Int. J.

Production Economics 134 (2011) 177–187

Contents lists available at ScienceDirect

Int. J. Production Economics

journal homepage: www.elsevier.com/locate/ijpe

A fuzzy logic approach to supply chain performance management

Gilberto Miller Devós Ganga a,1, Luiz Cesar Ribeiro Carpinetti b,n
a ~ Carlos, SP, Brazil
Department of Production Engineering, Federal University of Sao Carlos, Rod. Washington Luiz Km 235, 13565-905, Sao
b ~ Carlos, University of Sao Paulo, Av. Trabalhador Sancarlense 400, 13566-590, Sao
~ Carlos, SP, Brazil
Department of Production Engineering, School of Engineering of Sao

a r t i c l e i n f o abstract

Article history: The aim of this paper is to propose a supply chain performance model based on fuzzy logic to predict
Received 27 August 2010 performance based on causal relationships between metrics of the Supply Council Operations Reference
Accepted 17 June 2011 model (SCOR) model. The main contribution and originality of this proposal relates to the application of
Available online 20 July 2011
Fuzzy logic to predict performance based on performance metrics levels 1 and 2 of the SCOR model.
Keywords: Fuzzy logic is a technique suitable for dealing with uncertainty and subjectivity, which becomes an
Supply chain management interesting auxiliary approach to manage performance of supply chains. A descriptive quantitative
Fuzzy logic approach was adopted as research method, based on the prediction model. Statistical analysis of the
Performance measurement prediction model results confirmed the relevance of the causal relationships embedded in the model.
Decision processes
The findings reinforce the proposition that the adoption of a prediction model based on fuzzylogic and
on metrics of the SCOR model seems to be a feasible technique to help managers in the decision making
process of managing performance of supply chains.
& 2011 Elsevier B.V. All rights reserved.

1. Introduction proposes metrics to manage performance on multiple dimensions

in a hierarchical structure defined based on the causal relation-
Supply chain performance measurement has gained relevance ships of the business processes and sub-processes of the supply
and attention from the academic and practitioner communities chain management reference model. This is the case for instance
over the past decades. It is based on general theories of perfor- of the level one SCOR metrics for supply chain asset, reliability
mance measurement systems such as (Kaplan and Norton, 1992; and responsiveness performance measurement.
Bititci et al., 1997): These multi-dimensional and inter-organizational character-
istics of supply chain performance measurement systems can lead
 Derive metrics from strategic objectives and key performance to large number of metrics and difficulty in sharing data through-
factors. out the chain, making it complex to measure performance of
 Design leading and lagging metrics related to financial and the supply chain. An alternative to these is the use of Analytic
non-financial dimensions of performance. Hierarchy Process—AHP to calculate an overall measure of supply
 Adopt a business process management approach to derive chain performance based on the weights of different measures in
metrics. a hierarchy of relationships (Sharma and Bhagwat, 2007; Varma
 Design a measurement system with leading and lagging et al., 2008). Other multi-criteria techniques are also applied with
metrics interconnected by causal relationships. the same purpose. For instance, Agarwal et al., 2006 propose the
use of Analytic Network Process—ANP to calculate a weighted
A distinctive characteristic of supply chain performance mea- index that analyzes the combined influence of four supply chain
surement is that the measurement system should span the supply performance determinants. Chan et al. (2003) and Chan and Qi
chain, including metrics with interdependencies that cross the (2003) propose a combination of AHP with fuzzy sets to calculate
borders of organizations (Beamon, 1999; Beamon and Chen, 2001; an aggregated supply chain performance index. Fuzzy set theories
Gunasekaran et al., 2001; 2004; Kleijnen and Smits, 2003; (Zimmermann, 1991; Liao, 2005) have been increasingly
Shepherd and Günter, 2006). The Supply Chain Operations Refer- employed in supporting decision making processes as it is
ence model—SCOR of the Supply Chain Council (SCC, 2009) appropriate to dealing with uncertain and subjective situations,
typical of measures of qualitative nature or pair-wise compar-
n
ison-based method, such as AHP and ANP (Lin et al., 2006).
Corresponding author. Tel.: þ55 16 33739421.
E-mail addresses: [email protected] (G.M.D. Ganga),
However, calculation of global metrics of supply chain perfor-
[email protected] (L.C.R. Carpinetti). mance by aggregation of unrelated dimensions of performance
1
Tel.: þ55 16 33519535. can be meaningless for the purpose of performance management.

0925-5273/$ - see front matter & 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijpe.2011.06.011
178 G.M.D. Ganga, L.C.R. Carpinetti / Int. J. Production Economics 134 (2011) 177–187

A different approach to supply chain performance manage- supply chain processes (plan, source, make and return).
ment is using fuzzy logic inference rules to build a prediction Gunasekaran et al. (2004) propose metrics also related to supply
model that anticipates results of supply chain lagging metrics chain generic processes to manage performance on operational,
based on leading metrics and if-then scenarios. Unahabhokha tactical and strategic levels.
et al. (2007) propose a fuzzy expert system to design a predictive The metrics defined by the SCOR model follow the same
performance measurement system that derives predictive output principles. The SCOR model (SCC, 2009) proposes to analyze a
values from input values. An approach such as this can be used to supply chain from three perspectives (SCC, 2009): process,
set targets on leading indicators based on prediction of perfor- metrics and best practices. The SCOR framework maps the
mance of results. The study proposed here follows the approach connections between the inter-organizational processes in each
presented by Unahabhokha et al. (2007). However, differently company in a supply chain. One of the advantages of this model is
from their proposal, performance prediction was modeled accord- the creation of a common and standardized language among the
ing to causal relationships between metrics defined by the Supply companies within a supply chain, thus enabling companies to
Chain Operations Reference Model SCORs Version 8.0 (2006). The compare supply chain performance as a whole.
SCOR model was chosen since it incorporates some key business Top level SCOR metrics focus on five performance attributes, as
processes throughout the supply chain, reinforcing the need to follows (Bolstorff and Rosenbaum, 2007):
develop a systemic performance evaluation (Holmberg, 2000).
Adopting the SCOR model also allows the use of benchmarking  Reliability: the performance related to the delivery, i.e., whether
reference information to define the universe of discourse of each the correct product (according to specifications) is delivered to
input and output variables. the correct place, in the correct quantity, at the correct time, with
This paper is organized as follows: Section 2 briefly revises the the correct documentation and to the right customer.
SCOR model in order to illustrate the hierarchical view of  Responsiveness: the speed at which a supply chain provides
processes and performance measurements suggested for the the products to customers.
performance attributes in supply chain. Section 3 aims to clarify  Flexibility: the agility of a supply chain to respond to market
some fundamental concepts regarding fuzzy systems used in the changes in demand in order to gain or maintain its competitive
prediction model. Section 4 describes the proposed prediction advantage.
model, while Section 5 presents the results of the prediction  Cost: involves all the costs related to the operation of a supply
model. Next, Section 6 presents analysis of the results based on chain.
the surface response method. Final considerations about this  Asset management efficiency: the efficiency of an organization
research work are made in Section 7. in managing its resources to meet demand. This includes the
management of all the resources: fixed and working capital.

2. Supply chain performance measurement and the SCOR

metrics For each metric established in level 1, according to the
measurement dimension, there is a hierarchy of interrelated
Application of the theories of measurement systems for metrics. The unfolding of these metrics and their cause and effect
managing performance of supply chain are reported in several relationships makes it possible to analyze the performance of a
papers (Brewer and Speh, 2000; Park et al., 2005; Yeh; et al., supply chain from different perspectives, from a purely opera-
2007; Cai et al., 2009). Although it is right to say that there is no tional perspective to a strategic one.
measurement system for managing performance of the entire Fig. 1 illustrates this behavior for the metric upside supply
supply chain (Lee and Billington, 1993), recommendations for chain flexibility, related to the attribute of performance flexibility.
supply chain performance measurement include the adoption of The Upside Flexibility Supply Chain is a metric, which defines the
metrics that cross the borders of organizations, focusing on amount of time, generally in days, which the supply chain would
dimensions of performance related to inter and intra organiza- take to respond to unplanned change in demand of around 20%,
tional processes (Lapide, 2000). Beamon (1999) and Beamon and without cost or penalties for not delivering within the agreed
Chen (2001) propose metrics for managing resources, output and time. The SCOR model unfolds this metric in three others in level
flexibility of conjoined supply chain. Shepherd and Günter (2006) 2: Upside Source Flexibility; Upside Make Flexibility; and Upside
proposes inter and intra organizational metrics related to generic Deliver Flexibility (SCC, 2009). Such metrics represent the capacity

Supply Chain Flexibility Performance Attribute

Upside Supply Chain Flexibility Level 1

Upside Source Flexibility Upside Make Flexibility Upside Delivery Flexibility Level 2

Time needed to obtain additional capital (a)

Additional source volumes – supplier constraints (b) Constraints

Factors

Time needed to recruit/hire/train additional staff (c)

Fig. 1. Upside source flexibility: hierarchical metric structure (Supply Chain Council, 2009).
 G.M.D. Ganga, L.C.R. Carpinetti / Int. J. Production Economics 134 (2011) 177–187 179

of response (in days) of the processes source, make and deliver to contribution of a specialist in the area of the phenomenon to be
adequate to changes in demand, without extra costs or penalties modeled so that the pertinence functions can be constructed for
for not delivering within the agreed time. Each metric will be the description of the inputs.
constrained by different variables of process performance. For
example, the measure Upside Source Flexibility (Fig. 1) could be Definition 4. Membership function
formed by: (a) amount of time needed to obtain capital to fill the A pertinence function is a numeric, graphic function, which
gap between current capital availability and capital needed to attribute values of pertinence to discrete values of a variable, in
sustain a 20% increase in quantities ordered; (b) amount of time its universe of discourse, which represent the numerical interval
for the supplier to deliver its product (raw materials or products) of all the possible real values which a specific variable may
for order processing, including safety stock to sustain a 20% assume.
increase in quantities to be supplied; and (c) amount of time
needed to recruit/hire/train additional staff to fill gap between Definition 4.1. Triangular Membership Function (TMF)
current staff and staff needed to sustain a 20% increase in A fuzzy number A~ is a triangular fuzzy number if its pertinence
quantities delivered. function mA given by
0 for x o a,
8
>
> xa
> ba
3. Fuzzy logic definitions >
< for a r x rb,
mA ðxÞ ¼ for b rx r c,
ð2Þ
> cx
>
In this section, some basic definitions of fuzzy logic were
>
: cb 0 for x 4 c,
>
revised from Zadeh (1978), Dubois and Prade (1980),
Zimmermann (1991), Klir and Yuan (1995) and Bojadziev and
Bojadziev (1997). The definitions and notations presented in this
In which a, b and c are real numbers with aoboc. Outside the
section were used throughout this work and are essential to
interval [a,c], the pertinence degree is null, b represents the point in
understand the proposed model.
which the pertinence degree is maximum, as illustrated in Fig. 2.
Definition 1. A fuzzy set A~ in X is defined by The notation used to represent the triangular pertinence func-
~
Afx, mA ðxÞ=x A Xg, ð1Þ tions in this work follows de logic expressed in Fig. 3, which
exemplifies the use of pertinence functions for the performance
metric Accurate Documentation.
In which mA ðxÞ : X-½0,1 is called pertinence function of A~ and
As it can be observed, this metric is described by a composition
~
mA(x) is the degree of pertinence of x in A. of pertinence functions defined by the points (xi ; yi ), where xi,
Definition 2. Linguistic Variable: In this variable the value is denotes the universe of discourse of the variable and yi the
expressed qualitatively by linguistic terms and quantitatively by a pertinence level for each given measure level. By analyzing
pertinence function. The service level, for example, is a linguistic Fig. 3, it can be noticed that the highlighted points define the
variable whose values are: very low, low, medium, high or very high. limits of the triangular functions. The points (84;1) and (91;0),
Definition 3. Fuzzification: It is the process that relates the define the pertinence of the crisp variable to the linguistic level
numeric values of the crisp input variables to the values of ‘‘low’’. In the same way, the points (84;0), (91;1) and (98;0),
the linguistic variables. It is extremely important to have the define the pertinence of the crisp variable to the linguistic level
‘‘medium’’. The pertinence function ‘‘high’’ was defined by follow-
ing the same logic scheme. There are different types of pertinence
functions, such as the triangular, trapezoidal and Gaussian. How-
ever, the triangular pertinence is the most used (Wu and Lee,
2007; Ross, 1997; Liu and Liu, 2001), because it is intuitively easy
to use and calculate for decision makers (Dubois and Prade, 1980;
Lin et al., 2007).

Definition 5. Ifyandy then Rules

The rules of inference relate the information that forms the
x knowledge base in a fuzzy system. With the definition of the
a b c
pertinence functions, of both input and output variables, it is
Fig. 2. Triangular membership function. possible to define the rules of inference, generally from specialists’

low medium high

1.0
(84;1)
membership level

0.8 (98;0)
(91;1)
0.6
low
0.4
(91;0) medium
0.2
(84;0) (91;0) (98;0) high
0.0
84 % 88% 91% 94% 98%
universe of discourse

Fig. 3. Composition of pertinence functions for Accurate Documentation.

180 G.M.D. Ganga, L.C.R. Carpinetti / Int. J. Production Economics 134 (2011) 177–187

Table 1 4. The proposed model

Decision table: IFyandythen rules.
The research presented in this paper is descriptive, empirical
B1 ? Bj Bj þ 1 ? Bm
and quantitative (Bertrand and Fransoo, 2002), which is a class of
A1 C11 CB1j ? C1,m model-based research that is primarily interested in creating a
^ ^ ? ^ ^ ^ model (a prediction model in this case) that adequately describes
Ai Ci1 ? Cij Ci,j þ 1 ? Ci,m the causal relationships that may exist in reality, aiming at
Ai þ 1 Ci þ 1.1 ? Ci þ 1,j Ci þ 1,j þ 1 ? Ci þ 1,m
^ ^ ^ ^ ^
understanding the processes in the supply chain.
An Cn1 ? Cnj Cn,j þ 1 ? Cnm The model proposed in this work uses fuzzy logic to establish
causal relations among the performance variables (metrics)
defined by the SCOR model and thus it can predict the perfor-
opinions. The rules with possible fuzzy outputs represented by Cij mance of the processes of a supply chain. Fig. 4 illustrates the
are presented symbolically through a matrix n  m (n lines and m interaction between the modeling world and the real world. The
columns), called decision table, where Cij ,i ¼ 1,. . .,n, j ¼ 1,. . .,m, use of fuzzy logic in the modeling world acts as a mechanism to
are renamed elements of the set fC1 ,. . .,Cl g, as it can be observed in predict performance of the response variables (lagging), as spe-
cific input values (leading) are determined by estimative or by the
Table 1.
available data. Thus, fuzzy logic was used to model the effect of an
Thus, the meaning of ify andythen rules is
input variable (leading) over a response variable (lagging), which
If x is Ai and y is Bj then z is Ck ð3Þ can be observed in Fig. 4.
By using this model, it is possible, for example, to predict the
behavior of the metric Upside Source Flexibility (lagging) from
Definition 6. Inference Engine: The fuzzy inference engine takes the leading metrics leading (a); (b) and (c), illustrated in Fig. 1.
all possible combinations of the pre-determined fuzzy sets, The development of the fuzzy model highlighted in Fig. 4 can be
compares them with the fuzzy rule base, and assigns to understood (step by step) through the process depicted in Fig. 5
each combination the corresponding fuzzy region of influence. and discussed in the next paragraphs.
The inference engine is fundamental for the success of the Step 1: The definition of the performance metrics is a critical task
fuzzy system, since it supplies the output from each fuzzy input in the development of a performance measurement system. Due to
and from the relation defined by the base of rules. The literature the lack of logical mathematical equation among the leading
about fuzzy logic presents basically two methods of fuzzy performance measures and the consequent lagging performance
inference: The Mamdani Method and the Kang–Takagi–Sugeno measures this work resorted to the SCOR model 8.0 in order to
Method. The basic difference between these two methods is the obtain a reference framework to understand the cause and effect
type of response variable (consequent) and the Fuzzification relationships between the performance attributes (level 1 and level
procedure. This work has adopted the Mamdani method in its 2 metrics) related to the performance of a supply chain. The
integrity. understanding of such logical relations was essential in proposing
the present model, as it can be observed in Fig. 6. The SCOR 8.0,
Definition 6.1. Mamdani method through the document SCORcard, supplied a wide range of perfor-
Mamdani proposed a binary relation fuzzy M between xem to mance metrics (levels 1, 2 and 3) for the performance attributes in
mathematically model the base of rules. Such method is based on the supply chain. It would not be viable to develop a model for all
the metrics suggested by the SCOR 8.0. Therefore, to simplify the
a max–min rule of interference composition, following these
model, the choice of performance metrics from the SCOR model, in
procedures:
terms of variety and quantity, was based on benchmarking informa-
tion from several sources such as Supply Chain Council, APQC (2009)
 in each rule Rj, of the base fuzzy rules, a conditional ‘‘if x is Aj and the Manufacturing Performance Institute (2009). Each one of
then m is Bj ’’ is modeled by application 4 (minimum);
the metrics included in the model (Fig. 6) was modeled by a fuzzy
 the t-norma 4 (minimum) is adopted for the logic connective function as given in Fig. 6.
‘‘and’’;
 for the logic connective ‘‘or’’ t-conorma 3 (maximum) is 4.1. Supply chain cost performance
adopted, connecting the fuzzy rules of the base of rules.

Therefore, the binary relation M is a fuzzy subset of X  U The cost performance was modeled by using a fuzzy function f~ :
whose pertinence function is given by R6 -R so that,8ðx,
~ z~ Þ A R6 ¼ R5  R, the value of the cost perfor-
mance is obtained W ~ ¼ f~ ðy~ , z~ Þ, where y~ is a fuzzy variable, obtained
jMðx, mÞ ¼ max1 r i r r ½jRi ðx, mÞ ¼ max1 r i r r ½jAj ðxÞ4jBj ðmÞ ð4Þ from the components of the total cost of the, described by the fuzzy
function g~ : R5 -Rso that y~ ¼ g~ ðxÞ. The fuzzy variable z~ is defined as
cost of goods sold (COGS) and the components of the total cost of
In which r is the number of rules, which compose the base of the supply chain are x~ 1 ¼ ðx~ 11 , x~ 12 , x~ 13 , x~ 14 , x~ 15 Þ, in which x~ 11 is the
rules,A~ j and B~ j are the fuzzy subsets of rule j. Each one of the planning and finance cost, x~ 12 , the inventory carrying cost, x~ 13 the
values jAj(x) and jBj(m) is interpreted as the degrees with which information technology cost, x~ 14 , the material acquisition cost and
x and m are in the fuzzy subsets, A~ j and B~ j , respectively. x~ 15 , the order management cost, respectively.

Definition 7. Defuzzification: It is the process of transforming the 4.2. Supply chain asset performance
linguistic concept, obtained by the inference procedure, in a well
defined numeric value (crisp variable), which is used as an The asset performance was modeled by the fuzzy function
effective output of the fuzzy system. The best known methods f~ : R3  R2 ¼ R5 -R, so that, 8ðx,
~ y~ Þ A R3  R2 , is the asset perfor-
for defuzzification are the Center-of-area method (CoA); and the ~ ~
mance W ¼ f ðp~ , q Þ, in which the fuzzy variables p~ and q~ are
~
Center-of-maximum (CoM); and the Medium of Maximum (MoM) obtained from the components of cash to cash (x2) and return
method, all used in this work. over assets (y2), respectively, and can be described by the fuzzy
 G.M.D. Ganga, L.C.R. Carpinetti / Int. J. Production Economics 134 (2011) 177–187 181

Corrective Corrective
Goal/Standart
mechanism mechanism

Actual Actual
input Adjust output
Adjust
measure measure
value value

Input (leading) Output (lagging)

measure measure

Real World

Predictive Performance Measurement System

Estimade or
Fuzzy Expert Predict output
actual input
System performance
measure value

Modelling World

If...then
Rules

Mamdani
Fuzzification Inference Defuzzification
Engine

Database
Membership Supply Chain Membership
Functions Benchmarking Functions

Fig. 4. Predictive supply chain performance measurement. Font: Proposal by authors based on Unahabhokha et al. (2007) and Bojadziev and Bojadziev (1997).

1. Define performance atribbutes, leading measures and

SCOR 8.0 and lagging measures
Benchmarking
Metrics
2. Define universe ofdiscourse, membership functions
and linguistic terms for performance measures

3. Define If..then...rules

Expert
4. Define and test deffuzification methods

5. Tests and analysis of

results

Fig. 5. Method to develop the prediction model.

functions g~ : R3 -Rand h~ : R2 -R, so that 8x~ A R3 e y~ A R2 , p~ ¼ g~ ðxÞ 4.3. Supply chain reliability performance
e q ¼ h~ ðyÞ. The variable components x~ 2 and y~ 2 are x~ 21 : days sales
outstanding,x~ 22 : inventory days of supply,x~ 23 : days payable out- The reliability performance was modeled by the fuzzy function
standing, y~ 21 : asset turns and y~ 22 : net profit. f~ : R4 -R, so that, 8x~ A R4 , we have W
~ ¼ f~ ðz~ Þ in which the fuzzy
182 G.M.D. Ganga, L.C.R. Carpinetti / Int. J. Production Economics 134 (2011) 177–187

Supply Chain
Performance
Supply Chain

Reliability Flexibility Responsiveness Cost Asset

Performance
Atributes
Customer-Facing Internal-Facing

Upside Total Cost

Perfect Order
Flexibility
Order
Supply Chain COGS Cash to Cash ROA
SCOR Level 1
Fulfillment Fulfillment Metrics
Supply Chain Management

Orders in Upside Source Source Cycle Finance and Days Sales

Asset Turns
Full Flexibility Time Plannig Cost Outstanding

Delivery to Upside Make Make Cycle Inventory Days Payable

Net Profit
Commit date Flexibility Time Carrying Cost Oustanding

Perfect Upside Delivery Delivery Cycle IT Cost for Inventory Days

Condition Flexibility Time Supply Chain of Supply
SCOR Level 2
Metrics
Accurate Material
Documentation Acquisition Cost

Order
Management
Cost

Fig. 6. The proposed model for predicting SC performance.

Table 2
A sample of the pertinence functions and linguistic terms in the input variables

Level 1 metrics Unit Triangular membership functions and linguistic terms

Low Medium High

Accurate documentation % (84;1) (91;0) (84;0) (91;1) (98;0) (91;0) (98;1)

Delivery commit date % (85;1) (91.5;0) (85;0) (91.5;1) (98;0) (91.5;0) (98;1)
Orders in full % (85;1) (91.5;0) (85;0) (91.5;1) (98;0) (91.5;0) (98;1)
Perfect condition % (85;1) (91.5;0) (85;0) (91.5;1) (98;0) (91.5;0) (98;1)
Asset turns Turns (0.1;1) (1.55;0) (0.1;0) (1.55;1) (3;0) (1.55;0) (3;1)
Net profit % of revenue (0.1;1) (3.7;0) (0.1;0) (3.7;1) (7.3;0) (3.7;0) (7.3;1)
Inventory days of supply Days (27.6;1) (53.8;0) (27.6;0) (51.9;1) (80;0) (53.8;0) (80;1)
Days payable outstanding Days (30;1) (51;0) (30;0) (51;1) (72;0) (51;0) (72;1)
Days sales outstanding Days (25;1) (47.5;0) (27;0) (47.5;1) (70;0) (47.5;0) (70;1)
Delivery cycle time Days (1;1) (4;0) (1;0) (4;1) (7;0) (4;0) (7;1)
Make cycle time Days (1;1) (4;0) (1;0) (4;1) (7;0) (4;0) (7;1)
Source cycle time Days (1;1) (3.5;0) (1;0) (3.5;1) (6;0) (3.5;0) (6;1)
Upside deliver flexibility Days (10;1) (45;0) (10;0) (45;1) (80;0) (45;0) (80;1)
Upside make flexibility Days (10;1) (45;0) (10;0) (45;1) (80;0) (45;0) (80;1)
Upside source flexibility Days (10;1) (45;0) (10;0) (45;1) (80;0) (45;0) (80;1)
Finance and planning cost % of revenue (0.8;1) (1.3;0) (0.8;0) (1.3;1) (1.8;0) (1.3;0) (1.8;1)
Inventory carrying cost % of revenue (2.6;1) (4.15;0) (2.6;0) (4.15;1) (5.7;0) (4.15;0) (5.7;1)
IT cost for supply chain % of revenue (1.7;1) (2.7;0) (1.7;0) (2.7;1) (3.7;0) (2.7;0) (3.7;1)
Material acquisition cost % of revenue (5.2;1) (8.3;0) (5.2;0) (8.3;1) (11.4;0) (8.3;0) (11.4;1)
Order management cost % of revenue (8.2;1) (13.1;0) (8.2;0) (13.1;1) (18;0) (13.1;0) (18;1)
Cost of goods sold % of revenue (49;1) (59.5;0) (49;0) (59.5;1) (70;0) (59.5;0) (70;1)

variable z~ is obtained from the components of perfect order Components of the order fulfillment cycle time are: x~ 41 source
fulfillment ðx 3 Þ, which can be described by the fuzzy function cycle time, x~ 42 : make cycle time and x~ 43 : delivery cycle time.
g~ : R4 -R, so that z~ ¼ g~ ðx 3 Þ. The components of perfect order
fulfillment are: x~ 31 : delivery to customer commit date, x~ 32 : orders 4.5. Supply chain flexibility performance
delivery in full, x~ 33 : perfect condition e, x~ 34 : accurate
documentation. Flexibility performance was modeled by a fuzzy function f~ :
R4 -R, so that8ðx~ , z~ Þ A R4 ¼ R3  R, from which the flexibility
4.4. Supply chain responsiveness performance performance value is obtainedW ~ ¼ f~ ðy~ , z~ Þ, where the fuzzy vari-
able y 3 is obtained from the components of upside supply chain
~
Responsiveness performance was modeled by a fuzzy function flexibility, described by the fuzzy function g~ : R3 -R, so that
f~ : R3 -R , so that 8x~ A R3 , we have responsiveness performance y~ ¼ g~ ðx Þ:A fuzzy variable z~ is defined as overlap and the compo-
W ~ ¼ f~ ðz~ Þ, where the fuzzy variable z~ is obtained from the nents of upside supply chain flexibility are: y~ 3 ¼ ðy~ 31 , y~ 32 , y~ 33 Þ, in
components of order fulfillment cycle time ðx 4 Þ, which can be which y~ 31 is the upside source flexibility, y~ 32 , the upside make
described by the fuzzy function g~ : R3 -R, so that z~ ¼ g~ ðx 3 Þ. flexibility and y~ 33 , the upside delivery flexibility.
 G.M.D. Ganga, L.C.R. Carpinetti / Int. J. Production Economics 134 (2011) 177–187 183

Step 2: The definition of the universe of discourse (minimum Step 4: The defuzzification methods were defined empirically,
and maximum values) of each input or output variables was evaluating mainly the concept of continuity and discontinuity
based on benchmarking information from several sources such in fuzzification. Therefore, in most of the cases, the defuzzifi-
as Supply Chain Council, APQC and the Manufacturing Perfor- cation methods CoM and CoA were used because they are
mance Institute. In this work triangular pertinence functions considered continuous methods. The CoM method is the most
were used predominantly. Table 2 illustrates a sample of the recommended for supporting quantitative decisions, as it is
pertinence functions for the input variables, i.e., the level the case in our research. Table 4 shows a sample of the
2 metrics of SCOR. The pertinence functions are described pertinence functions for the response variables and defuzzifi-
according to the format proposed in Section 3. cation methods used.
Step 3: The definition of the rules of inference was based on Step 5: The model was tested using random numbers and
the causal relationships proposed by the SCORcard (SCOR, adopting the categories of performance proposed by the
2006). The linguistic terms (for example very low, low, SCORmark (advantage, parity and superior) (Francis, 2009).
medium, high and very high) of the output (lagging) metrics Also, the effect of the level of leading metrics on performance
were defined by the authors (acting as experts) as function of was analyzed using DoE techniques, particularly surface
the linguistic terms of the input (leading) metrics. For exam- response analysis, as discussed in Section 6.
ple, on the first line in Table 3, the lagging metric asset
performance was defined as low when the levels of the leading 5. Results
metrics cash to cash and return over assets were better and
very low respectively. With the aim of analyzing the response behavior generated by
the model, some experimental data was simulated by generating
Table 3 random values for the level 2 leading metrics of the SCOR model.
Rules of inference of the metric asset management. In order to do that, the random numbers were generated based on
IF THEN
expression 8, by a method of inverse transformation, for a
uniform, continuous distribution.
Cash to cash Return over assets Asset performance
X ’½a þ ðbaÞ  randomðsÞ ð5Þ
Better Very low Low The range of variation (maximum and minumum values) for
Better Low Medium
Better Medium High
random number generation of the leading metrics was based on
Better High Very high benchmarking information from several sources such as Supply
Better Very high Very high Chain Council, APQC and the Manufacturing Performance
Good Very low Medium Institute.
Good Low Medium
The same classification used by the Supply Chain Council,
Good Medium Medium
Good High High named Supply Chain SCORmark tool (SCC, 2009), was adopted to
Good Very high Very high analyze the results. It stratifies the process performance according
Medium Very low Very low to three positions:
Medium Low Low
Medium Medium Medium
Medium High Medium
 superior: is the performance (median value) on a specific
Medium Very high High indicator attained by 10% of the best classified SC’s in relation
Bad Very low Very low to the total of SCs surveyed;
Bad Low Low  advantage: is the performance (median value) among the top
Bad Medium Low
10 companies and the median of all the SCs studied;
Bad High Medium
Bad Very high Medium  parity: is the performance (median value) of all the SCs
Worst Very low Very low studied.
Worst Low Very low
Worst Medium Low
The objective was to analyze the limits of the proposed model,
Worst High Medium
Worst Very high High and therefore only the categories superior and parity were used.
Table 5 shows the results of these tests for the response variables

Table 4
A sample of the pertinence functions and defuzzification methods in the lagging measures.

Response variable Unit Defuzzification Triangular membership functions and linguistic terms
method
Very low Low Medium High Very high

Perfect order fulfillment % CoM (85;0) (87.2;0) (89.3;0) (91.5;0) (93.6;0)

(89.3;1) (89.3;1) (91.5;1) (93.6;1) (98;1)
(91.5;0) (93.6;0) (95.8;0)

Reliability supply chain % CoM (0;0) (87.2;0) (89.3;0) (91.5;0) (93.6;0)

(33.3;1) (89.3;1) (91.5;1) (93.6;1) (98;1)
(91.5;0) (93.6;0) (95.8;0)

Return over assets (ROA) % of revenue CoA (0.01;0) (3.7;0) (7.3;0) (11;0) (14.7;0)
(7.3;1) (7.3;1) (11;1) (14.7;1) (22;1)
(11;0) (14.7;0) (18.3;0)

Total SCM cost % of revenue CoM (24;0) (25.8;0) (27.7;0) (29.5;0) (31.4;0)
(27.7;1) (27.7;1) (29.6;1) (31.4;1) (35.1;1)
(29.6;0) (31.4;0) (33.2;0)
184 G.M.D. Ganga, L.C.R. Carpinetti / Int. J. Production Economics 134 (2011) 177–187

(lagging). The results obtained with the model are consistent Central Composite Design—CCD was used to design the experi-
when compared with the performance indicators obtained from ment, which uses a combination of factorial arrangements (com-
the benchmarking sources of the SCOR, which suggests that the plete or fractionated), mean and axial points.
fuzzy model was satisfactory in predicting performance in the As an example of the tests and analysis performed, the
supply chain. evaluation of the effect of the leading metrics Inventory Days of
Supply, Days Payable Outstanding and Days Sales Outstanding, on
the lagging metric Cash to Cash was carried out according to a 23
6. Statistical analysis of fuzzy prediction model responses factorial design plus six axial points (a ¼1.68179) and two central
points, as it can be observed in Table 6.
In order to understand the effect of the base of rules on the Fig. 7 shows the response surface for cash to cash metric as a
cause and effect relationships between the input variables (lead- function of Days sales outstanding and Days payable outstanding.
ing) and the output variables or responses (lagging), the Response Table 7 presents the results of Analysis of Variance (ANOVA) for a
Surface Methodology—RSM (Montgomery, 2005) was used to significance level p r0.05. By analyzing the p-value in Table 7,
design experiments and statistical analysis of the responses. it can be seen that effects on the response are statistically

Table 5
Results of the proposed prediction model.

Metrics Superior (top 10%) Advantage Parity (100%)

Median Mean Max Min Median Mean Max Min

Assets turns 2.60 2.65 2.88 2.52 1.99 1.38 1.43 2.88 0.14
Net profit 6.87 6.87 7.14 6.61 4.91 2.96 3.48 7.14 0.41
Return over assets (ROA) 14.56 15.17 17.98 13.57 12.74 10.93 10.38 17.98 2.91
Inventory days of supply 28.50 29.75 34.00 28.00 44.75 61.00 56.94 79.00 28.00
Days payable outstanding 35.50 34.75 37.00 31.00 43.75 52.00 52.97 72.00 31.00
Days sales outstanding 30.00 29.50 32.00 26.00 40.00 50.00 49.23 69.00 26.00
Cash to cash 28.69 29.01 34.05 24.60 41.72 54.75 51.84 74.91 24.60
Assets supply chain performance 69.14 70.86 79.76 65.39 54.83 40.51 43.92 79.76 14.16
Delivery cycle time 1.00 1.00 1.00 1.00 2.50 4.00 3.97 7.00 1.00
Make cycle time 1.00 1.25 2.00 1.00 2.00 3.00 3.63 7.00 1.00
Source cycle time 1.00 1.00 1.00 1.00 2.00 3.00 3.37 6.00 1.00
Order fulfillment cycle time 5.84 5.84 8.67 3.00 8.67 11.50 11.01 14.33 3.00
Responsiveness SC performance 83.76 85.00 99.99 72.48 66.88 50.00 53.47 99.99 26.84
Inventory carrying cost 2.78 2.77 2.88 2.64 3.39 3.99 4.01 5.55 2.64
IT supply chain cost 1.83 1.85 1.93 1.82 2.10 2.37 2.61 3.69 1.82
Finance and planning cost 0.86 0.86 0.90 0.81 1.07 1.27 1.25 1.70 0.81
Material acquisition cost 5.49 5.52 5.90 5.22 6.83 8.16 8.17 11.20 5.22
Order management costs 8.95 8.95 9.42 8.50 11.26 13.56 13.04 17.83 8.50
Total SCM costs 26.38 26.10 26.59 25.05 27.98 29.58 29.44 32.35 25.05
COGS 49.98 50.03 50.54 49.62 55.26 60.54 60.06 69.86 49.62
Cost supply chain performance 69.12 71.11 80.13 66.06 56.61 44.09 46.30 80.13 24.19
Upside deliver flexibility 14.00 14.25 16.00 13.00 30.50 47.00 44.77 73.00 13.00
Upside make flexibility 16.00 15.50 17.00 13.00 32.50 49.00 44.69 76.00 13.00
Upside source flexibility 13.50 13.75 18.00 10.00 27.75 42.00 43.34 80.00 10.00
Upside supply chain flexibility 83.56 88.47 106.13 80.63 109.28 135.00 132.19 168.37 80.63
Flexibility supply chain performance 74.49 72.16 75.89 63.75 62.25 50.00 51.34 75.89 34.11

Table 6
Experimental central composed planning for Cash to Cash.

Trials Non codified Codified Response (x2)

x~ 22 x~ 23 x~ 21 x~ 22 x~ 23 x~ 21

1 42.80 40.00 34.50  1.00000  1.00000  1.00000 47.93

2 42.80 40.00 60.50  1.00000  1.00000 1.00000 58.46
3 42.80 62.00 34.50  1.00000 1.00000  1.00000 36.97
4 42.80 62.00 60.50  1.00000 1.00000 1.00000 44.25
5 64.80 40.00 34.50 1.00000  1.00000  1.00000 56.36
6 64.80 40.00 60.50 1.00000  1.00000 1.00000 63.6
7 64.80 62.00 34.50 1.00000 1.00000  1.00000 42.13
8 64.80 62.00 60.50 1.00000 1.00000 1.00000 61.57
9 35.30 51.00 47.50  1.68179 0.00000 0.00000 43.93
10 72.30 51.00 47.50 1.68179 0.00000 0.00000 56.93
11 53.80 32.50 47.50 0.00000  1.68179 0.00000 74.67
12 53.80 69.50 47.50 0.00000 1.68179 0.00000 50.3
13 53.80 51.00 25.64 0.00000 0.00000  1.68179 41.34
14 53.80 51.00 69.36 0.00000 0.00000 1.68179 59.29
15 (C) 53.80 51.00 47.50 0.00000 0.00000 0.00000 50.3
16 (C) 53.80 51.00 47.50 0.00000 0.00000 0.00000 50.3

In which: ðx~ 22 Þ Inventory days of supply (days); ðx~ 23 Þ Days payable outstanding (days); ðx~ 21 Þ Days sales outstanding (days);
(x2) Cash to Cash (response) (days).
 G.M.D. Ganga, L.C.R. Carpinetti / Int. J. Production Economics 134 (2011) 177–187 185

significant for the metrics Inventory days of supply and Days Sales Payable Outstanding, with linear behavior and a negative effect
Outstanding for a linear adjustment model (po0.05), and Days on the response variable. By following the same logic, significant
Payable Outstanding presented (po0.05 for both linear and effects can be found in the metric Days Sales Outstanding, through
quadratic model). The highest effect on the model was Days a linear model and positive effect; and Inventory Days of Supply,

Fig. 7. Response surface for Cash to Cash as a function of Days sales outstanding and Days payable outstanding.

Table 7
ANOVA for Cash to Cash (R2 ¼0.92557; Adj: 0.87596).

Factors SS df MS F P (r0.05)

Inventory days of supply (L) 245.587 1 245.5871 20.29740 0.001478

Inventory days of supply (Q) 4.687 1 4.6868 0.38735 0.549135
Days payable outstanding (L) 497.354 1 497.3545 41.10558 0.000124
Days payable outstanding (Q) 116.811 1 116.8105 9.65421 0.012577
Days sales outstanding (L) 408.355 1 408.3552 33.74993 0.000256
Days sales outstanding (Q) 5.241 1 5.2414 0.43319 0.526902
Error 108.895 9 12.0994

Total SS 1463.135 15

Fig. 8. Level curve of Cash to Cash (days) in relation to Days sales outstanding (days) and Days payable outstanding (days).
186 G.M.D. Ganga, L.C.R. Carpinetti / Int. J. Production Economics 134 (2011) 177–187

Table 8 Although it is a quite different approach, when compared to

P-value of ANOVA of the SCOR model level metrics. other multi-criteria techniques used in supply chain performance
measurement such as AHP, ANP or fuzzy-AHP/ANP, some advan-
Lagging measures Leading measures P value ( o0.05)
tages can be pointed:
Perfect order fulfillment Accurate documentation 0.000202
Delivery to Commit Date 0.000372  It predicts performance on specific dimensions such as cost,
Orders in full 0.000112 responsiveness, reliability, etc. Calculation of a global indicator
Perfect condition 0.000111
of supply chain performance, as proposed by other techniques,
Order fulfillment cycle time Delivery cycle time 0.008983 because of aggregation of unrelated performance dimensions,
Make cycle time 0.008983
can be meaningless and useless to help managers in the
Source cycle time 0.229607
decision-making process.
Return over assets Asset turns 0.000021
 It is a predictive tool so it can help elucidating decision-
Net profit 0.000019
making problems that are faced by managers based on ‘‘what
Upside flexibility supply chain Upside source flexibility 0.070937 would happen ify’’;
Upside make flexibility 0.070624
Upside deliver flexibility 0.070624
 Using SCOR metrics enables comparisons of predictive results
with benchmarking information from several sources such as
Total cost supply chain Finance and planning cost 0.428424
Supply Chain Council, APQC and the Manufacturing Perfor-
management
Inventory carrying cost 0.091043 mance Institute.
IT cost for supply chain 0.340195
Material acquisition cost 0.000006
Order management cost 0.000000
Further research should be carried out so as to validate the
model with real data. Also, it would be interesting to customize
the model with the metrics of a specific supply chain, as well as
model the base of rules according to the point of view of the
also presenting linear behavior and a positive effect. Finally, the
supply chain managers of the same company.
metric Days Payable Outstanding presented a significantly positive
effect, and quadratic behavior. The analysis of the normal dis-
tribution of residues for the model Cash to Cash suggests that the References
residues can be evaluated as normally distributed, and therefore
the hypothesis of normality in the response is confirmed.
Agarwal, A., Shankar, R., Tiwari, M.K., 2006. Modeling the metrics of lean, agile and
Fig. 8 illustrates a level curve for the model cash to cash. The leagile supply chain: an ANP-based approach. European Journal of Operational
cash convertion time was ‘‘optimum’’ for a value below 32 days. Research 173, 211–225.
The optimum point occurred when the metrics Days sales out- APQC, Supply Chain Surveys. Available at: /www.apqc.orgS (accessed 10 June
2009).
standing and days payable outstanding performed 20 and 63 days Beamon, B.M., 1999. Measuring supply chain performance. International Journal of
respectively. Such performance perfectly illustrates the influence Operations & Production Management 19 (3), 275–292.
of the two variables with the greatest effects on the model, i.e., Beamon, B.M., Chen, V.C.P., 2001. Performance analysis of conjoined supply chains.
International Journal of Production Research 39 (14), 3195–3218.
Days sales outstanding and days payable outstanding. Bertrand, J.W.M., Fransoo, J., 2002. Operations management research methodolo-
The same procedure of experimentation and analysis of results gies using quantitative modeling. International Journal of Operations and
was performed for the other lagging variables (SCOR model level Production Management 22 (2), 241–264.
Bititci, U.S., Carrie, A.S., Mcdevitt, L., 1997. Integrated performance measurement
1 metrics as indicated in Fig. 6). Table 8 presents the significance systems. International Journal of Operations & Production Management,
level (p-value) for linear adjustment model of the other five level Bradford 17 (5), 522–534.
1 metrics. For all of them, it can be seen that most of the leading Bojadziev, G., Bojadziev, M., 1997. Fuzzy Logic for Business, Finance, and Manage-
ment. World Scientific, Danvers.
metrics has a significant effect on the lagging (level 1) metrics Bolstorff, P., Rosenbaum, R., 2007. Supply chain excelence. A handbook for
(po0.05). Dramatic Improvement Using the SCOR Model. AMACOM, NY.
Brewer, P.C., Speh, T.W., 2000. Using the Balanced Scorecard to measure supply
chain performance. Journal of Business Logistics 211, 75–93.
Cai, J., Liu, X., Liu, J., 2009. Improving supply chain performance management: a
7. Conclusions systematic approach to analyzing iterative KPI accomplishment. Decision
Support Systems 46, 512–521.
Following the research work by Unahabhokha et al. (2007), Chan, F.T.S., Qi, H.J., Chan, H.K., Lau, H.C.W., Ip, R.W.L., 2003. A conceptual model of
performance measurement for supply chains. Management Decision 41 (7),
this paper bridges an important gap using standardized and 635–642.
benchmarking metrics such as the SCOR model metrics to Chan, F.T.S., Qi, H.J., 2003. An innovative performance measurement method for
evaluate the application of fuzzy logic to develop a model to supply chain management. An International Journal Supply Chain Manage-
ment 8 (3/4), 209–223.
predict performance of supply chain lagging metrics based on Dubois, D., Prade, H., 1980. Fuzzy Sets and Systems. Academic Press, New York.
leading metrics and inference rules. Random data used to test the Francis, J., 2009. Keeping SCORs in Your Supply Chain: Benchmarking, Presenta-
model generated results which were shown to be very consistent tion Supply-Chain Council. Available at: /http://www.supply-chain.org/bench
marking/presentationS (accessed 15 June 2009).
when compared to the scorecards of many companies that use the Gunasekaran, A., Patel, C., Mcgaughey, R.E., 2004. A framework for supply chain
SCORmark as a standard for benchmarking. Also, analysis of performance measurement. International Journal of Production Economics 87
variance (ANOVA, with p o0.05) show the significance of the (3), 333–347.
Gunasekaran, A., Patel, C., Tirtiroglu, E., 2001. Performance measures and metrics
effects of most of the leading metrics on the responses, confirm- in a supply chain environment. International Journal of Operations & Produc-
ing the relevance of the causal relationships embedded in tion Management 21 (1/2), 71–87.
the model. Holmberg, S., 2000. A system perspective on supply chain measurements. Inter-
national Journal of Physical Distribution & Logistics Management 30 (10),
In general, the findings reinforce the proposition that the
847–868.
adoption of a predictive model based on fuzzy logic and on Kaplan, R.S., Norton, D.P., 1992. The Balanced Scorecard — Measures That Drives
metrics of the SCOR model seems to be a feasible approach to Performance. Harvard Business Review, Boston, pp. 71–79.
predict performance of supply chains so as to help managers in Kleijnen, J.P.C., Smits, M.T., 2003. Performance metrics in supply chain manage-
ment. Journal of the Operational Research Society 54 (5), 507–514.
the decision making process of managing performance of supply Klir, G.J., Yuan, B., 1995. Fuzzy Sets and Fuzzy Logic: Theory and Applications.
chains. Prentice Hall, Upper Saddle River, NJ.
 G.M.D. Ganga, L.C.R. Carpinetti / Int. J. Production Economics 134 (2011) 177–187 187

Lapide, L., 2000. What about measuring supply chain performance? Achieving Shepherd, C., Günter, H., 2006. Measuring supply chain performance: current
Supply Chain Excellence through Technology, vol. 2, pp. 287–297. research and future directions. International Journal of Productivity and
Lee, H.L., Billington, C., 1993. Material management in decentralized supply chain. Performance Management 55 (3/4), 242–258.
Operations Research 41 (5), 835–847. Supply Chain Operations Reference Model SCORs Version 8.0, 2006. Supply-Chain
Liao, S.H., 2005. Expert system methodologies and applications: a decade review Council. Available at: /http://www.supply-chain.org/resources/scor/8.0S
from 1995 to 2004. Expert Systems with Applications 28 (1), 93–103. (accessed 3 June 2009).
Lin, C.T., Chiu, H., Chu, P.Y., 2006. Agility ı́ndex in the supply chain. International Supply-Chain Council (SCC). Available at: /http://www.supply-chain.orgS
Journal Production Economics 100, 285–299. (accessed 3 June 2009).
Lin, H.Y., Hsu, P.Y., Sheen, G.J., 2007. A fuzzy-based decision-making procedure for Unahabhokha, C., Platts, K., Tan, K.H., 2007. Predictive performance measurement
system. A fuzzy expert system approach. An International Journal Benchmark-
data warehouse system selection. Expert Systems with Applications 32 (3),
ing 14 (1), 77–91.
939–953.
Varma, S., Wadhwa, S., Deshmukh, S.G., 2008. Evaluating petroleum supply chain
Liu, S.C., Liu, S.Y., 2001. An efficient expert system for air compressor trouble-
performance: application of analytical hierarchy process to balanced score-
shooting. Expert Systems 18 (4), 203–214.
card. Asia Pacific Journal of Marketing & Logistics 20 (3), 343–356.
Manufacturing Performance Institute. Available at: /www.mpi-group.netS Wu, W.W., Lee, Y.T., 2007. Developing global managers’ competencies using the
(accessed 8 June 2009). fuzzy DEMATEL method. Expert Systems with Applications 32 (2), 499–507.
Montgomery, D.C., 2005. Design and Analysis of Experiments: Response Surface Yeh, D.Y., Cheng, C.H., Chi, M.L., 2007. A modified two-tuple FLC model for
Method and Designs. John Wiley and Sons, New Jersey. evatuating the performance of SCM: by the six sigma DMAIC process. Applied
Park, J.H., Lee, J.K., Yoo, J.S., 2005. A framework for designing the balanced supply Soft Computing 7, 1027–1034.
chain scorecard. European Journal of Information Systems 14 (4), 335–346. Zadeh, L.A., 1978. Fuzzy sets as basis for a theory of possibility. Fuzzy Sets and
Ross, T.J., 1997. Fuzzy Logic with Engineering Applications. McGraw-Hill, New York. Systems 1 (1), 3–28.
Sharma, M.K., Bhagwat, R., 2007. An integrated BSC-AHP approach for supply chain Zimmermann, H.J., 1991. Fuzzy Set Theory and Its Applications. Kluwer Academic,
management evaluation. Measuring Business Excellence 11 (3), 57–68. Boston.