International Journal of Information Management Data Insights 4 (2024) 100279

Contents lists available at ScienceDirect

International Journal of Information Management Data

Insights
journal homepage: www.elsevier.com/locate/jjimei

How can Artificial Intelligence (AI) be used to manage Customer Lifetime

Value (CLV)—A systematic literature review
Edo Belva Firmansyah a , Marcos R. Machado b ,∗, João Luiz Rebelo Moreira a
aUniversity of Twente, Faculty of Electrical Engineering, Mathematics, and Computer Science, AE Enschede, 7500, Netherlands
b
University of Twente, Faculty of Behavioural, Management and Social Sciences, Department of High-Tech Business and Entrepreneurship, AE
Enschede, 7500, Netherlands

ARTICLE INFO ABSTRACT

Keywords: Customer Lifetime Value (CLV) represents the total worth of a customer to a company over time, aiding
Customer value businesses in resource allocation and tailored marketing for profitability. This literature review fills a research
Risk-Adjusted Revenue gap by examining how customer risk factors are integrated into CLV calculations. We conducted a systematic
Information systems
literature review across databases, adhering to strict criteria for relevance and quality. The review analyzed
Machine learning
CLV methodologies and outcomes, highlighting the use of mean–variance analysis to optimize customer
Systematic literature review
portfolios, with customer income fluctuations identified as a major risk factor. The study also explores the
evolution of CLV research, particularly in the application of Machine Learning (ML) for risk-adjusted CLV. Our
findings offer a comprehensive overview, laying the groundwork for future research and helping businesses
refine risk management strategies, identify high-risk customers, and enhance customer value through more
dynamic, data-driven models.

1. Introduction utilized to assess a company’s financial performance and approximate

its valuation (Gupta et al., 2004; Hogan et al., 2002).
Customers are the fundamental basis of any business. They func- Measuring CLV is a crucial undertaking for both researchers and
tion as the primary source of income, while also serving as a vital professionals in the marketing area. Several approaches have been
resource that impacts the company’s performance, profitability, and ex- devised to forecast and evaluate CLV in different scenarios and sectors.
pansion. This comprehension has instigated a transformation in corpo- These methods include determining CLV for the purpose of allocating
rate methodologies toward a more customer-focused strategy, resulting marketing resources in the airline industry (Rust, Lemon, & Zeithaml,
in the creation of Relationship Marketing (RM) in the 1980s (Gupta, 2004), in business-to-business settings (Kumar, Venkatesan, Bohling,
Lehmann, & Stuart, 2004; Kumar & Reinartz, 2016). The objective of & Beckmann, 2008; Venkatesan & Kumar, 2004), and in the online
RM is to allure, retain, and improve client relationships (Ryals & Knox, retail market (Dahana et al., 2019). These strategies often rely on
2005). It is crucial for the organization to ascertain the worth of their fundamental assumptions on customer retention rate and profit margin,
clients and design suitable measures to ensure their continued loyalty. while also taking into account the costs associated with customer
The client Lifetime Value (CLV) is the predominant metric utilized acquisition, retention, and servicing. Although this approach has been
to assess the value of a client to a firm (Glady, Lemmens, & Croux,
extensively utilized and demonstrated its utility in various contexts, it
2015; Gupta et al., 2004; Kumar & Reinartz, 2016). Customer Lifetime
does possess significant constraints. A significant drawback of CLV is its
Value (CLV) is a metric that quantifies the worth of each customer by
assumption that a corporation remains unresponsive to market changes
considering their historical, current, and anticipated future interactions
once it has made an investment in a client, which is not a realistic
with the organization (Gupta et al., 2004; Kumar & Reinartz, 2016).
scenario (Méndez-Suárez & Crespo-Tejero, 2021). The CLV calculation
Consequently, CLV has gained importance in Revenue Management
does not take into account the variations in client preferences, such
(RM) as it enables the identification of the most valuable customers and
as churn or a decrease in product/service purchases from the com-
serves as a vital measure for crafting customized marketing strategies
pany. These alterations pose a potential hazard that may impact the
for each individual customer (Dahana, Miwa, & Morisada, 2019; Kumar
& Reinartz, 2016; Qi, Zhou, Chen, & Qu, 2012). In addition, CLV can be customer’s future profitability. Hence, in order to achieve a precise

∗ Corresponding author.
E-mail addresses: [email protected] (E.B. Firmansyah), [email protected] (M.R. Machado), [email protected]
(J.L.R. Moreira).

https://doi.org/10.1016/j.jjimei.2024.100279

Available online 3 September 2024

2667-0968/© 2024 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

assessment of customer value, it is imperative that valuation methods and customer segmentation techniques, that support the inclusion of
consider not just customer profitability but also the accompanying customer risk. Additionally, it can provide insight into the influence
risks (Buhl & Heinrich, 2008; Yun & Yan, 2013). Therefore, it is essen- of customer risk on the long-term viability of an organization and the
tial to consider risk considerations when evaluating customer value, as management of customer relationships. (2) Methods for Incorporating
neglecting to include these variables in the CLV calculation may result Customer Risk into Customer Value Calculation: This section of the
in inaccurate evaluations of customer value. evaluation will concentrate on identifying, classifying, and assessing
While there has been extensive study on the computation of CLV, various approaches and models used in different industries to integrate
there is a lack of studies that have considered the impact of risk on customer risk into customer value assessments. The scope of this might
this calculation (Singh, Thakur, & Sharma, 2016). An influential study encompass several methodologies, ranging from quantitative models
in this field was conducted by Dhar and Glazer (2003), who empha- such as credit scoring and risk assessment algorithms, to qualitative
sized that numerous organizations neglect to assess if all their prized techniques like consumer feedback and behavior analysis. This explo-
clients are collectively advantageous from a risk standpoint. In order ration can facilitate comprehension of industry-specific strategies and
to fill this void, Dhar and Glazer (2003) introduced a novel measure evaluate the efficacy of these techniques in diverse circumstances. (3)
called the risk-adjusted lifetime value (RALTV), which considers the Classification of Risks and Customer Segments in Different Industries:
influence of risk on customer worth. Following that, additional scholars This section of the evaluation can offer a comprehensive examination
have endeavored to include the influence of risk when determining of the many sorts of hazards linked to different consumer groups across
customer value. For instance, Ryals (2003) and Ryals and Knox (2005) diverse industries. For instance, the evaluation could examine the
investigated the risk of churn and the volatility of future revenue in an financial risks associated with banking, the risks involved in service de-
insurance company. Similarly, Tarasi, Bolton, Hutt, and Walker (2011) livery in e-commerce, or the health-related risks in the pharmaceutical
explored the variability of revenues generated by segments as a risk industry. Gaining a comprehensive understanding of these risks enables
factor in a business-to-business (B2B) context. organizations to customize their strategy in order to effectively manage
Considering that customer risk is present in every industry, albeit and reduce specific hazards connected with various client segments.
in varying manifestations, it is crucial to include the influence of Furthermore, the rationale for conducting this study is strongly em-
risk when calculating customer value in order to generate precise phasized by notable deficiencies in current scholarly works, specifically
predictions. The progress in technology and the increase in big data with the incorporation of customer risk indicators into evaluations of
have made it easier to gather and analyze large amounts of customer Customer Lifetime Value (CLV). Recent study emphasizes the urgent
data. Consequently, Customer Relationship Management (CRM) has requirement for more sophisticated valuation models that consider the
experienced a substantial metamorphosis, enabling firms to gain a ever-changing nature of consumer interactions and market situations.
more accurate and actionable understanding of customer behavior. Nevertheless, there is a significant lack of research that comprehen-
Furthermore, the advent of Machine Learning (ML) has brought up sively examine the integration of risk into customer lifetime value
novel possibilities for forecasting the influence of customer risk on (CLV) calculations, especially in dynamic industries such as telecoms.
value. To enhance customer portfolios, it is crucial to modify and This gap not only constrains the theoretical comprehension but also
expand current models by integrating data-driven methods that allow impedes the actual implementation of CLV models in real-world situa-
for more accurate forecasts of future customer worth. tions where client behavior is progressively erratic. This study aims to
This study does a comprehensive analysis of existing literature to fill these gaps by providing a thorough examination of the techniques
examine the methodology used to assess customer value while con- used to evaluate customer value from an integrated risk standpoint. It
sidering customer risk. This study will analyze the factors that led will also discuss recent developments and ongoing difficulties in this
to the acceptance of this approach, the specific sectors in which it field.
has been implemented, the different types of risks considered, and the Moreover, the significance of this study is heightened by recent
methodologies employed in the models. In addition, this research seeks discussions in literature advocating for the incorporation of risk in-
to combine the techniques and elements associated with risk that have dicators, such as customer attrition, revenue volatility, and changes
been identified and utilized in other studies. It intends to present, to in consumption patterns, into standard CLV models. Dhar and Glazer
the best of our understanding, the initial comprehensive analysis of (2003), Ryals (2003) have proposed a risk-aware approach to customer
a methodology for determining the CLV while considering the cus- assessment, emphasizing its significance as a crucial subject for fu-
tomer risk component. In order to address the research objective, this ture investigation. However, there is still a lack of empirical research
study aims to investigate the following research question (RQ) using a that effectively define and put into practice these notions, especially
Systematic Literature Review (SLR): How to incorporate customer risk in industries that are undergoing significant technical advancements
while assessing customer value in the (telecommunication) industry? and facing heightened market competition. This analysis addresses
The sub research questions addressed in this literature review are: these concerns by analyzing the present state of risk-adjusted cus-
tomer lifetime value (CLV) models, offering valuable insights into their
1. What is the rationale for including customer risk in the calcula- effectiveness and suitability. By undertaking this action, it not only
tion of customer value? addresses a significant gap in academic knowledge but also provides
2. What are the prevailing techniques for integrating customer risk practical advice for organizations seeking to improve their customer
into the computation of customer value in different industries? relationship management strategies in response to these dangers.
3. When evaluating customer value in various industries, what This systematic literature evaluation aims to address the theoretical
are the most important types of risks associated with distinct and practical gaps that have been discovered in recent discussions. The
customer segments? process methodically reviews the integration of numerous risk factors in
current models, evaluates their performance in diverse industries, and
Performing a systematic examination of existing literature about the identifies potential areas for further research. The study aims to provide
incorporation of customer risk into the calculation of customer value a comprehensive resource for both academics and practitioners by thor-
has the potential to contribute within several domains: (1) Compre- oughly evaluating these elements. It seeks to guide the development of
hending the Justification for Incorporating Customer Risk: This facet customer valuation frameworks that are more robust and predictive.
of the study can offer valuable insights into the reasons why firms The primary objective is to cultivate a more profound comprehension
and industries regard customer risk as an essential determinant of of how risks can be efficiently controlled and included into the financial
customer value. The review can examine many theoretical frameworks assessment of clients, therefore improving the strategic decision-making
and practical justifications, such as risk mitigation, profit maximization, procedures inside enterprises.

2
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

The remaining organization of this paper is as follows: Section 2 out- This SLR employs Scopus1 and Web of Science (WoS)2 as its main
lines the research technique, which encompasses the scientific databases sources of information, both of which are prominent academic databases.
that were explored, the specific keywords employed for the searches, The selection of these databases was based on their capacity to offer
the criteria employed for study selection, and the process of picking extensive coverage of scholarly literature, encompassing both contem-
studies. Section 3 provides a comprehensive presentation of the study’s porary and historical articles that are pertinent to this subject (Harzing
findings and analysis. This includes the process of extracting, synthe- & Alakangas, 2016). Scopus and WoS are highly esteemed as rep-
sizing, and evaluating data in order to address the research question. utable sources for academic research due to their implementation of
Section 4 concludes with an in-depth analysis of the findings and offers stringent evaluation procedures, which guarantee the inclusion of only
suggestions for further research. top-notch articles in their databases (Baas, Schotten, Plume, Côté, &
Karimi, 2020; Pranckute, ̇ 2021). Moreover, both databases are indexed
2. Theoretical background & motivation and abstracted, indicating that they include bibliographic details and
article summaries rather than the complete text (Pranckute, ̇ 2021).
Relationship Marketing (RM)’s Customer Lifetime Value (CLV) the- These databases provide advanced search tools that enable users to
ory has evolved since the 1980s. Increasing emphasis on client relation- apply specific criteria, facilitating the discovery of pertinent literature.
ships and their value has characterized this progression. The premise In general, Scopus and WoS are suitable sources for this SLR study
that consumers are the basis of business operations has led this change since they offer extensive coverage and employ rigorous evaluation
from transactional to relationship-focused methods. CLV calculation is procedures.
essential to assess a customer’s lifetime value to a company. Strategic Moreover, the search query utilized in the database is constructed
resource allocation, marketing, and corporate growth decisions depend using a collection of keywords that are relevant to the research in-
on such evaluations (Kumar & Reinartz, 2016; Ryals & Knox, 2005). quiries. The primary aim of this SLR is to investigate the accuracy
Adding risk elements to CLV calculation is a major achievement. of predicting customers’ value by considering customers’ risk. Conse-
Traditional CLV models ignored consumer behavior variability and quently, all possible versions of the name will be incorporated into
volatility, which could dramatically impact profitability estimates. Dhar the search criteria. Nevertheless, abbreviations will be excluded as
and Glazer (2003)’s risk-adjusted lifetime value (RALTV) highlighted they have the potential to provide irrelevant outcomes or distort the
the need to account for customer value trajectory downturns owing intended significance of the inquiry. Moreover, it can be inferred that
to risks including turnover and purchase frequency fluctuations. Many any papers or journal articles that employ abbreviations have already
studies have broadened this concept to include other industries and risk been included in the search results, as the complete name has been
categories, recognizing the complexity of customer interactions and the included in the query. The search strings used for all sources consist of
changing nature of market situations (Ryals, 2003; Tarasi et al., 2011). the following strings:
Such theoretical advances highlight the need to integrate statistical
models and qualitative assessments into CLV assessments to better 1. (‘‘Customer’’ AND (‘‘Risk Adjusted Revenue’’ OR ‘‘Risk adjusted
reflect real-world complexities and provide a more comprehensive lifetime value’’ OR ‘‘Risk-adjusted Lifetime value’’))
perspective of customer value. 2. (‘‘Customer Lifetime Value’’ OR ‘‘Customer Value’’ OR ‘‘Cus-
This systematic literature analysis also examines how recent studies tomer Portfolio’’ OR ‘‘Customer Asset’’) AND (‘‘Optimization’’
across sectors have refined and expanded customer risk-based CLV cal- OR ‘‘Risk-Adjusted’’ OR ‘‘Risk Adjusted’’)
culation methods. This review analyzes the evolution of CLV assessment
procedures and highlights gaps and prospective areas for theoretical The two search strings were merged to obtain all possible combi-
and practical breakthroughs by reviewing recent research approaches. nations of results. To enhance the precision of the search result, the
Exploring industry-specific risks and how they affect customer value search query is employed in the article’s title, abstract, and keywords.
assessments will reveal CLV models’ adaptability and applicability.
This method supports academic and practical discourse on improving 3.2. Selection phase
customer valuation models in light of changing market conditions and
customer behaviors. Only peer-reviewed studies published in respectable journals or con-
ference proceedings were included to guarantee the research’s quality
3. Systematic literature review methodology and dependability. The papers complied with the prescribed formatting
criteria, which encompassed comprehensive author identification and
The technique of Systematic Literature Review (SLR) is a methodical publication details. The absence of a constraint on the publication year
and clear approach that enables researchers to acquire a more pro- was intended to offer a thorough and all-encompassing examination
found comprehension of previous studies in their field and recognize of the subject matter. The study only considered papers that were
areas that have not been adequately addressed, as well as promising published in English, while those published in languages other than
directions for future study (Kraus et al., 2022). The process generally English were eliminated. Furthermore, articles without a distinct cor-
consists of four stages, commencing with the identification of pertinent relation to the research inquiries of this study, as evident from their
databases and keywords for the search procedure. Next comes the title, abstract, or content, were also eliminated. Excluded were items
selection step, during which inclusion and exclusion criteria are defined that were duplicates, meaning they had the same titles or content, and
and implemented to choose relevant studies. The third phase entails were identified in several databases. Ultimately, the study did not take
the extraction and integration of data from chosen research, while the into account publications that were either incomplete or unduly brief.
fourth phase centers on interpreting the results to discern patterns and Table 1 presents a concise overview of the criteria for including and
themes within the data (Mengist, Soromessa, & Legese, 2020). excluding data that were used in this study throughout the SLR.
In order to ensure the pertinence of the papers to this study and
3.1. Search phase to avoid wasting time reading irrelevant publications, the collected
articles underwent a comprehensive evaluation process. This proce-
The search phase, which is the initial step in the SLR process, is dure entailed several stages, commencing with the execution of the
of utmost importance in assessing the overall efficacy and credibility
of the review. This phase involves various essential duties, such as
finding appropriate databases for obtaining research and determining 1
www.scopus.com
2
the keywords to be used in the search process. www.webofscience.com

3
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

Table 1 4. Main findings

SLR - Inclusion and exclusion criteria.
Inclusion criteria Exclusion criteria This section organizes the findings of this study thematically, fol-
Studies published in conferences Papers that are not complete lowing the structure proposed by Varsha, Chakraborty, and Kar (2024),
proceeding and journals
Votto, Valecha, Najafirad, and Rao (2021), and used by Amato, Os-
Adhere to proper formatting guidelines, Studies that are not related to the
including complete author identification main RQ from title, abstract, and
terrieder, and Machado (2024), Natarajan et al. (2024). We begin by
and publication information content summarizing the key quantitative and qualitative aspects of the ana-
English-based peer-reviewed studies Duplicate articles by title or lyzed literature. Next, we review the definitions and features relevant
content to this field. We then examine the various sources of customer risk
Studies published until Dec 2024 Articles eliminated based on the
and the time horizons employed in different modeling frameworks, as
quality of the content.
Studies limited to document type of well as the areas and methods used in the different application cases.
journal articles, that includes abstract, Subsequently, we discuss how AI (specifically, machine learning) is
and includes full-text utilized in prescriptive or predictive frameworks for assessing Risk-
Adjusted Customer Lifetime Value (CLV) labels. Finally, we conclude
this section by highlighting the main contributions and limitations of
this systematic literature review (SLR), and by defining a research
specified search queries on each scientific database, and subsequently
agenda for the field
implementing the inclusion and exclusion criteria outlined in Table 1.
The redundant research from various databases were eliminated, and
4.1. Quantitative & qualitative summary of the outputs of this study
the irrelevant studies were rejected based on the evaluation of their
title and abstract. Subsequently, the remaining research were assessed
The findings of the SLR conducted in this study identified publi-
by thoroughly analyzing the complete text and excluding studies that
cations from 2003 to 2022, as shown in Fig. 2. The first study that
were brief, excessively broad, or lacking in completeness. Ultimately,
was recognized was undertaken by Dhar and Glazer (2003) in the
the researchers carefully chose the main papers that were utilized in
field of health economics and decision sciences. An average of one
this study.
to two publications were published per year on the issue, with a
After completing the evaluation process, 22 articles (16 from Sco-
total of four studies identified in 2013. Following 2016, there was
pus, and 6 from WoS) were chosen from a total of 386 articles discov-
ered in the initial round. The SLR selection flow chart utilized in this a period of five years during which no publications were released.
investigation is illustrated in Fig. 1. It is important to highlight that However, in 2021, one study was identified, and an additional two
this selection step, reducing from 386 to 22 papers, is done similarly studies were created by Machado and Karray (2022a) and Machado
to the methodology approached by Behera, Bala, and Dhir (2019). We and Karray (2022b). The preponderance of the research was published
read all titles and abstracts, and based on quality and relevance of the in journals focused on marketing and management. Specifically, 41%
content to the research questions of this study, we include papers in of the studies were published in marketing journals such as the Jour-
our review/data extraction phase. nal of Marketing and the Journal of Marketing Research, while 36%
were published in management publications such as the Management
3.3. Data extraction Research Review. The remaining papers were published in alternative
venues, such as the European Journal of Operational Research and
Following the selection of important publications, the SLR study Decision Support Systems journal. The early research done between
proceeded to its third phase, which involved extracting and synthe- 2003 and 2011 obtained a substantial amount of citations, with most
sizing data to obtain information that is relevant for addressing the cases reaching 100. Citations are frequently employed as a metric to
research questions. This entailed a comprehensive examination of all gauge the utility, significance, or sway of a publication. The quantity
papers, employing a research framework specifically crafted to address of citations garnered by an article is sometimes seen as a measure of
the research inquiries. Table 2 displays the 22 chosen papers, together its impact (Aksnes, Langfeldt, & Wouters, 2019). Table 3 provides a
with the employed research methodology and the findings of the study. comprehensive analysis of how citations are distributed.
The research methodologies employed were Literature Review (LR), Ex- A citation network analysis was performed on the results of the SLR
periment (E), and Comparative Study/Validation Study (CS/VS), each and is illustrated in Fig. 3. This analysis was conducted to investigate
of which yielded different output categories, namely Theoretical (T), the connections between the different studies. The size of each circle
Conceptual Model (CM), and Empirical findings (ER). Most of the litera- in this graphic represents the number of papers that reference the
ture included theoretical viewpoints on the topic area. The studies were corresponding paper within the network. Moreover, the arrow lines
categorized as theoretical (T) if they addressed overarching concepts, serve to illustrate network correlation: the paper located at the head
advantages, downsides, or design principles for including risk in the of the arrow is the one that has been cited, whereas the document at
calculation of consumers’ value. In addition, certain research presented the tail of the arrow is the citing paper. The analysis indicates that the
a conceptual model (CM) that enhanced the theoretical framework by study conducted by Dhar and Glazer (2003) has the highest level of
offering visual representations of the design choices embedded inside influence among the results of the SLR. Out of the 21 papers included
the model. All of the studies also generated empirical results (ER) in the SLR, 17 of them cite Dhar and Glazer (2003). The research
through the use of calculations (C), predictions (P), or both (CP) to conducted by Tarasi et al. (2011) and Buhl and Heinrich (2008) are
determine consumers’ value, taking into account risk. Furthermore, this of great significance and have been referenced by 9 and 8 studies,
study generated Risk-adjusted Revenue (RAR) estimates through both respectively. Co-citation analysis reveals a strong association among
calculation and prediction methods. these three studies, as 8 out of 9 papers that Tarasi et al. (2011) and
In addition to the main search phase and data extraction, we 7 out of 8 studies that Buhl and Heinrich (2008) also Dhar and Glazer
conducted searches on Web of Science and Scopus using the same (2003). The presence of a robust correlation and emphasis on research
search queries (Section 3.1), but specifically targeted papers published is evident among the three most significant studies (Annarelli, Battis-
in 2023 and 2024, this year, to complement our results. Initially, we tella, Nonino, Parida, & Pessot, 2021). Five studies neither referenced
collected 30 papers in total, but after reviewing the titles and abstracts, nor were referenced by the other research, with the majority of them
we included 9 (all from Scopus) papers in the most recent literature. being published after 2011. Two papers, namely Hai-wei et al. (2006)
Their main findings are also comprehensively discussed and presented and Petersen and Kumar (2015), were identified as not referencing and
in Section 4.1.2. referencing the other research identified in the SLR.

4
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

Fig. 1. SLR - Selection flow chart.

Fig. 2. Number of studies by year of publication.

The primary theme is assessed by analyzing the most frequent words articles commonly feature the phrases ‘‘customer’’, ‘‘risk’’, ‘‘portfo-
in the abstracts and keywords of the studies. The title of selected lio’’, ‘‘value’’, and ‘‘model’’. As for the keywords, the most frequent

5
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

Fig. 3. Literature review - Citation network.

Table 2 The studies mentioned in Table 3 were further explored, and in-
An overview of research methods and outputs of the different studies in the area. formation was extracted to find the relevant evidence to answer the
Reference Research Method(s) Output research question. The detailed results of the data extraction are shown
LR E CS/VS T CM ER in Table 4, which will be structured with the following elements:
Dhar and Glazer (2003) ✓ ✓ C Reference, Research theme (the key aim of the study), Industry setting
Ryals (2003) ✓ ✓ C of the study, Type of risk used in CLV calculation, Methodology used in
Ryals and Knox (2005) ✓ ✓ ✓ C the study, Time frame of the data; and Total observed customer/data.
Wangenheim and Lentz (2005) ✓ ✓ ✓ CP
Hai-wei, Ming-hui, and Ya-lin (2006) ✓ P
Ryals and Knox (2007) ✓ ✓ ✓ ✓ C
4.1.1. Studies throughout time
Buhl and Heinrich (2008) ✓ ✓ ✓ C The findings of the SLR indicate that most of the research that
Buhl and Heinrich (2008) ✓ ✓ ✓ CP include customers’ risk in CLV calculations are based on principles from
Sackmann, Kundisch, and Ruch (2010) ✓ ✓ ✓ C financial portfolio theory. This collection of studies is based on the
Tarasi et al. (2011) ✓ ✓ ✓ C
research conducted by Nobel Laureate William Sharpe, who is known
Ruch and Sackmann (2012) ✓ ✓ C
Singh, Murthi, and Steffes (2013) ✓ ✓ ✓ C for his contributions to Modern Portfolio Theory (MPT) and Capital
Juhl and Christensen (2013) ✓ ✓ ✓ C Asset Pricing Model (CAPM) (Sharpe, 1964). Additionally, these studies
Albadvi and Norouzi (2013) ✓ CP also depend upon the work of Markowitz (1952), who is linked with
Yun and Yan (2013) ✓ C portfolio selection theory (PST). MPT is founded on the principle of
So, Thomas, Seow, and Mues (2014) ✓ P
constructing portfolios of shares and suggests that investors should
Petersen and Kumar (2015) ✓ ✓ C
Norouzi and Albadvi (2016) ✓ ✓ ✓ CP strive to maximize their profits while maintaining a specific degree
Singh et al. (2016) ✓ ✓ ✓ C of risk. The CAPM is based on the premise that investors exhibit risk
Viviani, Komura, and Suzuki (2023) ✓ ✓ ✓ ✓ P aversion, meaning they require greater returns in exchange for taking
Machado and Karray (2022a) ✓ P
on increasing levels of risk. The CAPM also posits that all assets possess
Machado and Karray (2022b) ✓ ✓ ✓ CP
two separate categories of risk: systematic and unsystematic. Systematic
risk, as defined by Sharpe (1964), is a type of risk that impacts the
entire market and all assets within it. On the other hand, unsystematic
terms include ‘‘customer relationship management’’, ‘‘customer port- risk is specific to a single asset or a small group of assets. The CAPM
folio management’’, ‘‘risk’’, ‘‘customer lifetime value’’, and ‘‘customer illustrates that the risk specific to individual assets can be mitigated
by maintaining a portfolio that is well-diversified. However, the risk
portfolio’’. The findings indicate that most of the research concentrates
associated with the overall market, known as systematic risk, cannot
on examining the correlation between ‘‘risk’’ and ‘‘customer’’, with a be avoided by diversification (Buhl & Heinrich, 2008). Moreover, in
specific emphasis on ‘‘customer relationship management’’, ‘‘customer the context of the CAPM, systematic risk is quantified using beta, which
portfolio management’’, and ‘‘customer lifetime value’’. denotes the degree of responsiveness of an asset’s returns to fluctuations

6
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

Table 3
SLR - Number of citations.
Number of citations Number of studies List of studies
0–50 14 Hai-wei et al. (2006), Juhl and Christensen (2013), Machado
and Karray (2022a, 2022b), Norouzi and Albadvi (2016),
Ruch and Sackmann (2012), Sackmann et al. (2010), Singh
et al. (2013, 2016), Viviani et al. (2023), Wangenheim and
Lentz (2005)
50–100 1 Buhl and Heinrich (2008)
100–150 2 Ryals and Knox (2005), Tarasi et al. (2011)
150–200 3 Homburg, Steiner, and Totzek (2009), Petersen and Kumar
(2015), Ryals (2003)
>200 2 Dhar and Glazer (2003), Ryals and Knox (2007)

in the broader market. Therefore, in the studies that utilize CLV models have combined customer portfolio theory and customers’ segment dy-
containing risk and are based on financial portfolio theory, consumers namic theory to suggest that static portfolio models may inaccurately
are seen as risky assets. Each customer or customer group has a unique assess the value of top-tier customers and underestimate the value of
level of risk that must be effectively managed in order to maximize bottom-tier customers by neglecting the changing value dynamics in
returns. customer relationships. Viviani et al. (2023) have expanded upon this
The PST has been utilized in other research projects for client port- strategy to ascertain the most favorable client composition inside the
folio management, including Sackmann et al. (2010) for e-commerce, hospitality sector.
Tarasi et al. (2011) and Juhl and Christensen (2013) for B2B settings.
The authors Norouzi and Albadvi (2016) have enhanced previous re- 4.1.2. Recent literature (2023–2024)
search by proposing a hybrid model that integrates stochastic CLV Recent research (2023–2024) shows that customer value evaluation
modeling and ex-ante customer portfolio optimization. This model approaches are evolving toward more dynamic, risk-aware strategies
predicts future return streams from customers and incorporates these across industries. Researchers are using advanced data analytics to
predictions into the process of optimizing the customer portfolio. Ul- improve customer segmentation and strategic decision-making.
timately, Machado and Karray (2022a) suggest modifying the PST Saberi, K. Hussain, and Saberi (2023) used a Customized Assort-
model introduced by Buhl and Heinrich (2008) by incorporating ML ment by Customer Churn (CA&C) model for online grocery purchas-
techniques to classify customers into groups and forecast the RAR value ing. This model uses mixed multinomial logit, survival analysis, and
for each category. These studies illustrate the importance of utilizing dynamic programming to manage inventory by targeting high-value
financial portfolio theory in the management of customer portfolios. consumers at risk of churning, enhancing customer retention and prof-
An alternative method for assessing the value of customers, taking itability (Saberi et al., 2023). Ben Ncir, Ben Mzoughia, Qaffas, and
into account the level of risk, is the Multiple Source of Risk (MSR) ap- Bouaguel (2023) created a proprietary genetic algorithm for multi-
proach. This technique considers the various risk variables related with objective segmentation in banking to better classify clients based on
the industry and business environment (Machado & Karray, 2022a). predictive and descriptive behaviors. This method improves client seg-
The authors Singh et al. (2013) have noted that many studies that mentation and marketing strategy customization (Ben Ncir et al., 2023).
consider risk in customers’ value calculations, such as those based on Using machine learning to forecast Customer Lifetime Value (CLV)
the CAPM in finance, only consider one type of risk – specifically, shows the industry’s shift toward more advanced analytical frame-
income volatility – without identifying and quantifying various sources works. Elveny, Syah, and Nasution (2024) used resilient M-estimation
of risk that impact customers’ value. As a result, they suggested a to optimize CLV predictions under data variability and outliers, demon-
methodology for evaluating the worth of credit card clients by consider- strating the potential of sophisticated statistical techniques for com-
ing several risk factors, such as the instability of diverse income sources plicated customer data. Studies in these lines show that customer
and the customers’ probability of default. In a subsequent study, Singh valuation is becoming more granular and risk-adjusted. Advanced clus-
et al. (2016) utilized a non-parametric method to create a risk-adjusted tering algorithms, genetic algorithms, and machine learning are being
Recency Frequency and Monetary (RARFM) index for individual cus- used to adapt to client behavior and economic instability to drive
tomers. This index considered returns from different income sources customer relationship profitability and sustainability. This move shows
and accounted for various types of risk, including the potential for the sophistication of data analysis tools and the relevance of adaptable
customer churn, the likelihood of achieving a minimum sales threshold, strategy in fast-changing markets.
and the volatility of customer purchases. Additional research, such as
the study conducted by Ryals and Knox (2005), has examined variables 4.2. Customer sources of risk
such as consumers’ individual insurance claims and customer attrition.
Ultimately, Machado and Karray (2022b) integrate the PST and MSR The SLR study identified nine distinct sources of risk that customers
methodologies to compute the RAR in FSI. Subsequently, they employ may bear, with the most frequently utilized risks being the fluctuation
statistical tests to assess the outcome. in customers’ income, beta risks, and customers’ churn. The variability
Several studies, including those conducted by Dhar and Glazer of clients’ income is a quantifiable risk that exists in all industry
(2003) and Ryals (2003), have utilized the CAPM to determine a contexts, which accounts for its widespread presence in the literature.
customer-specific discount rate for integration into CLV models. Never- As an illustration, Wangenheim and Lentz (2005) included the unpre-
theless, Wangenheim and Lentz (2005) and Buhl and Heinrich (2008) dictability of customers’ transactions as a type of risk that is not related
argue that the CAPM has specific limitations, including its failure to to the overall market in their calculations. This was done to address the
consider unsystematic risk in the computation. In order to address shortcomings of the CAPM model, which does not consider such non-
these constraints, the researchers utilize the Portfolio Selection Theory systematic risks. In addition, they emphasized that when determining
(PST) proposed by Sharpe (1964) for client portfolio management. They the value of an asset, the anticipated value is reduced by a factor that is
argue that applying Sharpe (1964)’s PST model offers a more efficient impacted by the level of risk connected to the asset, such as its historical
method to alleviate these limits. The authors Homburg et al. (2009) or anticipated future volatility. Hence, the initial metric to be examined

7
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

Table 4
SLR - Data extraction and summary of prior studies.
Study Research theme Industry Risk Method Time frame Total customers
(years)
Dhar and Glazer (2003) Find the best customers’ segment by B2B Beta risk Mean Variance 5 –
calculating Risk Adjusted Lifetime Value
(RALTV)
Ryals (2003) Introducing risk in CLV calculation FSI (insurance) Churn Mathematical 4 2
and Statistical
models
Ryals and Knox (2005) Introducing risk into CLV calculations FSI (insurance) (1) Insurance claim Mathematical 4 12
within the insurance industry involves risk, (2) Churn and Statistical
forecasting revenue, risk, and costs, and models
then calculating the risk-adjusted CLV
Wangenheim and Lentz (2005) Attempt to segment the customers based B2C (Airlines) (1) Beta risk, (2) Regressions, 4 23 652
on metrics that describe the dynamics of Variance of Correlation
their lifecycle development, including customers analysis, and
return and risk characteristics transactions, (3) Clustering
Inactive periods (k-Means)
Hai-wei et al. (2006) Incorporating risk into the measurement B2B (1) Churn, (2) NBD, and 2 2343
of customers’ value through a Bivariate Decline of purchase, Mathematical
Hierarchical Bayesian approach (3) Volatility of and Statistical
income models
Ryals and Knox (2007) Adjusting the CLV calculation by FSI (insurance) Relationship risk Mathematical 4 10
incorporating relationship risk. (customer and Statistical
Developing a customer relationship relationship, models
scorecard as a managerial tool to assess account
risks in customer relationships and relationship, and
formulate risk mitigation strategies knowledge of
customer)
Buhl and Heinrich (2008) Balancing the customer portfolio across FSI (insurance) Beta risk Mean Variance 10 –
various customers’ segments based on
their occupations, estimating the return,
risk, and correlation coefficients of
different groups according to their
annual revenue. This balancing is
achieved through statistical analysis
(mean–variance) and consideration of
correlations between segments
Homburg et al. (2009) Developing a CLV model for dynamic FSI (B2C: Volatility of income Regression and 1.25 (Telco), 300000 (Telco),
customer portfolio analysis that Telecommunication, (probability of Mean-Variance 4 (others) 3422 (Pharm), 826
considers switching patterns between Pharmaceutical, changing segment (Chem), and
customers’ segments over time chemical and over time) 100000 (Bank)
Banking)
Sackmann et al. (2010) Finding the optimal mix of customers’ B2C (commerce) Volatility of income Mean-Variance 0.75 2357
segments (relationship-oriented and
transaction-oriented) within a
non-contractual customer portfolio using
Mean-Variance (MV) optimization
Tarasi et al. (2011) Segment the company based on cash B2B Beta risk Mean-Variance 7 516
flow and utilize Mean-Variance analysis
to determine the most profitable
portfolio considering the risk-return ratio
Ruch and Sackmann (2012) Calculate the risk-adjusted price for B2C (Commerce) (1) Payment risk (2) NBD and – –
customers in an e-tailer. The price is Churn Mathematical
adjusted based on customers’ value, and Statistical
turnover risk, and payment risk models
Singh et al. (2013) Calculating Risk Adjusted Revenue in FSI (Credit card) (1) Beta risk, (2) Data 3 1700
the credit card industry using DEA, and Volatility of income Envelopment
then ranking customers from the most to (3) Probability of Analysis (DEA)
the least profitable Default
Juhl and Christensen (2013) Building upon the work of Tarasi et al. B2B Beta risk Mean-Variance 4 968
(2011), this study aims to enhance it by
incorporating growth rate considerations.
It involves segmenting customers based
on geographical and customers’
segments, utilizing an expanded dataset,
and introducing additional metrics such
as reward-to-volatility (RtVo) and
Jensen’s alpha
(continued on next page)

8
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

Table 4 (continued).
Study Research theme Industry Risk Method Time frame Total customers
(years)
Albadvi and Norouzi (2013) Calculating risk-adjusted CLV and FSI (1) Beta risk (2) NBD and 2.75 2632
comparing it with the traditional CLV. Downside beta Mathematical
The process involves clustering and Statistical
customers based on RFM variables, models
estimating beta and downside beta for
each segment, and subsequently
calculating a risk-adjusted discount rate
for each segment. The CLV is computed
using the Pareto/NBD model
Yun and Yan (2013) Developing a customer portfolio B2B (Commodities) Volatility of income Mathematical 1 20
optimization model under risk and Statistical
conditions, utilizing multi-phase models
marketing resource inputs as decision
variables. This model is solved using a
particle swarm optimization (PSO)
method, aiming to assist companies in
optimizing customer equity and
enhancing their marketing strategies
So et al. (2014) Employing credit card users as either FSI Probability of Scorecard 4 6000
transactors or revolvers, this study default models
demonstrates the process of constructing
a scorecard to predict the likelihood of
each applicant being a transactor. The
impact of this distinction on profitability
modeling is then showcased
Petersen and Kumar (2015) Conduct research and experimentation to B2C (Commerce) Risk of product Mathematical 3 26 000
develop a novel CLV metric that return and Statistical
incorporates product returns as a risk models
factor in the calculation. Compare the
results with the company’s existing
strategy and a benchmark model. Utilize
the Pareto/Non binomial distribution
(NBD) model to statistically estimate the
volatility of future returns generated by
customers. Calculate the weight of each
customers’ segment and optimize the
composition of customers
Norouzi and Albadvi (2016) Utilizing the Pareto/Non binomial B2C (Cosmetics) Volatility of income NBD and 2 1411
distribution (NBD) model as a statistical Mean-Variance
assumption to estimate the volatility of
future returns generated by customers,
calculate the weight of each customers’
segment, and optimize the composition
of customers
Singh et al. (2016) Develop a risk-adjusted RFM method for B2B (1) Churn, (2) DEA 2 2000
calculating CLV and employ DEA to Volatility of income
determine the most profitable customer
composition
Viviani et al. (2023) Enhance the profitability-risk B2C (Hotel) Volatility of income Mathematical 5 5000
relationship within the hotel industry by and Statistical
optimizing customers’ segments. models
Employing Hidden Markov Model
(HMM) for customers’ segmentation, the
customer portfolio is subsequently
optimized using a statistical approach
Machado and Karray (2022a) Utilize ML to model RAR in the FSI (loan settings) (1) Recency of ML 8 2 million
financial sector, integrating Customer delinquency, (2)
Portfolio Theory (CPT) and Multiple Volatility of
Sources of Risk (MSR) approaches multiple sources of
income, (3) Credit
rating, and (4)
Probability of
default
Machado and Karray (2022b) Implementing a hybrid approach to FSI (loan settings) (1) Volatility of Mathematical 8 2 million
evaluate Customer RAR in the financial sources of income, and Statistical
sector. Calculate RAR using the CPT (2) Beta risk, (3) and ML
approach and group customers based on Minimum expected
all variables in the dataset rate of return

in this particular scenario is the fluctuation in client transactions over Where 𝑝 represents the number of periods, 𝑅𝑉𝑖𝑡 is greater than
a period of time, also known as the variability of income derived from zero, and 𝑐 ranges from 1 to 𝑗. 𝑅𝑉𝑖𝑡 denotes the revenue generated
the customer, which can be described as: by customer 𝑖 during period 𝑡, while 𝐶𝑅𝑉 represents the revenue
1 ∑
𝑝
conditioned on the customer. The study conducted by Homburg et al.
𝐶𝑉 𝐴𝑅𝑗𝑖 = (𝑅𝑉𝑖𝑡 − 𝐶𝑅𝑉𝑖𝑗 )2 (1)
𝑝 − 1 𝑡=1 (2009) examines the utilization of income volatility as a measure of risk

9
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

among customers. They took into account the unpredictability of client According to Singh et al. (2016), customer churn was recognized as
income over a specific timeframe and the probability of a customer the third most common risk in the assessment. Customer churn is a risk
switching their category as risks to be included in the analysis. Singh that can occur in any business and is currently seen as a significant con-
et al. (2013) suggested incorporating the volatility of various revenue cern for many firms. Customer churn is the likelihood that a customer
sources as one of the several risks. The term "volatility of income" was will cease to be engaged or loyal to a firm. Singh et al. (2016) warn
defined as: that using a client’s previous purchases as the only basis for predicting
√ their future behavior might be deceptive. This approach may wrongly
𝜎 = 𝐸(𝑋 2 ) − (𝐸(𝑋))2 (2)
presume that the consumer is still actively involved with the firm,
where 𝐸(𝑋) represents the anticipated mean value of 𝑋. Further- while in fact they may have already stopped using its services. This
more, Machado and Karray (2022a) employ income volatility as a risk might result in companies squandering time and funds on customers
factor to forecast the probability of default (PD) for a client in the who have lost interest in their products or services. Hai-wei et al.
financial services industry (Equation 2.2). (2006) observe that conventional models such as recency, frequency,
Beta risk was shown to be the second most frequently encountered monetary (RFM) typically rely just on the recency of a customer’s
risk factor in the SLR analysis. The majority of the studies reviewed most recent purchase to assess the likelihood of churn. However, this
were based on financial portfolio theory, namely the CAPM and MPT. approach may not adequately capture the actual risk of churn since it
It is important to note that CAPM only considers systematic risk, which disregards fluctuations in customer purchase patterns. In order to tackle
is measured by beta. This information is supported by the studies con- this issue, they provide a more extensive characterization of churn risk
ducted by Buhl and Heinrich (2008), Wangenheim and Lentz (2005). that considers both the duration since a customer’s previous purchase
Beta is a quantitative measure that indicates the relative level of price and their whole purchase record. Consequently, they define churn risk
volatility of an asset in relation to a benchmark. More precisely, it refers as:
to the proportion of the covariance between an asset and the market to 𝑅𝑒𝑐𝑒𝑛𝑐𝑦
Churn Risk = √ (6)
the variance of the market. When an asset has a positive beta value, it 𝜎𝑛
means that the stock price is expected to move in the same direction as
the overall market. On the other hand, if the beta value is negative, the The symbol 𝜎𝑛 denotes the variance of interpurchase time. In their
stock price is projected to move in the opposite direction of the market. research, Ryals (2003), Ryals and Knox (2005) consider the probability
A stock with a beta of 1.0 will have returns that closely follow the of client retention as the measure of risk. The customer retention
movements of the market, whereas a stock with a beta of 1.5 is expected rate is estimated based on the managerial knowledge of the customer
to exhibit movements 1.5 times greater than the market (Wangenheim relationship, and this risk is then used to generate the risk-adjusted
& Lentz, 2005). Therefore, during periods of increasing stock markets, CLV.
high beta shares are favored, and during periods of declining stock Various studies have identified distinct categories of risks that are
markets, low beta shares are favored (Ryals, 2003). Subsequently, this unique to individual businesses. In the financial services business,
concept is expanded to the client portfolio, where the beta value is such as credit card or loan providers, PD (Probability of Default) was
calculated for each individual customer or segment of customers, and utilized as a risk assessment measure, as demonstrated by Machado
employed as an indicator of their risk. and Karray (2022a), Singh et al. (2013), So et al. (2014). PD is an
In order to determine the beta value of client segments, it is essential acronym that represents the likelihood of a consumer failing to make
to establish the market portfolio. This portfolio encompasses all assets their credit card or loan payments. Every financial institution has its
that are currently available, with each asset being held in proportion own criteria for classifying a customer as being in default on payments,
to its market value in relation to the overall market value of all with the majority of banks defining default as the failure to pay the
assets. However, it is challenging or frequently impractical to establish minimum balance for a consecutive period of 90 days (Singh et al.,
a singular market portfolio that encompasses all companies (Buhl & 2013). Additional sources of risk were detected in many industries. For
Heinrich, 2008; Tarasi et al., 2011). Thus, the majority of research example, in the insurance industry, kinds of risk such as customer claim
utilize the concept of the market portfolio to compute beta values, risk and relationship risk were observed Ryals and Knox (2005, 2007).
which is defined as the company’s existing customer base (Albadvi & Significant risks in e-commerce include product returns and payment
Norouzi, 2013; Buhl & Heinrich, 2008; Dhar & Glazer, 2003; Machado risk, as outlined by Petersen and Kumar (2015), Ruch and Sackmann
& Karray, 2022b; Singh et al., 2013; Tarasi et al., 2011; Wangenheim (2012).
& Lentz, 2005). Each of these research employs a same formula to
compute the beta value: 4.3. Time horizon and input data in risk-adjusted CLV models
𝑐𝑜𝑣(𝜙𝑐𝑡 , 𝜙𝑚𝑡 )
𝛽= (3) The time of observation or data utilized to compute CLV shown
𝑣𝑎𝑟(𝜙𝑚𝑡 )
substantial variation across the chosen research. The study conducted
The variables 𝜙𝑐𝑡 and 𝜙𝑚𝑡 represent the returns for customer 𝑐 and by Sackmann et al. (2010) had the briefest duration of observation,
market 𝑚, respectively, for a specific time period 𝑡. The calculation employing a mere 39 weeks of questionnaires to examine customer
of the discount rate may vary among studies, as demonstrated by the behavior and categorize clients into segments. In contrast, the study
works of Dhar and Glazer (2003) and Wangenheim and Lentz (2005). conducted by Buhl and Heinrich (2008) had the most extensive data
They define the discount rate as: collection period, spanning 10 years of customer income data, in order
𝑑 = 𝜓𝑚 × 𝛽 (4) to forecast the CLV. The length of the observation data was associated
with the specific industry in which the study was carried out. As an
Where 𝜓𝑚 is the anticipated market rate of return. In contrast, Buhl illustration, Homburg et al. (2009) employed distinct time intervals for
and Heinrich (2008) enhance the discount rate calculation formula each sector in their investigation. In the telecommunications sector, a
by incorporating the ‘‘risk-free asset’’ or the customer with the least period of observation is defined as a duration of one quarter of a year,
amount of risk into the equation, as follows: while in the banking industry, it is defined as a duration of one year.
Typically, the B2C sector had a shorter duration for data observation,
𝑑 = 𝜓𝑓 + 𝛽(𝜓𝑚 − 𝜓𝑓 ) (5)
whereas the FSI sector had the greatest duration. The mean duration
𝜓𝑓 is the minimal expected rate of return for customers with the lowest for B2C, B2B, and FSI was determined to be 2.85, 3.5, and 4.9 years,
risk, while 𝜓𝑚 reflects the expected rate of return of the market. respectively.

10
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

The observed number of customers in the studies shown variation In their study, Tarasi et al. (2011) examined the variation in a
based on the industry and methodology employed. The minimum num- client portfolio, forecasted the similarity between several market seg-
ber of clients seen was 10 in the study conducted by Ryals and Knox ments such as automobile, consumer goods, food and beverage, and
(2007) on an insurance firm, but Machado and Karray (2022a, 2022b) investigated the application of market segment weights in an opti-
utilized the largest dataset comprising 2 million instances spanning mized portfolio. Subsequently, mean–variance analysis was employed
8 years. The authors Ryals and Knox (2007) utilized a limited set of to construct the optimal frontier for the composition of the client
customer data and applied managerial expertise to assess the customer’s portfolio. Similarly, Viviani et al. (2023) employed the Hidden Markov
risk worth. They subsequently computed the risk-adjusted CLV using Model (HMM) to divide consumers into segments. They then utilized
mathematical methodologies. Meanwhile, Machado and Karray (2022a, mean–variance analysis to create an optimal customer portfolio, taking
2022b) employed ML algorithms to forecast the risk of customers, into account the customers’ likelihood of switching segments as a
necessitating a larger dataset for improved accuracy. Table 4 contains measure of risk. Aside from mean–variance analysis, other techniques
comprehensive data regarding the duration of the time period and the like particle swarm optimization are employed to ascertain the most
number of clients involved in each study. advantageous customer portfolio (Yun & Yan, 2013).
Other studies in various application areas utilize distinct approaches
4.4. Application areas & methods to incorporate the customers’ risk in CLV to integrate client risk. As an illustration, Ryals and Knox (2005)
models utilizes a mathematical method to directly compute risk-adjusted CLV.
The study conducted by So et al. (2014) utilizes a statistical approach
This study explores the incorporation of customers’ risk into the to construct a scorecard predictive model, which is then employed
customers’ value inside the industrial setting. The research identified in to identify customers who are more likely to generate higher profits.
the SLR can be categorized into three major groups based on industry The authors in Albadvi and Norouzi (2013) employ the RFM model
settings: FSI (Financial Services Industry), B2C (Business-to-Consumer), to cluster consumers and subsequently utilize a mathematical method-
and B2B (Business-to-Business). The SLR findings indicate that the bulk ology to compute the risk of these customers. The risk-adjusted CLV
of studies focused on Financial Services Industry (FSI), accounting for associated with each segment is estimated using Pareto/NBD model-
44% of the research. This encompassed several areas such as insurance ing. The authors Singh et al. (2013) and Singh et al. (2016) utilized
(e.g., Ryals and Knox (2005)), credit card (e.g., Singh et al. (2013)), and Data Envelopment Analysis (DEA) to calculate the RAR in a credit
peer-to-peer lending (Machado & Karray, 2022b). The second largest card company and the CLV from a RAR Forecasting Model (RARFM),
group was found in business-to-consumer (B2C) settings (30%) outside respectively.
of the financial services industry (FSI), encompassing several businesses
such as e-commerce (e.g., Ruch and Sackmann (2009)), telecommuni- 4.5. Machine learning in risk-adjusted CLV studies
cations (Homburg et al., 2009), and airlines (Wangenheim & Lentz,
2005). Meanwhile, the remaining 26% of the research focused on Out of the studies examined, only four of them employ ML methods
business-to-business (B2B) settings, namely in the context of providing to forecast the risk-adjusted CLV. The initial study that employs a ML
medical instruments to hospitals (Hai-wei et al., 2006) and in the algorithm in their research is presented by Wangenheim and Lentz
commodities industry (Yun & Yan, 2013). (2005). Multiple linear regression is employed to forecast customer in-
In terms of application area, the studies conducted on RAR pri- come, taking into account the customers’ risk. The anticipated value is
marily aimed at finding the optimal customer portfolio composition to then utilized to construct customer segmentation using the K-Means al-
maximize returns. Analyzing the most frequent words, we noticed that gorithm. The authors of the study (Homburg et al., 2009) employed ML
‘‘customer portfolio management’’ is one of the dominant keywords in methods, namely a regression tree, to divide their customer data into
the area of research. This application area of the studies are seen in segments. These segments were then applied to optimize the client port-
various industries such as the financial services industry (Albadvi & folio using the Mean-Variance approach. In a recent study, Machado
Norouzi, 2013; Buhl & Heinrich, 2008; Dhar & Glazer, 2003; Homburg and Karray (2022b) included ML techniques into their model to esti-
et al., 2009), the business-to-consumer (B2C) setting (Norouzi & Al- mate customers’ RAR in the financial lending industry. They utilized
badvi, 2016; Sackmann et al., 2010; Viviani et al., 2023; Wangenheim logistic regression to obtain the probability of default (PD) of lenders,
& Lentz, 2005), and the business-to-business (B2B) setting (Juhl & which was then incorporated as one of the factors in assessing the
Christensen, 2013; Tarasi et al., 2011; Yun & Yan, 2013). In addition customer’s risk. The credit rating was utilized to forecast the probability
to these industries, the studies also explored other areas of application, of default (PD) of the lender. Two distinct approaches were imple-
such as identifying the most profitable customers in the credit card mented to represent the client credit rating: the Weight of Evidence
industry (Singh et al., 2013; So et al., 2014), adjusting product prices (WOE) method and the normalized interest rate. The outcomes of both
in e-commerce (Ruch & Sackmann, 2012), and also find more accu- methods were then compared. The previous approaches demonstrate
rate risk-adjusted CLV metric (Machado & Karray, 2022b; Petersen & an innovative effort to incorporate several risk factors into CLV calcu-
Kumar, 2015; Ryals, 2003; Singh et al., 2016). lations, emphasizing the advantages of merging conventional statistical
The inclusion of customers’ risk in customer assessment relies on methods with ML techniques.
the particular field of study being examined. Several studies employ Furthermore, the subsequent investigation conducted by Machado
financial portfolio ideas to optimize consumer portfolios, often utilizing and Karray (2022a) represents the initial study that comprehensively
mean–variance analysis to identify the ideal composition. In order employed ML techniques to forecast the risk-adjusted CLV. The authors
to determine the optimal customer portfolio composition, these stud- suggest utilizing a hybrid ML system to accurately forecast the RAR
ies commonly divide customers into segments based on demographic value of consumers’ sector. The term ‘‘hybrid framework’’ in this con-
characteristics such as their occupation, educational attainment, home- text denotes the combination of two distinct ML algorithms, typically
ownership, marital status, and employment situation (Buhl & Heinrich, using both supervised and unsupervised learning, with the aim of
2008; Homburg et al., 2009; Machado & Karray, 2022b; Sackmann predicting a specific value. Hybrid models are advantageous in terms
et al., 2010; Tarasi et al., 2011; Viviani et al., 2023). This segmentation of predicted accuracy due to their ability to efficiently handle datasets
is then utilized to calculate or forecast the most suitable customer with a large number of dimensions. They are able to integrate the
portfolio composition. Segmenting consumers is essential as the actions strengths of individual models while minimizing their flaws. The study
of individual customers can be unpredictable, however the behavior conducted by Machado and Karray (2022a) involved a comparison
of a sizable group of customers can be more foreseeable (Yun & Yan, between individual models and hybrid models using consumer data
2013). from the peer-to-peer lending business. The hybrid approach initially

11
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

employs unsupervised learning to cluster the customer data, subse- 4.7. Research gaps and research agenda
quently utilizing the outcomes as a variable to forecast the customer
RAR value through supervised learning. The researchers conducted a Although there is an increasing amount of literature on customer
comparative analysis of various combinations of clustering and pre- valuation, there are still various areas of research that require attention.
diction algorithms, including k-Means++, k-Means random, DBSCAN, Most of the research on customer valuation that consider customers’
clustering with predetermined variables for clustering algorithms, and risk have mostly focused on the financial services industry (FSI) and
Adaptive Boosting (AB), Gradient Boosting (GB), Decision Tree (DT), have not adequately included other industries. Specifically, there is
Random Forest (RF), Support Vector Machine (SVM), and Artificial a dearth of research in the business-to-consumer (B2C) sector, which
Neural Network (ANN) for the prediction algorithms. The trials un- possesses distinct characteristics and confronts unique issues in contrast
dertaken in their research demonstrate that the hybrid ML algorithms to the financial services industry (FSI). Moreover, there exists only a
surpass individual models in terms of predictive capacity and pro-
single study undertaken in the communications business that examines
cessing time across most frameworks. The efficiency of the combined
the incorporation of customers’ risk in customer assessment. The signif-
models of DT, RF, or GB with k-Means++ is higher than that of the
icance of this study gap lies in the distinct consumer behavior observed
separate models. Thus, the study conducted by Machado and Karray
in the communications business, typified by substantial client acquisi-
(2022b) showcases the capabilities of hybrid ML models in predicting
risk-adjusted CLV and offers valuable insights for future research in this tion costs and minimal switching costs. Utilizing customer valuation
field. techniques in the telecommunication sector, particularly in forecast-
ing client lifetime value, can offer useful information for optimizing
4.6. Main contributions & limitations of this study customer acquisition and retention efforts.
Furthermore, although ML has seen a growing application in cus-
4.6.1. Academic and practical contributions tomer valuation studies, there is a dearth of research that fully har-
This paper examines the current status of academic research on nesses the potential of ML algorithms. Table 2 indicates that the ma-
risk-adjusted customer value, highlighting its significance for both jority of studies primarily concentrate on quantifying the influence
theoretical and practical aspects. From an academic perspective, this of customers’ risk on CLV, whereas just a minority utilize predictive
study is, to our knowledge, the first SLR that investigates the concept techniques to anticipate the impact of customers’ risk. Furthermore,
of customer risk in customer valuation. The systematic review based on Table 4 indicates that despite the increasing popularity of ML (ML),
existing research papers enhances the existing body of knowledge by only two research effectively employ this method to forecast the value
offering a theoretical analysis of customer risk in customer assessment. of consumers after adjusting for risk. This offers academics a chance to
Moreover, this study offers a comprehensive organization and analysis investigate the potential of ML in forecasting the value of consumers
of previous research on risk-adjusted CLV, encompassing various indus- adjusted for risk. Given the abundance of big data and the substantial
try contexts, methodologies, hazards, and techniques employed. These computational power now accessible, ML possesses the capacity to pro-
mappings can be utilized to highlight areas of research that need further vide robust forecasts pertaining to customer behavior and risk. Hence,
investigation and provide guidance for future research growth.
conducting a comparative examination between ML and conventional
This study also yields relevant insights for practitioners. To begin
approaches for computing the worth of consumers adjusted for risk will
with, the derived application area provides a clear understanding of
yield useful information regarding their efficacy and possible utilization
the current state of development in the sector. Organizations might
in diverse businesses. Additional investigation in this field has the
utilize this data to remain current with the most recent advancements
and techniques in risk-adjusted customer value. Furthermore, the study potential to facilitate the development of more exact and reliable
identifies many application areas and procedures that can assist firms in CLV estimations, ultimately assisting firms in making well-informed
selecting the most suitable research topic and approach for incorporat- decisions regarding customer acquisition and retention tactics.
ing customer risk into customer assessment for a certain objective. For Furthermore, there is a notable deficiency in research regarding the
instance, if a company operates in a volatile industry with a significant implementation of XAI in customer value studies, as only one study
likelihood of customers leaving, they may opt to include a larger dis- has been identified in this particular area. Furthermore, no research
count rate in their CLV calculations to factor in this risk. Furthermore, has been discovered that has utilized XAI for the purpose of predicting
all of the identified studies have yielded empirical findings that might risk-adjusted income. The lack of research in XAI (Explainable Artificial
enhance firms’ comprehension of the practical implementation of the Intelligence) on customer valuation is of utmost importance, given the
sector. The study’s discovery highlights the significance of incorporat- increasing significance of interpretability, regulatory compliance, and
ing customer risk into the assessment of customer value. Consequently, ethical considerations in the explainability of ML models. Moreover,
it is strongly advised for organizations to appraise and analyze their the absence of Explainable Artificial Intelligence (XAI) applications in
customer risk in order to have a more comprehensive comprehension the field of RAR prediction is a significant deficiency. This is because
of customer value. RAR forecast is a crucial measure for assessing customer value, as it
incorporates the risk associated with customers. Integrating Explainable
4.6.2. Limitations of this study Artificial Intelligence (XAI) into RAR forecast can improve the clarity
This study faced two primary limitations. Firstly, the scope of of ML models and yield more precise and dependable evaluations of
this systematic literature review (SLR) was restricted by the specific
client value.
selection of keywords and research databases used. Despite careful
To summarize, the existing body of research on customer valua-
evaluation of these search parameters to ensure a comprehensive re-
tion has identified numerous areas that require more investigation.
view of the topic, it is possible that some relevant studies were not
These include the absence of representation from diverse industries,
included. This limitation suggests that future research might expand the
search criteria to include a broader array of databases and keywords, the restricted application of ML techniques, and the scarcity of studies
potentially uncovering additional pertinent research. Secondly, the that employ explainable artificial intelligence (XAI). To maximize the
inclusion and exclusion criteria of the review were narrowly defined, utility of machine learning (ML) in predicting risk-adjusted Customer
focusing only on specific types of articles. As a result, general articles Lifetime Value (CLV) in non-contractual settings, future research should
that address the area of interest but do not specifically concentrate concentrate on several key aspects. Primarily, there is a need to develop
on it were excluded, possibly causing the omission of crucial insights. and validate ML models tailored to manage the significant volatility
Additionally, the discovery of extensively referenced publications that and unpredictability of customer behavior in these sectors (e.g., apply-
do not cite or mention the research included in this SLR indicates the ing models such as deep learning and Bayesian inference might shed
existence of other bodies of work that integrate customer risk into CLV light on improved results). Future studies should focus on constructing
estimations but were not captured in this review. robust predictive algorithms capable of accurately forecasting changes

12
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

in customer loyalty and spending patterns, even in the absence of CRediT authorship contribution statement
long-term contractual relationships. Comparative studies should also be
undertaken to evaluate the performance of these ML models against Edo Belva Firmansyah: Writing – original draft, Visualization,
traditional statistical methods in predicting risk-adjusted CLV. Such Validation, Software, Investigation, Formal analysis, Data curation,
studies would provide valuable insights into when ML models are more Conceptualization. Marcos R. Machado: Writing – review & editing,
effective than traditional approaches, thereby guiding businesses on the Supervision, Project administration, Funding acquisition, Conceptu-
strategic implementation of these advanced analytical tools. Moreover, alization. João Luiz Rebelo Moreira: Writing – review & editing,
incorporating multi-dimensional risk factors, such as customer churn, Supervision, Project administration.
probability of default, and variability in transactions, into ML models
will enhance their predictive accuracy and reliability. This integration Declaration of competing interest
will better equip businesses to make informed decisions about customer
acquisition and retention strategies. The authors declare that they have no known competing finan-
cial interests or personal relationships that could have appeared to
5. Conclusion influence the work reported in this paper.

This study demonstrates that when including client risk into CLV Acknowledgment
calculations, the commonly used approach entails applying financial
(customer) portfolio theory and considering different sources of risk We gratefully acknowledge the support of the Marie Skłodowska-
evaluations. Moreover, the primary objective of studies investigating Curie Actions (MSCA) under the European Union’s Horizon Europe
a customer portfolio theory is to employ the mean–variance technique research and innovation program for the Industrial Doctoral Network
and customer revenue volatility as the primary risk element in order on Digital Finance, acronym: DIGITAL, Project No. 101119635. Their
to ascertain the most advantageous composition of client portfolios. significant contribution has been instrumental in advancing our re-
Conversely, studies that are based on a multi-source risk approach search and fostering collaboration within the digital finance field across
concentrate on different categories of risk, such as customer churn Europe. Next to that, we also would like to acknowledge that this
and probability of default. Despite the widespread application of these research has received funding from the European Union’s Horizon
studies in several industries, such as FSI, B2C, and B2B, there is still Research and Innovation Actions programme under grant agreement
room for research in this topic. Additional studies can be conducted number 101138473, related to the project ‘‘IS2H4C - From Industrial
to conduct a more thorough investigation in order to explore this field Symbiosis to Hub for Circularity’’.
more deeply. This systematic literature review suggests that further in- These partnerships and funding sources have greatly contributed to
vestigations could explore other sectors, particularly in non-compulsory our ability to conduct rigorous and impactful research. Our findings are
settings, such as telecommunications and commerce, to expand the our own and do not necessarily represent the views of the supporting
scope of these research. institutions.
The majority of the identified studies primarily examine the quan-
tification of the influence of customer risk on CLV, whereas only a References
minority utilize predictive methods to anticipate the effects of customer
Aksnes, D. W., Langfeldt, L., & Wouters, P. (2019). Citations, citation indicators, and
risk. In addition, despite the increasing popularity of ML, only two
research quality: An overview of basic concepts and theories. Sage Open, 9(1),
research effectively employ this method to forecast risk-adjusted cus- Article 2158244019829575.
tomer value. This offers academics a chance to investigate the potential Albadvi, A., & Norouzi, A. (2013). Using downside CAPM theory to improve customer
of ML in forecasting risk-adjusted customer value. ML has the capacity lifetime value prediction in non-contractual setting. Management Science Letters,
to provide robust predictions about customer behavior and risk, thanks 3(12), 3003–3012.
Amato, A., Osterrieder, J. R., & Machado, M. R. (2024). How can artificial intelligence
to data accessibility and advanced computing capabilities. Hence, con- help customer intelligence for credit portfolio management? A systematic literature
ducting a comparative examination of ML and conventional approaches review. International Journal of Information Management Data Insights, 4(2), Article
for computing risk-adjusted customer value will yield useful infor- 100234.
mation regarding their efficacy and prospective utilization in diverse Annarelli, A., Battistella, C., Nonino, F., Parida, V., & Pessot, E. (2021). Literature
review on digitalization capabilities: Co-citation analysis of antecedents, conceptu-
businesses. Additional investigation in this field has the potential to
alization and consequences. Technological Forecasting and Social Change, 166, Article
facilitate the development of more exact and reliable CLV estimations, 120635.
ultimately assisting firms in making well-informed decisions regarding Baas, J., Schotten, M., Plume, A., Côté, G., & Karimi, R. (2020). Scopus as a curated,
customer acquisition and retention tactics. high-quality bibliometric data source for academic research in quantitative science
studies. Quantitative Science Studies, 1(1), 377–386.
To enhance the implementation of ML in predicting risk-adjusted
Behera, R. K., Bala, P. K., & Dhir, A. (2019). The emerging role of cognitive computing
CLV, particularly in non-contractual industries where income volatility in healthcare: A systematic literature review. International Journal of Medical
and churn are challenging to measure, a research question for future Informatics, 129, 154–166. http://dx.doi.org/10.1016/j.ijmedinf.2019.04.024, URL
investigation could be formulated as follows: What are the most effec- https://www.sciencedirect.com/science/article/pii/S1386505619301911.
Ben Ncir, C.-E., Ben Mzoughia, M., Qaffas, A., & Bouaguel, W. (2023). Evolutionary
tive ways to use ML techniques to forecast the risk-adjusted CLV in the
multi-objective customer segmentation approach based on descriptive and
non-contractual domain-independent (any industries) scenario? predictive behaviour of customers: application to the banking sector. Journal
To summarize, this review has highlighted the primary topics and of Experimental and Theoretical Artificial Intelligence, 35(8), 1201–1223.
relationships found in the literature surrounding the integration of http://dx.doi.org/10.1080/0952813X.2022.2078886, URL https://www.scopus.
com/inward/record.uri?eid=2-s2.0-85131357000&doi=10.1080%2f0952813X.
customer risk in customer valuation. The historical progression of this
2022.2078886&partnerID=40&md5=a8edc1ca5cf3ec508261afb082541b9b, Cited
discipline, techniques for integrating customer risk, and the categories by: 0.
of risk taken into account have been systematically analyzed and Buhl, H. U., & Heinrich, B. (2008). Valuing customer portfolios under risk-return-
expanded upon. We identified these main types of risks: beta risk, aspects: a model-based approach and its application in the financial services
churn, variance of customer transactions, and volatility of income, industry. Academy of Marketing Science Review, 12(5).
Dahana, W. D., Miwa, Y., & Morisada, M. (2019). Linking lifestyle to customer lifetime
besides other types such as inactive periods and risk of product return. value: An exploratory study in an online fashion retail market. Journal of Business
The study has also examined the constraints, scholarly and practical Research, 99, 319–331.
ramifications, and prospective avenues for future research. Dhar, R., & Glazer, R. (2003). Hedsins customers. Harvard Business Review, 81(5), 86–92.

13
E.B. Firmansyah et al. International Journal of Information Management Data Insights 4 (2024) 100279

Elveny, M., Syah, R. B. Y., & Nasution, M. K. M. (2024). An boosting Qi, J.-Y., Zhou, Y.-P., Chen, W.-J., & Qu, Q.-X. (2012). Are customer satisfaction
business intelligent to customer lifetime value with robust M-estimation. and customer loyalty drivers of customer lifetime value in mobile data services:
IAES International Journal of Artificial Intelligence, 13(2), 1632–1639. http: a comparative cross-country study. Information Technology and Management, 13,
//dx.doi.org/10.11591/ijai.v13.i2.pp1632-1639, URL https://www.scopus.com/ 281–296.
inward/record.uri?eid=2-s2.0-85192993390&doi=10.11591%2fijai.v13.i2.pp1632- Ruch, M., & Sackmann, S. (2009). Customer-specific transaction risk management in
1639&partnerID=40&md5=8cdeb0f5c2ecf3d4fa3d05272efb079a, Cited by: 0; All e-commerce. In SIGeBIZ track of the americas conference on information systems (pp.
Open Access, Hybrid Gold Open Access. 68–79). Springer.
Glady, N., Lemmens, A., & Croux, C. (2015). Unveiling the relationship between Ruch, M., & Sackmann, S. (2012). Integrating management of customer value and risk
the transaction timing, spending and dropout behavior of customers. International in e-commerce. Information Systems and e-Business Management, 10, 101–116.
Journal of Research in Marketing, 32(1), 78–93. Rust, R. T., Lemon, K. N., & Zeithaml, V. A. (2004). Return on marketing: Using
Gupta, S., Lehmann, D. R., & Stuart, J. A. (2004). Valuing customers. Journal of customer equity to focus marketing strategy. Journal of Marketing, 68(1), 109–127.
Marketing Research, 41(1), 7–18. Ryals, L. (2003). Making customers pay: measuring and managing customer risk and
Hai-wei, W., Ming-hui, J., & Ya-lin, W. (2006). Adding risk in measuring customer value returns. Journal of Strategic Marketing, 11(3), 165–175.
using bivariate hierarchical Bayesian approach. In 2006 International conference on Ryals, L. J., & Knox, S. (2005). Measuring risk-adjusted customer lifetime value and
management science and engineering (pp. 89–93). IEEE. its impact on relationship marketing strategies and shareholder value. European
Harzing, A.-W., & Alakangas, S. (2016). Google scholar, scopus and the web of science: Journal of Marketing, 39(5/6), 456–472.
a longitudinal and cross-disciplinary comparison. Scientometrics, 106, 787–804. Ryals, L., & Knox, S. (2007). Measuring and managing customer relationship risk in
Hogan, J. E., Lehmann, D. R., Merino, M., Srivastava, R. K., Thomas, J. S., & Verhoef, P. business markets. Industrial Marketing Management, 36(6), 823–833.
C. (2002). Linking customer assets to financial performance. Journal of Service Saberi, Z., K. Hussain, O., & Saberi, M. (2023). Data-driven personalized
Research, 5(1), 26–38. assortment optimization by considering customers’ value and their risk
Homburg, C., Steiner, V. V., & Totzek, D. (2009). Managing dynamics in a customer of churning: Case of online grocery shopping. Computers & Industrial
portfolio. Journal of Marketing, 73(5), 70–89. Engineering, 182, http://dx.doi.org/10.1016/j.cie.2023.109328, URL https:
Juhl, H. J., & Christensen, M. (2013). Portfolio optimization and performance eval- //www.scopus.com/inward/record.uri?eid=2-s2.0-85163196378&doi=10.1016%
uation: An application to a customer portfolio. Journal of Marketing Analytics, 1, 2fj.cie.2023.109328&partnerID=40&md5=2941cae250faca83a5ae14988bfcaf9b,
156–173. Cited by: 1.
Kraus, S., Breier, M., Lim, W. M., Dabić, M., Kumar, S., Kanbach, D., et al. (2022). Sackmann, S., Kundisch, D., & Ruch, M. (2010). Customer portfolio management in
Literature reviews as independent studies: guidelines for academic practice. Review e-commerce: an analytical model for optimization. Management Research Review,
of Managerial Science, 16(8), 2577–2595. 33(6), 617–634.
Kumar, V., & Reinartz, W. (2016). Creating enduring customer value. Journal of Sharpe, W. F. (1964). Capital asset prices: A theory of market equilibrium under
Marketing, 80(6), 36–68. conditions of risk. The Journal of Finance, 19(3), 425–442.
Kumar, V., Venkatesan, R., Bohling, T., & Beckmann, D. (2008). Practice prize report— Singh, S., Murthi, B., & Steffes, E. (2013). Developing a measure of risk adjusted
The power of CLV: Managing customer lifetime value at IBM. Marketing Science, revenue (RAR) in credit cards market: Implications for customer relationship
27(4), 585–599. management. European Journal of Operational Research, 224(2), 425–434.
Machado, M. R., & Karray, S. (2022a). Applying hybrid machine learning algorithms to Singh, A., Thakur, N., & Sharma, A. (2016). A review of supervised machine learning
assess customer risk-adjusted revenue in the financial industry. Electronic Commerce algorithms. In 2016 3rd International conference on computing for sustainable global
Research and Applications, 56, Article 101202. development (pp. 1310–1315). Ieee.
Machado, M. R., & Karray, S. (2022b). Integrating customer portfolio theory So, M. C., Thomas, L. C., Seow, H.-V., & Mues, C. (2014). Using a transactor/revolver
and the multiple sources of risk approaches to model risk-adjusted revenue. scorecard to make credit and pricing decisions. Decision Support Systems, 59,
IFAC-Papersonline, 55(16), 356–363. 143–151.
Markowitz, H. (1952). The utility of wealth. Journal of Political Economy, 60(2), Tarasi, C. O., Bolton, R. N., Hutt, M. D., & Walker, B. A. (2011). Balancing risk and
151–158. return in a customer portfolio. Journal of Marketing, 75(3), 1–17.
Méndez-Suárez, M., & Crespo-Tejero, N. (2021). Why do banks retain unprofitable Varsha, P., Chakraborty, A., & Kar, A. K. (2024). How to undertake an impactful
customers? A customer lifetime value real options approach. Journal of Business literature review: Understanding review approaches and guidelines for high-impact
Research, 122, 621–626. systematic literature reviews. South Asian Journal of Business and Management Cases,
Mengist, W., Soromessa, T., & Legese, G. (2020). Method for conducting systematic 13(1), 18–35.
literature review and meta-analysis for environmental science research. MethodsX, Venkatesan, R., & Kumar, V. (2004). A customer lifetime value framework for customer
7, Article 100777. selection and resource allocation strategy. Journal of Marketing, 68(4), 106–125.
Natarajan, R., Ranjith, C. P., Mohideen, M. S. K., Gururaj, H., Flammini, F., & Viviani, J.-L., Komura, A., & Suzuki, K. (2023). Integrating dynamic segmentation and
Thangarasu, N. (2024). Utilizing a machine learning algorithm to choose a signif- portfolio theories for better customer portfolio performance. Journal of Strategic
icant traffic identification system. International Journal of Information Management Marketing, 31(1), 140–153.
Data Insights, 4(1), Article 100218. Votto, A. M., Valecha, R., Najafirad, P., & Rao, H. R. (2021). Artificial intelligence in
Norouzi, A., & Albadvi, A. (2016). A hybrid model for customer portfolio analysis in tactical human resource management: A systematic literature review. International
retailing. Management Research Review, 39(6), 630–654. Journal of Information Management Data Insights, 1(2), Article 100047.
Petersen, J. A., & Kumar, V. (2015). Perceived risk, product returns, and optimal Wangenheim, F. V., & Lentz, P. (2005). Customer portfolio analysis: applying financial
resource allocation: Evidence from a field experiment. Journal of Marketing Research, risk and volatility measures to customer segmentation and risk-adjusted lifetime
52(2), 268–285. value determination. Available at SSRN 782064.
Pranckute,̇ R. (2021). Web of science (WoS) and scopus: The titans of bibliographic Yun, C., & Yan, P. (2013). A customer portfolio model based on multi-phase marketing
information in today’s academic world. Publications, 9(1), 12. strategy and particle swarm optimization. In 2013 International conference on
management science and engineering 20th annual conference proceedings (pp. 957–962).
IEEE.

14