to minimize vaccine production and administration cost. Adida et
al. (2013) consider a monopoly business model for an imperfect vaccine with opera-
tional issues and system impacts and propose a two-section menu of endowments that
prompts a socially effective degree of coverage. Chandra and Vipin (2021) propose
a subsidy contract to coordinate the VSC of child immunization India.
262 D. Chandra and B. Vipin

5 Supply Chain Distribution

Supply chain distribution is one of the important phases in the VSC of developing
countries. It determines how the vaccines should be distributed to the health cen-
ters, how many levels are required to distribute the vaccines, points of distributing
the vaccines, how much inventory should be kept at each level, etc. (Lotty & van
Jaarsveld, 2018). Since last-mile delivery of vaccines is one of the complex stages
in the VSC, therefore, various issues often arise in delivering the vaccine during the
endpoint. Tana et al. (2014) point out that issues such as handling of vaccines, cold
chain monitoring, storage of vaccines are common issues faced by the immunization
programs in vaccines distribution. Gupta et al. (2013) mention that in India the coor-
dination issues in VSC have prompted issues concerning the supply chain and normal
accessibility of immunizations supply at outreach centers. Their study also indicates
that precise and timely forecasting of the materials required for immunizations is a
significant bottleneck in the distribution of vaccines on time with required doses.
As VSC in India comprises multiple levels and due to the large population and com-
plex demography, the VSC issues are inevitable. Hence, the supply chain design
in India should be robust that can deliver vaccines efficiently and effectively to the
delivery locations. Thus, the supply chain designers can rethink of redesigning the
existing VSC. For example, is it necessary to store vaccines at divisional vaccine
stores as the vaccines can be supplied directly to district vaccine stores from state
vaccine stores? In this regard, Chen et al. (2014) propose a planning model to improve
the vaccine distribution performance in developing nations. Their model consists of
various parameters such as demand of vaccines at each location, capacity at each
stage, inventory lost at each location
 and doses
  administered.
 The objective function
in their model is defined as j nd j +  i j t ( pi,R j,t + pi,F j,t ). The first term
in the expression ensures the number of full vaccination of children at a location j,
while the second term represents i vaccines used from a refrigerator at location j in
time t and likewise i vaccines used from a freezer at location j in time t. The model
assists to capture the current practice of procurement and distribution of vaccines in
developing nations and ensures that distribution efficiency is improved. This model
has been adopted by the immunization programs of Niger, Vietnam and Thailand,
and the implementation has shown positive results.
Ng et al. (2018) design a multi-criterion approach to improve vaccine distribution
efficiency by minimizing cost and reproduction number, improving social benefits of
influenza vaccination.
 One  of their
 objective functions to minimize vaccination cost
φ φ φ φ
is formulated as g∈ G φ∈ q∈ Q g qg r g sg l gq . Here qg represents the number of
φ
people that can be immunized by following a strategy φ, r g represents the costs of
vaccination a single person in any risk group, sg is the percentage of people in risk
φ
group g in the entire population, and l gq is the percentage of immunized people in
the risk group g by following a vaccination plan q under an allocation strategy φ.
They consider the SIR model and divide the population into risk groups like seniors,
infants, babies, etc. Allocation strategy is the type of vaccination strategy followed by
the immunization program to vaccinate a group of peoples. For example, they may
Investigating the Vaccine Supply Chain: A Review 263

follow a different strategy to vaccinate infant who needs more precaution during
vaccine administration as compared to senior citizens. The proposed model can help
to reduce immunization costs by 10% than the average annual costs of immunization
programs. Another instance of redesigning the VSC is in the case of an emergency
or outbreak. Today, due to the rise in infectious diseases, it is imperative that the
supply chain be adequately responsive so that the need for vaccines in outbreaks/
emergencies can be met with short delivery times. Therefore, the VSC in India should
have the capacity and capability to fulfill the demand of vaccines of any state with
a outbreak/emergency by their neighboring states. In such scenarios, the logistics
of private companies may help distribute vaccines to the children. Yarmand et al.
(2014) propose a two-phase vaccination policy model in case of an emergency or
an outbreak. The model helps to minimize production costs by reducing the usage
of vaccine doses. Such models should be designed in the scenario of India so that
deaths due to outbreaks can be reduced.
Yadav et al. (2014) suggest that integration of VSC with the supply chains of vari-
ous health commodities may help to minimize costs incurred due to storage, trans-
portation, warehousing and distribution. They propose that the VSC designers can
integrate activities such as procurement, transport, storage and ordering with other
health products supply chain to improve VSC efficiency and effectiveness. However,
it requires proper planning at each level of the supply chain as various uncertainties
may affect the overall performance of the VSC. Outreach can be another solution to
improve distribution performance. In India, the majority of the population resides in
the remote and rural areas. Therefore, due to transportation and other complexities,
bringing children to immunization health centers becomes difficult for the parents.
Hence, outreach can be an effective measure to distribute vaccines to the desired
locations.
Few studies propose mathematical models to address outreach location problems.
For instance, Lim et al. (2016) develop various coverage models to find the optimal
outreach location to improve vaccination coverage. In one of their binary coverage
models, the problem is formulated as follows: A population residing in a radius of R
will be vaccinated through outreach, and the remaining cannot be vaccinated through
outreach. Let k is the total number of villages that will be covered by outreach, pi
is the population in each village, ci is the cost at each outreach, di j is the distance
between two villages i, j, N is maximum outreach permissible to set up, ai ∈ {0, 1},
where 1 belongs to the village i if it has been finalized as outreach else 0, bi ∈ {0, 1},
where 1 indicates if village i is covered else 0. So the objective
n function is to maximize
the number ofpeople to be vaccinated through outreach i=1 pi bi , with constraints
 n
such as bi ≤ j∈ Si ai for Si = { j : di j ≤ R, j, i = 1, ..., n}, i=1 ai ≤ N . Hence,
such coverage models can be helpful in UIP to cover more population in rural and
remote areas.
Inventory management is also important in the vaccine distribution phase. Proper
inventory management ensures that vaccines do not go waste, and the demand can
be fulfilled with the available stock. In this aspect, Hovav et al. (2015) propose a
264 D. Chandra and B. Vipin

inventory model for influenza vaccine distribution. Tebbens et al. (2010) design an
optimal vaccine inventory tool for the control of eradicated diseases.

6 Observations and Insights of the Study

We perform a systematic literature review in VSC and observe that various con-
tracts and distribution models have been proposed by various authors that can help
to improve coordination and delivery performance. One major gap we identified that
with respect to child immunization in India is that no such models have been devel-
oped to improve coordination and distribution. Therefore, by looking at the results of
the VSC performance improvements of other nations through VSC contracts, it can
be summarized that such contracts can surely help child immunization in India to
optimize their supply chain performance. In VSC, as the supplier alone faces the risk
of supply uncertainty, therefore, a contract in which the government is incentivizing
the supplier in case of the overproduction can be helpful to achieve channel coordi-
nation. In this context, a subsidy contract in the presence of supply uncertainty can
be helpful to the UIP India to coordinate the supply chain. In a subsidy contract, UIP
may provide financial support to the supplier if the supplier’s production quantity
is greater than the UIP order quantity. This can motivate the supplier to produce
more and fulfill the order in each batch. Other contracts such as cost sharing will
also be helpful to UIP India. A late rebate contract in which the supplier gives a
rebate if the delivery period is late can be also appropriate in the VSC of UIP. UIP
lacks on-time financial support from the government which affects the release in the
purchase order; therefore, suppliers and UIP should also consider such aspects while
designing the contracts. For example, the supplier may fulfill the demand of UIP even
if payment is late and may charge a small amount of penalty if such a case happens.
This may help the UIP to meet the child immunization demand. Contracts such as
D-QF or option contract may pose some challenges in UIP due to the nature of the
contracts. In D-QF, the retailer can return some leftover inventory to the supplier at
a full price up to a certain level; therefore, in the case of child immunization, imple-
menting the D-QF can be difficult because for child immunization vaccines there is
less number of suppliers and returning the leftover stock with the full price condition
may not be acceptable by the supplier due to factors such as supply uncertainty,
high-order volume and low-profit margins. UIP may also consider some distribution
models to tackle uncertainty such as COVID-19 and other outbreaks. For example,
due to COVID-19, immunization rates have been affected drastically in major states
of India. Therefore, models in which such disruptions can be handled will be really
helpful. Models with outreach centers may help to improve immunization rates in
India.
Investigating the Vaccine Supply Chain: A Review 265

7 Discussion and Implications of the Study

An efficient and effective VSC can help improve child immunization program per-
formance; however, relatively low volume of literature on VSC highlights the fact
that the field is still to gain wide popularity among the researchers. In this paper, we
perform a thorough literature review on VSC and identify that only 263 documents
in four international citation indexes since 1965 have addressed the issue of VSC in
their study. We also report that only 18 papers targeted VSC of India—a low volume
of research on the child immunization program performance of India, indicating that
a significant research opportunities exists in VSC especially in the context of ongoing
pandemic. From Table 3, it can be observed that 18 documents, which focus on VSC
of India, have primarily focused on performance measurement, vaccine wastage,
economic characteristics, vaccine introduction, etc., however important aspects such
as vaccine demand forecasting and shortages have not been addressed in any of the
studies. Hence, researchers and academicians can focus on addressing forecasting
and shortage issues in VSC to assist UIP to improve vaccine availability.
The literature review reveals that the primary focus of researchers is to study vaccines
used for any age group, for illustration, Lin et al. (2020) study the cold chain trans-
portation decision improvements in VSC, and Hovav et al. (2015) design a network
flow model for inventory management of influenza vaccine and distribution. Only a
few studies focused on basic vaccines and their modeling and analysis to improve
child vaccination performance. For instance, Chandra and Kumar (2021) evaluate
the effect of KPIs of the VSC on the sustainable development of a health operation
of UIP India as Mission Indradhanush, Chen et al. (2014) study the planning model
for the WHO-EPI vaccine, and Tebbens et al. (2010) study the optimal polio vaccine
stockpile design. Focusing on other aspects such as decision-making problems in
VSC, Arji et al. (2019) report that around 40 papers (2005–2019) employ the fuzzy
concept as a decision-making tool in human infectious diseases, but in the field of
VSC for child immunization in developing countries like India, only a few studies
have considered the application of fuzzy for decision-making such as Chandra et al.
(2018) and Chandra and Kumar (2019).
We further discuss the importance of coordination in the supply chain and how
different contracts achieve coordination in different VSC settings. As compared to
the immunization programs of developed nations, coordination of VSC is not been
given substantial importance by the UIP India. Establishing coordinating contracts in
the vaccination programs can reduce cost of the supply chain and improve vaccines
delivery performance. We also discuss various issues in VSC distribution in, India
and how a well-designed distribution channel can help to improve VSC performance.
266 D. Chandra and B. Vipin

8 Conclusions, Limitations and Future Scope

In this study, we review the literature on VSC and investigate the literature addressing
the supply chain of vaccination programs in India. We identify 263 documents, of
which, 18 concentrated on VSC of India. VSC is undoubtedly one of the topics
in which a significant amount of research gaps exists. India runs the largest child
immunization program in the world reinforcing the need to investigate the issues in
vaccine supply chain provide solutions. Further, we provide comprehensive review
on the models on supply chain coordination and supply chain network design for
vaccines in a global setting.
In the current COVID-19 scenario with multiple organizations focusing on devel-
oping vaccines, governments need to ensure efficient and effective supply chains of
to be available vaccines by ensuring proper contractual terms and optimal network
design for the supply chain. One of the major limitations of our study is that the
research papers are drawn from SCIE, SSCI, A&HCI and ESCI indexed journals.
Future research review can consider other indexed journals and look into other issues
and aspects of VSC than the contract design and network design.
Based on the analysis of literature on VSC, we find the need for better modeling
and analysis of VSC considering the region-specific and product-specific character-
istics. Incorporating the stochastic nature of different parameters such as demand
and supply is essential in designing the VSC distribution network. Future research
can look into the different supply and demand uncertainties to be addressed while
designing contracts for VSC coordination. SIR modeling approach to better capture
the demand for vaccines can be considered as an extension to the proposed models.

Acknowledgements The authors would like to thank all the anonymous reviewers for the con-
structive comments and insightful suggestions to improve the quality of the paper.

References

Adida, E., Dey, D., & Mamani, H. (2013). Operational issues and network effects in vaccine markets.
European Journal of Operational Research, 231(2), 414–427.
Arifoglu, K., & Tang, C. S. (2020). Two-sided incentive program for coordinating the influenza
vaccine supply chain. SSRN. http://dx.doi.org/10.2139/ssrn.3361140.
Arji, G., Ahmadi, H., Nilashi, M., Rashid, T. A., Ahmed, O. H., Aljojo, N., & Zainol, A. (2019).
Fuzzy logic approach for infectious disease diagnosis: A methodical evaluation, literature and
classification. Biocybernetics and Biomedical Engineering, 39, 937–955.
Ashok, A., Brison, M., & LeTallec, Y. (2017). Improving cold chain systems: Challenges and
solutions. Vaccine, 35(17), 2217–2223.
Chandra, D., & Kumar, D. (2018). A fuzzy micmac analysis for improving supply chain performance
of basic vaccines in developing countries. Expert review of vaccines, 17(3), 263–281.
Chandra, D., & Kumar, D. (2019). Prioritizing the vaccine supply chain issues of developing coun-
tries using an integrated ism-fuzzy anp framework. Journal of Modelling in Management, 15(1),
112–165.
Investigating the Vaccine Supply Chain: A Review 267

Chandra, D., & Kumar, D. (2019). Two-way assessment of key performance indicators to vaccine
supply chain system in india. International Journal of Productivity and Performance Manage-
ment, 68(1), 194–230.
Chandra, D., & Kumar, D. (2021). Evaluating the effect of key performance indicators of vac-
cine supply chain on sustainable development of mission indradhanush: A structural equation
modeling approach. Omega, 101(June), 102258.
Chandra, D., & Vipin, B. (2021). On the vaccine supply chain coordination under subsidy contract.
Vaccine. https://doi.org/10.1016/j.vaccine.2021.05.081.
Chatterjee, S., Pant, M., Haldar, P., Aggarwal, M. K., & Laxminarayan, R. (2016). Current costs
& projected financial needs of india’s universal immunization programme. The Indian journal of
medical research, 143(6), 801–808.
Chen, S.-I., Norman, B. A., Rajgopal, J., Assi, T. M., Bruce, L. Y., & Brown, S. T. (2014). A planning
model for the WHO-EPI vaccine distribution network in developing countries. IIE Transactions,
46(8), 853–865.
Chick, S. E., Hasija, S., & Nasiry, J. (2017). Information elicitation and influenza vaccine production.
Operations Research, 65(1), 75–96.
Chick, S. E., Mamani, H., & Simchi-Levi, D. (2008). Supply chain coordination and influenza
vaccination. Operations Research, 56(6), 1493–1506.
Dai, T., Cho, S.-H., & Zhang, F. (2016). Contracting for on-time delivery in the u.s. influenza
vaccine supply chain. Manufacturing & Service Operations Management, 18(3), 332–346.
Dasgupta, R., Sharma, S., Sharma, N., Banerjee, K., Haran, E., Muliyel, J., Salunke, S., Masoodi,
M. A., Haldar, P., Bahl, S. et al. Successful switch (topv to bopv) in india: Tribute to a resilient
health system. Vaccine, 37(17), 2394–2400.
Gilbert, S. S., Thakare, N., Ramanujapuram, A., & Akkihal, A. (2017). Assessing stability and
performance of a digitally enabled supply chain: Retrospective of a pilot in uttar pradesh, india.
Vaccine, 35(17), 2203–2208.
Gupta, S. S., Nair, G. B., Arora, N. K., & Ganguly, N. K. (2013). Vaccine development and deploy-
ment: Opportunities and challenges in india. Vaccine, 31S, B43–B53.
Hanson, C. M., George, A. M., Sawadogo, A., & Schreiber, B. (2017). Is freezing in the vaccine
cold chain an ongoing issue? a literature review. Vaccine, 35(17), 2127–2133.
Harsh, S. D., Bonny, S. H., Ketan, P., & Surendra, J. (2016). Closing gaps in routine immunization-
impact and cost assessment on components of new vaccine policy in routine immunization in
gujarat, india. Journal of Research in Medical and Dental Science, 4(1), 70–74.
Hovav, S., & Tsadikovich, D. (2015). A network flow model for inventory management and distri-
bution of influenza vaccines through a healthcare supply chain. Operations Research for Health
Care, 5, 49–62.
Islam, M. T., Chowdhury, F., Qadri, F., Sur, D., & Ganguly, N. (2020). Trials of the killed oral
cholera vaccine (shanchol) in india and bangladesh: Lessons learned and way forward. Vaccine,
38, A127–A131.
Kristensen, D. D., Bartholomew, K., Villadiego, S., & Lorenson, K. (2016). What vaccine product
attributes do immunization program stakeholders value? results from interviews in six low-and
middle-income countries. Vaccine, 34(50), 6236–6242.
Lee, B. Y., Wedlock, P. T., Haidari, L. A., Elder, K., Potet, J., Manring, R., Connor, D. L., Spiker, M.
L., Bonner, K., Rangarajan, A. et al. (2017). Economic impact of thermostable vaccines. Vaccine,
35(23), 3135–3142.
Lee, B. Y., Wedlock, P. T., Mitgang, E. A., Cox, S. N., Haidari, L. A., Das, M. K., et al. (2019).
How coping can hide larger systems problems: the routine immunisation supply chain in bihar,
india. BMJ global health, 4(5).
Lim, J., Claypool, E., Norman, B. A., & Rajgopal, J. (2016). Coverage models to determine outreach
vaccination center locations in low and middle income countries. Operations Research for Health
Care, 9, 40–48.
Lin, Q., Zhao, Q., & Lev, B. (2020). Cold chain transportation decision in the vaccine supply chain.
European Journal of Operational Research, 283(1), 182–195.
268 D. Chandra and B. Vipin

Lotty, E. D., & van Jaarsveld, W. (2018). Literature review: The vaccine supply chain. European
Journal of Operational Research, 268, 174–192.
Mamani, H., Adida, E., & Dey, D. (2012). Vaccine market coordination using subsidy. IIE Trans-
actions on Healthcare Systems Engineering, 2(1), 78–96.
Mamani, H., Chick, S. E., & Simchi-Levi, D. (2013). A game-theoretic model of international
influenza vaccination coordination. Management Science, 59(7), 1650–1670.
Ministry of Health and Family Welfare. Mission indradhanush. Available at https://nhp.gov.in/
mission-indradhanush1_pg. Accessed: 2020-03-26.
Ministry of Health and Family Welfare. National vaccine policy. Available at https://main.mohfw.
gov.in/sites/default/files/108481119000.pdf. Accessed: 2020-03-26.
Ministry of Health and Family Welfare. Universal immunization program. Available at https://www.
nhp.gov.in/sites/default/files/pdf/immunization_uip.pdf. Accessed 26 March, 2020.
Murhekar, M. V., Dutta, S., Kapoor, A. N., Bitragunta, S., Dodum, R., Ghosh, P., Swamy, K. K.,
Mukhopadhyay, K., Ningombam, S., Parmar, K. et al. (2013). Frequent exposure to suboptimal
temperatures in vaccine cold-chain system in india: results of temperature monitoring in 10 states.
Bulletin of the World Health Organization, 91, 906–913.
Ng, C., Cheng, T., Tsadikovich, D., Levner, E., Elalouf, A., & Hovav, S. (2018). A multi-criterion
approach to optimal vaccination planning: Method and solution. Computers & Industrial Engi-
neering, 126, 37–649.
Ng, C., Cheng, T., Tsadikovich, D., Levner, E., Elalouf, A., & Hovav, S. (2018). A multi-criterion
approach to optimal vaccination planning: Method and solution. Computers & Industrial Engi-
neering, 126, 37–649.
Shamsi, N. G., Torabi, S. A., & Shakouri, H. G. (2018). An option contract for vaccine procurement
using the SIR epidemic model. European Journal of Operational Research, 267(3), 1122–1140.
Issues in vaccination delivery. (2014). L. J. Tan, M. A. Barnett, A. Eisenberg, H. F. Fields, J. M.
Hackell, M. M. Léger, B. Meadows, K. Moore, S. J. Rehm, M. C. Rothholz, C. Smith, T. K., and
D. L. Wexler. From refrigerator to arm. Vaccine, 32, 2389–2393.
Tebbens, R. J. D., Pallansch, M. A., Alexander, J. P., & Thompson, K. M. (2010). Optimal vaccine
stockpile design for an eradicated disease: Application to polio. Vaccine, 28(26), 4312–4327.
Tebbens, R. J. D., Pallansch, M. A., Alexander, J. P., & Thompson, K. M. (2010). Optimal vaccine
stockpile design for an eradicated disease: Application to polio. Vaccine, 28(26), 4312–4327.
Wedlock, P. T., Mitgang, E. A., Siegmund, S. S., DePasse, J., Bakal, J., Leonard, J., Welling, J.,
Brown, S. T., & Lee, B. Y. (2018). Dual-chamber injection device for measles-rubella vaccine:
The potential impact of introducing varying sizes of the devices in 3 countries. Vaccine, 36(39),
5879–5885.
Yadav, P., Lydon, P., Oswald, J., Dicko, M., & Zaffran, M. (2014). Integration of vaccine supply
chains with other health commodity supply chains: A framework for decision making. Vaccine,
32, 6725–6732.
Yan, X., & Zaric, G. S. (2017). Influenza vaccine supply chain with vaccination promotion effort
and its coordination. IISE Transactions on Healthcare Systems Engineering, 7(1), 53–72.
Yarmand, H., Ivy, J. S., Denton, B., & Lloyd, A. L. (2014). Optimal two-phase vaccine allocation to
geographically different regions under uncertainty. European Journal of Operational Research,
233(1), 208–219.
Index

A Cumulative distance, 51
Agri-food supply chain, 235, 244
Analytical model, 90, 91, 93, 95, 234, 235
Artificial intelligence, 120, 141, 162, 164,
D
177
Decentralized supply chain, 91, 97, 261
Artificial neural network, 159, 161, 162
Deterministic heuristic, 119–122, 124, 127
Digital technology, 226, 228, 229
Discount rate, 23, 25, 35
B
Backorder cost, 6, 7, 19 Discrete transportation lead time, 65
Base-stock level, 68, 94 Distribution, 16, 25, 29, 30, 32, 39, 44–46,
Behavioral experiments, 237, 238, 240 53, 90, 96, 189, 205, 208, 212, 236,
Behavioural operations management, 233– 252, 253, 255–257, 262–266
235, 244 Dummy node, 46, 47
Bound-based sampling heuristic, 105
Buyer-supplier relationship, 235, 237, 238,
242 E
Electronic data interchange, 44
Enterprise resources planning, 218
C E- procurement, 220, 222, 224, 228
Cancer diagnosis, 156 Equilibrium, 23, 25, 28, 31, 35, 38, 238
Capacity, 23–26, 28–35, 45, 47, 50, 51, 53, Expected payoff, 33
94, 207, 242, 253, 262, 263 Expected sales, 95, 96, 99, 100
Centralized supply chain, 91, 98, 100
Chebyshev’s inequality, 38
Cold chain transportation, 265
Comparative evaluation, 16 F
Confidence ellipsoids, 34 Finite horizon, 43–46
Consistency index, 187 Finite-time horizon, 58
Continuous-review policy, 16, 93 Fixed cost, 47, 58, 112
Contract mechanism, 89, 91, 92 Flowshop, 119–121, 124, 127
Cost-sharing contract, 259 Forecast information sharing, 23, 27, 29, 32
Credibility index, 23, 25, 34, 35 Forecast matrix, 32
Credibility test, 23, 25, 27, 30, 31, 40 Forward reduction technique, 127–129

© The Editor(s) (if applicable) and The Author(s), under exclusive license 269
to Springer Nature Singapore Pte Ltd. 2021
B. Vipin et al. (eds.), Emerging Frontiers in Operations and Supply Chain Management,
Asset Analytics, https://doi.org/10.1007/978-981-16-2774-3
270 Index

G Jobshop, 119–121, 124–129

Game theory, 25, 237–239 Joint economic lot size, 66
Game-theoretical model, 29, 261
Genetic algorithm, 4, 90, 91, 93, 104–106,
111, 117, 204 K
Global reporting initiative matrix, 159, 164, K-means algorithm, 142, 155
165, 167, 176
Goodwill cost, 33
Green supply chain management, 181 L
Lead-time, 4, 24, 31, 40, 68, 94
Lean supply chain, 177
H L’Hopital rule, 36, 38
Hamiltonian cycle, 46 Linear programming, 139, 142, 144, 156
Hamiltonian path, 46 Linear regression, 174
Health care management, 243 Logistics, 140, 143, 146, 148, 149, 167,
Heuristic, 43–45, 54, 57–62, 90, 91, 93, 104, 173, 177, 189, 203, 218, 224, 228,
105, 112, 114, 119, 120, 122, 124, 235–237, 239, 241, 263
235 Logistics management, 235–237, 239
Holding cost, 4, 6, 8, 12, 15, 16, 19, 44, 46,
47, 53, 94
Hotelling’s T 2 , 31, 33 M
Humanitarian operations, 243 Machine learning, 139, 141, 142
Human resource allocation problem, 204 Machine-routing, 119–121, 123, 124, 127–
Hybrid inventory ordering policy, 5, 20 130
Manufacturing line, 119–122, 124, 127–130
Manufacturing resources planning, 218
I Manufacturing SME, 217, 219, 221
Immunization health center, 253, 263 Material requirement planning, 218
Infinite horizon, 44 Mathematical models, 4–6, 8, 12, 14, 16, 20,
Influenza vaccine, 259–261, 264, 265 43, 45, 203, 239, 263
Information and communication technology, Meta-heuristic, 91, 93, 104, 105, 112, 113,
217, 228 116, 117, 120
Information asymmetry, 28, 29, 259 Mixed integer linear programming, 3, 43, 44,
Information quality, 219, 222, 224, 225 46, 51
Information sharing, 24–29, 40, 41, 93, 218– Mixed integer non-linear programming, 65,
222, 224, 225, 228, 229, 237, 239, 69
240 Multi-criteria decision-making, 161, 165,
Information technology, 25, 201 181, 183, 184, 186, 193
Internal Lean practices, 219, 222, 224, 225, Multi-echelon supply chain, 62, 92
228, 229 Multilayer perceptron, 159, 163, 164
Inventory, 3–8, 10–20, 24, 29, 40, 43–48, 50, Multi level supply chain, 92, 93
51, 53, 57, 58, 60, 62, 93, 94, 119, Multiple linear regression, 139, 142, 144,
120, 225, 227, 235, 237–239, 245, 156
260, 262–265 Multi-retailer, 40
Inventory replenishment, 44, 45 Multivariate statistics, 23, 25
Inventory routing problem, 43–45, 51, 54, 60
Inverse reduction technique, 128, 129
N
Nash equilibrium, 26, 91
J Nearest neighbourhood algorithm, 58
Job-index based insertion scheme, 119 No split-delivery, 44, 45, 51, 62
Index 271

NP-hard, 43, 44, 54, 62, 120 Repeated game, 23–26, 28, 31, 35, 38, 40
Replenishment, 6–8, 10, 15, 16, 26, 43–45
Replenishment cycle, 69, 70, 72
O Resource allocation, 202–204, 210, 213, 245
On-hand inventory, 7, 10, 11, 46 Review length, 39, 40
On-order inventor, 7, 11 Review strategy, 23, 27–30, 35, 37, 39, 40
Operations research, 142, 235, 236
Optimal allocation, 204, 207
Optimal order quantity, 91, 92, 95–100
Ordering cost, 4–8, 10, 15, 16, 19, 46, 47,
S
50, 58
Safety factor, 65, 70, 78, 81–83
Order quantities test, 30
Order-up-to-level, 6, 12, 13, 19 Safety stock, 220, 222, 228
Outsourcing, 220, 222, 224, 228, 239 Sales rebate contracts, 89, 90, 92, 93, 96,
100, 101, 104, 113, 114, 116
Scheduling, 93, 119, 203, 204
P Service integrator, 242
Parallel simulated annealing, 105, 108, 110, Service level constraint, 20
111 Service provider, 204–206, 241–243
Parallel weighted genetic algorithm, 106 Service supply chain, 235, 241, 242
Pareto optimal, 23, 25, 26 Setup cost, 70, 74, 75
Pearson correlation coefficient, 159, 174 Shortage cost, 5, 8, 12, 15, 16, 94
Perfect public equilibrium, 23, 25, 35 Simulation, 201, 203, 207, 210, 211, 213
Periodic-review order-up-to policy, 3–5, 16, Simulation-based optimization, 207, 210
20 Staff allocation, 203, 212
Petroleum industry, 181–185, 189, 192, 193 Statement of work, 204, 212
Pipeline inventory, 7 Strategic collaboration and lean practices,
Positive externalities, 261 220, 222
Postponement, 219, 220, 222, 224, 225, 228,
Strategic planning, 220, 221, 223–225, 229
229
Subcontracting, 220, 223, 224, 228
Price-dependent stochastic demand, 65, 67
Sub tour elimination, 50, 51
Price-sensitive stochastic demand, 67
Process sequence, 119 Supersequence, 119–121, 123, 127–130
Profit margin, 70, 264 Supply chain, 3–6, 8, 12–20, 23, 24, 26–31,
Project management, 235, 236, 244 40, 41, 43, 44, 46, 50, 60, 62, 89–
Proportion of periods test, 30 102, 104–106, 111–116, 160, 161,
Punishment phase, 25, 28, 35 164, 165, 174, 182, 217–222, 226–
229, 234–242, 244, 245, 251, 252,
254–256, 258–260, 262–266
Q Supply chain coordination, 43, 90, 91, 93,
Quantitative modelling, 237, 238 96, 100, 116, 252, 256, 266
Quantity discount, 92 Supply chain management, 24, 26, 90, 119,
Quantity flexibility, 92, 260 184, 233, 235–239
Supply chain management practice, 181, 226
Supply chain performance, 159–161, 164–
R 166, 174, 217, 239, 252, 264
Regression analysis, 174, 178 Support vector machine, 142
Relational contract, 24, 25, 28, 29, 40, 41 Sustainability, 160, 162, 165, 177, 182–184,
Renewable energy, 170, 177, 181, 183, 192– 189, 193, 219, 244
194 Sustainable supply chain management, 160,
Reorder point, 6 163
272 Index

T U
Transportation cost, 44–46, 50, 53, 58, 60, Universal immunization program, 251
94
Travelling salesman problem, 54, 57, 58
Trigger strategy, 30, 39
Truthful information sharing, 24, 29, 40, 41 V
Two-pair adjacent interchange scheme, 119, Vaccine supply chain, 251, 266
127, 129 Variance inflation factor, 140, 143, 145
Two-sided incentive scheme, 261 Vendor managed inventory, 43, 93