Journal Search Engine
Search Advanced Search Adode Reader(link)
Download PDF Export Citaion korean bibliography PMC previewer
ISSN : 1598-7248 (Print)
ISSN : 2234-6473 (Online)
Industrial Engineering & Management Systems Vol.19 No.4 pp.744-757
DOI : https://doi.org/10.7232/iems.2020.19.4.744

A Stochastic Approach for Valuing Customers in Banking Industry: A Case Study

Hamid Bekamiri, Mohammad Mehraeen*, Alireza Pooya, Hossein Sharif
Management Department, Economics & Administrative Sciences Faculty, Ferdowsi University of Mashhad, Mashhad, Iran
Sharif University of Technology, Management and Economic Departement, Iran
*Corresponding Author, E-mail: m-lagzian@um.ac.ir
February 3, 2020 July 7, 2020 October 2, 2020

ABSTRACT


This study provides a stochastic dynamic programming model with a Markov chain that explicitly focuses on the customer as well as a new model for valuing the customers in the banking industry. The proposed framework calculates individual customer’s lifetime value dynamically. The study follows a stochastic dynamic programming model that is based on the Markov chain. The deduced findings are illustrated with supplementary context from an outstanding case study. The findings underline the importance of the stochastic model for calculating customer lifetime value based on customer behavior. The presented framework provides a beneficial way for future research and valuable insight for allocating promotional marketing strategies to customer groups. The presented framework provides a dynamic model for calculating the individual customer’s lifetime value. The main contribution of the study is the explicit calculation of individual customer’s lifetime value in the banking industry. Thus, this study provides a stochastic framework for customer segmentation and allocates appropriate marketing promotion strategies. Furthermore, the results of this study were supported by real customer data of one of the largest banks in the MENA region.



초록


    1. INTRODUCTION

    New marketing strategies aim to create business intelligence from customer information. This means that the purchase behavior of customers should be determined and accordingly, customers should be classified, and depending on the behavior of each section, special offerings should be provided for customers (Wang et al., 2019). Information and knowledge extracted from customer behavior data are sent to the marketing section continuously to determine a new solution for the operational section. Therefore, it can be stated that marketing section strategies are largely updated according to the changes in the behavior of customers in different sections.

    Since there is a growing requirement for marketers to forecast customers' future behavior accurately, it is necessary to create a mechanism to identify customers in the first step to develop strategies to log the interests of customers and the firm (Burelli, 2019;Bekamiri et al., 2020). The importance of this issue has been highlighted in recent decades with a change in the approaches of organizations and the position of the customer at the center of all organizational goals. In this regard, the most common model to identify customers is customer lifetime value model. Customer lifetime value (CLV or LTV), as a comprehensive key performance indicator (KPI), determines customers’ revenues during their relationship with the firm (Burelli, 2019). Generally, customer lifetime value is a value that the customer creates for the organization during his or her lifetime. In this line, many studies are conducted with different titles such as customer value, customer lifetime value, customer equity, and customer profitability (Hwang, 2016). The origin of these studies is the use of mathematical, statistical, and accounting models to identify the most important financial and behavioral dimensions of the customer in previous periods to determine strategies consistent with the present behavior of the customer to keep profit resulting from the relationship with a customer and to increase it.

    Every customer has a value cycle and the firm aims to estimate it to set its marketing plans. Customer lifetime value in one definition is estimated as part of cash flow using the average weighted cost of a customer over his/her lifetime relationship with the company (Kumar and Petersen, 2012).

    As pointed out, customer lifetime value estimation is important to determine organizational strategies and to set up various advertising campaigns. With the development of personal marketing, the necessity to identify customers has been highlighted based on customer value (Hwang, 2016). CLV calculation not only helps design marketing strategies but also optimizes financial planning. Managers can establish tailored campaigns to increase the firm's profit or decrease customers' costs with changes in customer behavior. These costs include operational cost, customer services, customer attrition (Wang et al., 2019). Direct marketing is one of the significant fields and its implementation requires the application of CLV. With customer lifetime value KPI, the company can solve marketing decision problems, such as the acquisition or the trade-off between acquisition and retention costs (Burelli, 2019;Blattberg and Deighton, 1996). Furthermore, marketers can segment customers in appropriate groups; thus, customers with the same behaviors will be grouped, and based on this segmentation, different marketing objectives promotional offers, personalized customer advertising, exclusive deals, and loyalty rewards programs are implemented (Wang et al., 2019).

    Given the complexity of calculating CLV, to date, there is no specific model for estimating it in practice for all industries. Besides, given the potential power of CLV as a KPI, building a proper model for CLV has been a challenge in different industries; accordingly, various approaches like statistical parametric analysis and machine learning have been applied (Burelli, 2019;Gupta et al., 2006;Sifa et al., 2018). The main objective of the present study is to provide a stochastic framework to estimate customer lifetime value using customer profitability.

    2. REVIEW OF LITERATURE

    As pointed out, customer value is studied with different titles such as customer period value, customer lifetime value, and customer profitability. However, no significant difference exists between the expected objectives of these studies (Hwang, 2016). According to the importance of customer segmentation based on the created customer value, considerable scientific studies are being conducted to develop statistical methods to determine how to estimate customer lifetime value. Most of the studies on the customer lifetime value are based on the net present value and the value of deals by customers is determined accordingly in these models (Hwang, 2016). Customer lifetime value is also defined as the present value of future purchases. Usually, customer lifetime value estimation is based on the expected purchases of a customer that is converted to the present time using the discount rate.

    The concept of customer lifetime value was introduced by Kutler (1974). Kutler introduced this concept as an instrument to estimate the discounted rate of future profits resulting from a customer. However, in many studies on this topic, customer lifetime value is considered as the estimation for all profits resulting from a customer from the beginning of a relationship with the company; in other studies, this index is estimated from the combination of two approaches (Chang et al., 2012). Nevertheless, it seems that the best definition for customer lifetime value is related to the method to estimate all dimensions of profit and loss from attraction to rejection (Hwang, 2016).

    Generally, the presented models to estimate customer lifetime value are classified into two general approaches including past and future models. The most important differences between these models are related to the consideration of the future and the costs incurred by customers (Hiziroglu and Sengul, 2012). In this section, five models for calculating CLV are investigated; models 1 and 2 are related to past models approach, and 3, 4, and 5 are related to future models approach:

    • 1) RFM model: This model is the most commonly used model am-ong previous models. According to the definition by Wansbeek and Bult (1995), RFM stands for recency, frequency, and monetary, respectively. Recency means the last purchase by the customer. Frequency means the number of purchases within a specific period. Monetary means the money that is spent within a specific period. In recent studies, some researchers have not considered the effect of three components equally, and using weighted RFM (WRFM), they have dedicated a specific weight consistent with their values (Khajvand et al., 2011;Peker et al., 2017; Kahreh et al., 2014; Hamdi and Zamiri, 2016).

    • 2) PCV method: Past Customer Value (PCV) is based on this assumption that the past performance of the customer indicates profitability in the future and previous results can be used as future customer value. In this method, customer value is estimated based on total profit resulting from past interactions. Therefore, the total profit considering the time value of money is transferred to the present and it is a basis for the potential value of the customer (Čermák, 2015; Jain and Singh, 2002). For example, suppose that a retailer is interested in estimating PCV for all customers to identify the best customer. This company gathers information about the value of purchased products and the profit obtained by different customers within a specific period. They could compare the value created by each customer by converting it into the present value (Esmaeili and Tarokh, 2013).

    • 3) Jain and Singh’s model: Jain and Singh (2002), to estimate customer lifetime value, assumed that all cash flows will be considered at the end of the period and according to net present value, future cash flow is predicted from the customer. They presented Eq.1 to estimate customer lifetime value; it has been taken into consideration by other researchers as a basic model to estimate customer lifetime value (Jain and Singh, 2002):

      CLV = i = 1 n R i C i ( 1 + d ) i 0.5
      (1)

    • i = period to create cash flow transaction by the customer

    • R = revenue obtained from the customer in the period i

    • C = total revenue R creation costs in the period i

    • n = total number of customer activity periods

    • 4) Hwang, Jung, and Suh’s Model: Hwang, Jung, and Suh (2004) presented a model to estimate customer lifetime value. Present value and potential value formulas in this model are estimated separately according to Eqs.2, 3, and 4 (Hwang and Suh, 2004; Chan, 2008):

      L T V = C u r r e n t   v a l u e + P o t e n t i a l   v a l u e
      (2)
      C u r r e n t   v a l u e = t i = 0 N i π p t i 1 + d N i t i
      (3)
      P o t e n t i a l   v a l u e = j = 0 N P r o b i j × P r o f i t i j
      (4)

    • ti customer service period i

    • d = discount rate

    • Ni: total customer service period i

    • π p ( t i ) = past profit of customer i during i

    • probij = the probability to use strategy j in customer i segmentation i

    • 5) Stochastic model: to estimate customer lifetime value in this model, it has been attempted to use Markov chain models to present a more accurate and better method to estimate future customer behavior. In this method, the profitability of customers is determined in different conditions from attraction to rejection to predict profitability in the future. Hyun-Seok Hwang (2016) presented a model to estimate customer lifetime value using stochastic methods (Hwang, 2016).

    However, in other studies, customer lifetime value models are divided into two general deterministic and stochastic approaches and both of them have been effective in the related setting (Esmaeili and Tarokh, 2017). In the following, the findings of two studies based on these approaches will be presented.

    Han et al. (2012) conducted a study to estimate value and segment customers using the data mining technique. In this study, it has been attempted to present a model to estimate customer value and customer segmentation in China’s Telecommunication Industry. According to the presented model, the value of customers of one of China’s telecom operators has been calculated from January to April 2009 has been estimated. In this study, customer segmentation strategies according to customer value have been used using data mining technology to extract customer valuation variables. The presented model in this study can predict the customer lifecycle when only the demographic information of a company’s customers is available. Also, to estimate customer loyalty and credit, the AHP method was used. Unlike other methods, the score of the extracted weights from AHP was estimated according to the experiences of experts and by considering data characteristics. This prevents scoring by some experts according to personal opinions.

    In a study by Kahreh et al. (2014), it has been attempted to estimate the lifetime value of a commercial bank’s customers and classify them in different classes. Besides, the purpose of this study was to achieve a suitable approach for market segmentation according to customer lifetime value. Profit segmentation is an approach for market segmentation that helps define causal factors relative to descriptive factors. In this study, the model presented by Berger and Nasr (1998) was used to estimate customer value. In this model, seven parameters were used including the probability to continue the relationship between customer and bank, average accounts, profit margin, direct expenses of accounts, indirect expenses of accounts, and many periods under study. In this study, the information of 10000 customers in four years from 2005 to 2008 was gathered. Then, using these models and parameters, customer value in each year was estimated and finally, according to the obtained values, customers were classified into six classes and the first class was named gold customers. Also, it has been pointed out that this classification can be used in determining attraction and maintenance strategies (Kahreh et al., 2014).

    A literature review of CLV studies can be divided into three different groups (Ashkezari et al., 2019). The first group focused on profitability calculation models. The main purpose of this group is to provide a suitable model for assessing customer value. Ref. (Burelli, 2019;Sifa et al., 2018;Chang et al., 2012;Esmaeili and Tarokh, 2017;Ashkezari et al., 2019;Gupta and Lehmann, 2003;Wang et al., 2019) can be grouped in this category. The second group focused on how to use these models to segment customers. Ref. (Peker et al., 2017; Chan, 2008;Han et al., 2012) can be grouped in this category. The third group evaluated how customers value the relationship between value interactions with CLV with marketing concepts, such as Customer Repurchasing Behavior (Baidya et al., 2019).

    What is absent from this literature is evidence to calculate the profitability of customers according to customer behavior analysis. In this paper, we proposed a Markov model for real business problems and used the RFM model for defining different states of Markov’s chain. This approach made it possible for us to calculate customer profitability based on predicting customer behavior. Based on this model, companies can establish their campaigns for different segments. In this research, we want to answer two questions:

    • 1) What are the value and status of every customer?

    • 2) What are the future value and the future status of these customers?

    3. METHODOLOGY: A STOCHASTIC MODEL FOR CALCULATING CLV

    The objective of this study is to propose a new framework to estimate customer lifetime value using customer profitability estimation. Since calculating CLV requires considering the state of customer relationship with the firm, the proposed model should encompass the customer situation. The different states of customer relationships with the firm include customer churn, potential customer, and active customer.

    In this method, for calculating time in customers' relationships with firms dynamically, Markov chain model is used. Indeed, we should define different states of customer relationship situation with the firm. According to Hyun-Seok Hwang research, these states can be classified into three situations, P (potential customer state), D (defected customer state), and A (active customer state). Figure 1 illustrates the Customer Relationship Dynamics (CRD) using a stochastic model (Hwang, 2016).

    CRD shows the situation of the customer relationship with the firm during the customer lifetime. Within this period, customers can be in different situations. For example, when a customer is acquired from the potential customer (P) state, the customer stands on the active state. This standard model can be adapted to different business models. For defining different states in this study, the RFM model is used.

    RFM analysis is a proven marketing model for identifying customer's behavior by using certain measures. RFM model helps categorize customers into different classes or states to choose customers who are more likely to respond to promotions and marketing campaigns similarly. The RFM model is calculated on three quantitative dimensions:

    • Recency (R): Has a customer recently deposited his or her money?

    • Frequency (F): How often does a customer deposit his or her money?

    • Monetary (M): How much money has a customer deposited?

    4. INDIVIDUAL CLV EQUATION

    During a customer lifetime, the customer can be in different states. According to the Markov chain model, we can calculate these dynamics. In every state, the customer generates some income and cost for the firm. In a very simple method, incomes include loan interest and costs include expenses for providing services. In summary, customers create a variety of incomes and costs tailored to their states of relationship with the firm. The profit of customer relationship states is unique for each customer. This feature of the model leads to improvement of the accuracy of the model and CLV is calculated for every customer precisely. Although calculating CLV with this framework is very necessary, the common structural model provides the static value of a customer. Thus, it is very important to propose an extended model to handle these dynamics. At the level of each customer, the proposed CLV model based on a stochastic model is as follows (Hwang, 2016;Yang and Lian, 2014):

    CLV = lim n i = 1 N T i R ( 1 + d ) i
    (5)

    • T is the one-step transition matrix of customer i

    • R is the reward vector of customer i

    • d is the discount rate

    For calculating the individual CLV, we use the infinite geometric series as shown in Equations 7 and 8. The matrix I is an identity matrix.

    CLV = T 0 ( 1 + d ) 0 +   T 1 ( 1 + d ) 1 + +   T i ( 1 + d ) i R  
    (6)

    Since the inverse matrix of [I−(1+d)−1T]−1 has been proven, individual CLV can be calculated as the cumulative profitability generated by the customer in all states (Yang and Lian, 2014).

    C L V i =   lim n C L V n
    (7)
    C L V i =   I   1 + d 1 T 1 R
    (8)

    A sequence of transition probability from one state to another in a Markov chain model (MCM) is called a transition probability matrix. The one-step transition matrix that collects all customer states and inter-state transition probabilities is shown in Figure 2. Besides, the reward vector of a customer is a square matrix that includes states' profit and cost for every customer individually. Every element of this matrix represents the profitability of the customer when the customer stays in that state. In this stochastic model, considering the time value of money, discount rate d calculates future profit into the present one. Based on the profit or cost of the customer in each state, all profits or costs are transferred into Net Present Value (NPV).

    For this part of the research, we suggested a new way based on customer profitability analysis (CPA), which can estimate the profit or cost of the customer during customer lifetime in every state.

    5. LIMITING PROBABILITY

    Since it is possible to transfer from one state to another state with positive probability, this model of Markov chain is ergodic or irreducible. If some power of transition matrix with finite-state has only positive elements, the Markov chain is regular. Given that Pn has all positive elements then:

    P T r a n s f r e i n g   f r o m   i   t o   j   i n   m   s t e p s > 0
    (9)

    Therefore, a regular chain is a subset of the ergodic Markov chain. When a regular transition matrix has a finite state space Markov chain, we can assume that lim n P n = W , with common row w. Thus, the r-by-r system of linear equations given by xP= x has a unique probability row vector solution, and this solution is the common row w.

    In this regard, transferring among all states and the expected time in a transition matrix of CLV is finite start-ing at state i, until the process returns to state i. Therefore, this transition matrix has the limiting probability. The limiting probability of the transition matrix shows the long-run ratio of the time within which a customer stays at a special state (Permana et al., 2018).

    Let πi denotes the limiting probability of state i in the transition matrix. πi is calculated by the following equation:

    π j = i π i P i j ,          j π i
    (10)

    where P is the transition probability from state i to state j

    Marketing managers can use the limiting probability to investigate CRD and accordingly, they can launch the marketing campaign for changing the customer costly relationships to profitable relationships (Hwang, 2016).

    6. A CASE OF STUDY IN MENA REGION BANKING INDUSTRY

    The most significant feature of the banking industry is the variety of customer behavior. This complexity has made it more difficult for the banks to navigate toward the appropriate situation. To cope with this problem, managers need a powerful KPI to detect customer behavior that launches appropriate loyalty programs. The customer lifetime value provides the most appropriate KPI for identifying customer behavior dynamically.

    Therefore, the present study aimed to present an intelligent structure to estimate customer lifetime value based on financial and credit behavior of customers. The different steps of this research are presented in Figure 3. This model can create a suitable approach to develop organizational strategies consistent with the behavior of customers. In this approach, the organization can focus on increasing the incomes and satisfy the needs of customers. However, disruptive customers can be managed with appropriate strategies in the same way.

    In this research, based on the RFM model shown in Table 2, eight states including potential customers, highly active customers, active customers, less active customers, churn customers, active service customers, no active service customer, and high-risk customer have been defined.

    Based on this customized CRD, the customer in the active situation may be in 6 states or transfer to churn or potential states. The customized CRD model for the Iranian banking industry is shown in Figure 4. In the active situation, customers can be highly active, active, less active, risky, and no active service state.

    A customer can be transferred between three different areas, which include Active, Churn, or Potential area. Given the complex relationship customer in the active area, we assume the customer can move in each state of this area to churn state. The graphical representation of the state transition is shown in Figure 4.

    For implementing the Markov chain model in this research, following from which we had given the defined finite states, we build built the one-step transition matrix. After this phase, at each customer's level, Reward Vector is calculated. Besides, the limiting probability for each customer will be estimated as well. Finally, individual CLV is computed based on the new extended model.

    7. DATA PREPARATION

    In this study, we used customer transaction data of one of the largest banks in the MENA region. This bank has about 23 million customers for implementing this model. In this research, data sampling was performed based on the cluster sampling method. We selected customers in Tehran and this cluster had about 212956 customers. In the cleaning step, customers’ data were reviewed and data related to 13714 customers with missing value were deleted. We defined features based on the RFM model and used Z-Score normalization for the transformation step. Customers’ data were extracted in the integrated form of a CSV file from the data warehouse.

    8. BUILDING THE ONE-STEP TRANSITION MATRIX

    For calculating the one-step transition matrix of customer i, various probabilities are estimated by customer’s historical past behavioral customer data for about 54 preceding months. Given that the size of the dataset for 199242 customers in this research was very large, we used the SQL server for computing the one-step transition.

    Table 3 shows the situation of customer i during 54 preceding months. The values of this table have been calculated for 199242 customers. According to what we have defined as 8 states for the Markov chain model, the onestep transition has 64 elements which are computed and normalized for estimating the probability of the stay period of a customer in each state. The one-step transition matrix for customer i is shown in Table 4.

    9. REWARD VECTOR OF CUSTOMER I

    Reward vector of customer i is calculated by the amount of provided revenue or expense in each state of the Markov chain model. The accuracy of this part of the research is very important and it significantly influences results in the model.

    Since the objective of the present phase of the study is to propose a new framework to estimate customer profitability, we use thematic analysis which is a method of analyzing qualitative data. It is usually applied to a set of texts, such as interview transcripts. In the analytical model of the study, the most important revenue- expense variables were identified and presented as the suggested model to estimate the reward vector of customers.

    The statistical population included 48 experts with more than 10 years of experience in marketing, accounting, or finance department with academic education in these fields. The statistical sample of this study included 11 banking experts who were selected based on snowball sampling in the content analysis section. In different sections of data collection such as “review and reference to documents”, databases of banks and review were used. Data were analyzed using “content analysis”. Content analysis is an instrument to determine the existence of words or concepts in the text or a set of texts. The content analysis includes the intellectual process for the classification of qualitative data into the clusters of conceptual classes to identify consistent patterns and the relationships between variables and topics (Given, 2008).

    The findings of this part of the study are presented in two general sections of content analysis and model presentation. Content analysis methods include two main methods of “conceptual analysis” and “relational analysis”. In this study, content analysis is based on thematic analysis. Thematic analysis is a method to determine, analyze, and express patterns (themes) that exist in data. This method organizes data and describes them in detail, but it can go beyond this and interpret different aspects of the issue.

    As mentioned earlier, to analyze the data of the interviews, the thematic analysis (TA) method has been used. This method has six steps, including familiarity with data, generating initial codes, searching categories, reviewing categories, defining and naming categories, and finally, preparing a report (Braun and Clarke, 2006).

    • 1. Familiarity with Data: To extract meaning patterns from the heart of the data, the first step is the comprehensive familiarization with data. Converting oral interview data to written text is a useful method to become fully acquainted with data. Regarding the importance of this section in the thematic analysis (Bird, 2005), reading and rereading data helps greatly to become familiar with the data.

    • 2. Generating Initial Codes: In this step, the data are coded systematically, then the related data is continuously updated based on new data (new interviews). In this part of the research, the data obtained from the text of the interviews are coded in several steps, so that the most concise categories yet comprehensible that at the same time are comprehensible are extracted.

    • 3. Searching Categories: In this step, the codes obtained from the previous step are combined in the form of potential categories; all data related to each potential category are collected. To search categories, encoding all data should be done first.

    • 4. Reviewing Categories: In this step, the relationship of the categories with the extracted codes and the entire data set is examined and, if necessary, is reviewed. In other words, this is the step of refining categories into two levels: the first level involves reviewing the extracted codes related to one category indicating that whether these codes have the necessary consistency to form a common pattern. The second level includes reviewing categories concerning the entire data set, indicating whether they reflect the obvious meaning in the data set. The purpose of this step is to analyze the reliability of the accuracy of categories in relation to the entire data and to add codes possibly ignored in the previous steps to the relevant category. The end of this step is the presentation of a thematic map of the analyses. Reviewing the results of the previous step led to the integration of subcategories and its reduction to 5 sub-categories, which were placed in the form of 2 categories according to the similarities and differences.

    • 5. Defining and Naming Categories: This step is the last in refining the categories. Each category is defined by resorting to all the data in its subset, and its relationship to other data and categories is expressed to answer the research question.

    • 6. Preparing a Report: This step is the final step of the analysis that includes the selection of examples from the quotations, final analysis of the selected quotations, returning to the relationship between the analysis and the research question and literature, and finally preparing the scientific report of the analysis.

    The most important variables for the estimation of customer lifetime value based on profitability, which have been explained by experts, are presented in Table 6.

    In this section, the model extracted from the identified variables in content analysis is presented. According to the identified variables by experts, customer profitability analysis is proposed in Table 7. All dimensions of customer revenue- cost relationships are formulated by Kumar and Rajan (2009) and Gupta and Lehmann models (Gupta and Lehmann, 2003;Kumar and Rajan, 2009;Berger and Nasr, 1998).

    C P A j = t = 1 n C o n t r b u t i o n o f c u s t o m e r j t 1 + i t
    (11)

    Contribution of customer j is the profit the firm makes from serving customer j in a given period t, and i is the discount rate when the Weighted Average Cost of Capital (WACC) is considered.

    Now, according to estimations, the total profit from each customer can be estimated:

    IEMS-19-4-744_Eq12.gif
    (12)

    According to this extended CPA model, we can calculate the reward vector for every customer in all states. Thus, the benefit-cost analysis of a firm can be achieved based on all customers’ aggregate reward vectors. The reward vector customer i is shown in Table 8. This customer has created 4020944 IRR during 54 preceding months.

    The total reward vector for 199242 customers is represented in Table 9. This reward vector shows that these customers have created 19.940 billion IRR during their lifetime with this bank.

    Hence, we can compute the one-step transition matrix and the reward matrix for each customer individually. Then, given Equation 8, individual CLV will be calculated based on all customer profits in the past and the future.

    10. LIFETIME VALUE OF CUSTOMER I

    In this section, the presented model was implemented using real information. This proposed model calculated individual CLV for 199242 customers. The lifetime value of customer i is estimated through Equation 8 as represented above. The discount rate, d, is set as 0.0083 (%/month).

    The result shows that the CLV of this customer is 6635656 IRR. Given this stochastic model, marketing managers can provide an appropriate promotion to this customer based on the customer behavior in different states. Thus, they can make a tailored action plan for each customer. According to one to one marketing, this approach proposes an operation plan based on customer behavior to achieve a win-win situation.

    11. THE PRACTICAL SIDE OF RESEARCH

    Proposing an appropriate model for analyzing customer behavior to implement loyalty programs can be very useful in increasing sales and reducing marketing costs. Creating this platform can be very useful in the operational implementation of marketing concepts such as tailored pricing, upselling, cross-selling, etc.

    Marketing managers can define company strategies based on identifying and predicting changes in their customer behavior to manage long-term relationships and customer satisfaction. In the dynamic environment, understanding and predicting customer behavior can help managers to establish effective advertising campaigns (Chen et al., 2004).

    For example, marketing managers can identify customers who are probable to churn in the future, and with segmentation, these customers based on their values can be established in different campaigns to save them in an active relationship with a company.

    Through this model, companies can evaluate the effectiveness of their marketing promotion campaigns; for instance, they can show through holding campaigns how many churned customers are re-attracted to active zone and how much profitability increased during these campaigns.

    What is more, companies can make resource allocation more rational, with the least resource investment for maximum profit return.

    13. LIMITATIONS AND SCOPE OF FURTHER RESEARCH

    Access to customer demographic data (gender, age, occupation, etc.) can be very effective in model development. However, for information security reasons, we could not access these parts of customers' information. Due to limited resources (hardware) for analysis, customers' data of a city were analyzed. It is suggested that using a big data framework, the data related to all customers be analyzed based on this model.

    14. CONCLUSION

    Extensive changes in information technology and the emergence of new concepts in this context have given importance to the implementation of intelligent methods to optimize strategic decisions for different organizations. Therefore, the present study aimed to propose a dynamic structure to estimate customer lifetime value based on the financial and transaction behavior of the customer. The use of this model can create a suitable approach to develop organizational strategies consistent with the behavior of customers. In this approach, the organization can focus on incomes increase as well as satisfy the needs of customers. However, the loss of customers can be managed with appropriate strategies in the same way.

    The most important innovation presented in this study is the development of a dynamic customer lifetime value model consistent with the business model of the organization through content analysis of qualitative interviews to present an efficient pattern for estimating customer value. Moreover, the implementation of the suggested model in the real context of one of the largest Iranian banks can be considered as the special advantage of this study.

    Eventually, it has been attempted to design a dynamic customer lifetime value analysis algorithm based on profitability analysis. According to the importance of accurate identification of factors affecting profitability estimation, it is suggested to investigate the model for predicting customer behavior based on the Markov chain model. Based on this proposed model we can answer research questions:

    • 1. What are the value and status of every customer? Based on this model, we were able to calculate the profitability of each customer. This model also allowed us to identify the status of each customer's interaction with the bank. This means that we can highlight the relationship between each customer with the bank like churn, active zone, or potential situation.

    • 2. What are the future value and the future status of these customers? Based on this model, we can predict the future value and future status of relationships with a bank for each customer.

    To summarize our contribution, the present work proposed a hybrid model with a Markov model and RFM model to calculate CLV individually in the banking industry. Besides, based on identifying customer behavior stochastically, by this model, we can design a new model for establishing marketing programs.

    Figure

    IEMS-19-4-744_F1.gif

    Customer relationship dynamics (Hwang, 2016).

    IEMS-19-4-744_F2.gif

    The one-step transition matrix (Hwang, 2016).

    IEMS-19-4-744_F3.gif

    The different steps of research.

    IEMS-19-4-744_F4.gif

    Graphical representation of the markov chain for a case study.

    Table

    Different approaches to CLV calculation

    States of markov chain model based on RFM model

    Historical behavior of Customer i

    Transition Probabilities of Customer i

    Statistical society information

    Content analysis of interviews with banking experts

    Extended formula based on CPA

    The Reward vector customer i

    The total Reward vector for 199242 customers

    The CLV of customer i.

    REFERENCES

    1. Ashkezari, M. S. , Albadavi, A. and Ashkezari, M. S. (2019), Presenting a model for calculating ISACO customers lifetime value by markov chain using data mining techniques, International Journal of Industrial Engineering & Production Research, 30, 429-441.
    2. Baidya, K. M. , Maity, B. , and Ghose, K. (2019), Innovation in marketing strategy: A customer lifetime value approach, Journal of Business and Management, 25(2), 71-97.
    3. Bekamiri, H. , Mehraeen, M. , Pooya, A. , and Sharif, H. (2020), The banking industry foresight using the scenario planning approach and the cross-effects matrix, Journal of BI Management Studies, 8(33), (In Persian), 2020.
    4. Berger, P. D. and Nasr, N. I. (1998), Customer lifetime value: Marketing models and applications, Journal of Interactive Marketing, 12(1), 17-30.
    5. Bird, C. M. (2005), How I stopped dreading and learned to love transcription, Qualitative Inquiry, 11(2), 226-248.
    6. Blattberg, R. C. and Deighton, J. (1996), Manage marketing by the customer equity test, Harvard Business Review, 74(4), 136-144.
    7. Braun, V. and Clarke, V. (2006), Using thematic analysis in psychology, Qualitative Research in Psychology, 3(2), 77-101.
    8. Bult, J. R. and Wansbeek, T. (1995), Optimal selection for direct mail, Marketing Science, 14(4), 343-466.
    9. Burelli, P. (2019), Predicting customer lifetime value in free-to-play games, Data Analytics Applications in Gaming and Entertainment, CRC Press, Chapter 5, 79-107, cited 2019 July 11,
    10. Čermák, P. (2015), Customer profitability analysis and customer life time value models: Portfolio analysis, Procedia Economics and Finance, 25, 14-25.
    11. Chan, C. C. H. (2008), Intelligent value-based customer segmentation method for campaign management: A case study of automobile retailer, Expert Systems with Applications, 34(4), 2754-2762.
    12. Chang, W. , Chang, C. and Li, Q. (2012), Customer lifetime value: A review, Social Behavior and Personality: An International Journal, 40(7), 1057-1064.
    13. Chen, M. C. , Chiu, A. L. , and Chang, H. H. (2004), Mining changes in customer behavior in retail marketing, Expert Systems with Applications, 28(4), 773-781.
    14. Delafrooz, N. and Farzanfar, E. (2016), Determining the customer lifetime value based on the benefit clustering in the insurance industry, Indian Journal of Science & Technology, 9(1), 1-8.
    15. Esmaeili, G. M. and Tarokh, M. J. (2013), Customer lifetime value models: A literature survey, International Journal of Industrial Engineering & Production Research, 24(4), 317-336.
    16. Esmaeili, G. M. and Tarokh, M. J. (2017), A stochastic approach for valuing customers, International Journal of Software Engineering and its Applications, 9(3), 59-66.
    17. Given, L. M. (2008), The SAGE Encyclopedia of Qualitative Research Methods, Thousand Oaks, SAGE Publications, Inc. CA.
    18. Gupta, S. and Lehmann, D. R. (2003), Customers as assets, Journal of Interactive Marketing, 17(1), 9-24.
    19. Gupta, S. , Hanssens, D. M. , Hardie, B. G. S. , Kahn, W. , Kumar, V. , Lin, N. , Ravishanker, N. , Sriram, S. , Sriram, N. R. S. , Kumar, V. , and Gupta, S. (2006), Modeling customer lifetime value, Journal of Service Research, 9(2), 139-155.
    20. Hamdi, K. and Zamiri, A. (2016), Identifying and segmenting customers of pasargad insurance company through RFM model (RFM), International Business Management, 10(18), 4209-4214.
    21. Han, S. H. , Lu, S. X. , and Leung, S. C. (2012), Segmentation of telecom customers based on customer value by decision tree model, Expert Systems with Applications, 39(4), 3964-3973.
    22. Hiziroglu, A. and Sengul, S. (2012), Investigating two customer lifetime value models from segmentation perspective, Procedia-Social and Behavioral Sciences, 62, 766-774.
    23. Hu, H. , Lin, J. , Qian, Y. , and Sun, J. (2018), Strategies for new product diffusion: Whom and how to target?, Journal of Business Research, 83, 111-119.
    24. Hwang, H. S. (2016), A stochastic approach for valuing customers: A case study, International Journal of Software Engineering and Its Applications, 10(3), 67-82.
    25. Hwang, H. , Jung, T. , and Suh, E. (2004), An LTV model and customer segmentation based on customer value: A case study on the wireless telecommunication industry, Expert Systems with Applications, 26(2), 181-188.
    26. Jain, D. and Singh, S. S. (2002), Customer lifetime value research in marketing: A review and future directions, Journal of Interactive Marketing, 16(2), 34-46.
    27. Jain, D. and Singh, S. S. (2002), Customer lifetime value research in marketing: A review and future directions, Journal of Interactive Marketing, 16(2), 34-46, 2002.
    28. Kahreh, M. S. Tive, M. , Babania, A. , and Hesan, M. (2014), Analyzing the applications of customer lifetime value (CLV) based on benefit segmentation for the banking sector, Procedia-Social and Behavioral Sciences, 109, 590-594.
    29. Kahreh, S. M. , Tive, M. , Babania, A. , and Hesan, M. (2014), Analyzing the applications of customer lifetime value (CLV) based on benefit segmentation for the banking sector, Procedia - Social and Behavioral Sciences, 109, 590-594.
    30. Khajvand, M. , Zolfaghar, K. , Ashoori, S. , and Alizadeh, S. (2011), Estimating customer lifetime value based on RFM analysis of customer purchase behavior: Case study, Procedia Computer Science, 3, 57-63, 2011.
    31. Kotler, P. (1974), Marketing during periods of shortage, Journal of Marketing, 38(3), 20-29.
    32. Kumar, V. and Petersen(, J. A. (2012), Statistical methods in customer relationship management, Wiley Online Library, cited 2012 August 7, Available from: http: //www.scopus.com/inward/citedby.url?scp=84891584262&partnerID=8YFLogxK.
    33. Kumar, V. and Rajan, B. (2009), Profitable customer management: Measuring and maximizing customer lifetime value, Management Accounting Quarterly, 10(3), 1-18.
    34. Peker, S. , Koçyiğit, A. , and Eren, P. (3017), LRFMP model for customer segmentation in the grocery retail industry: A case study, Marketing Intelligence & Planning, 35(4), 544-559.
    35. Permana, D. , Pasaribu, U. S. , Indratno, S. W , and Suprayogi, S. (2018), Convergence of transition probability matrix in CLV Markov models, IOP Conference Series: Materials Science and Engineering, 335.
    36. Sifa, R. , Runge, J. , Bauckhage, C. , and Klapper, D. (2018), Customer lifetime value prediction in non-contractual freemium settings: Chasing high-value users using deep neural networks and SMOTE, Proceedings of the 51st Hawaii International Conference on System Sciences, 923-932.
    37. Wang, X. , Liu, T. , and Miao, J. (2019), A deep probabilistic model for customer lifetime value prediction, Available from: https://arxiv.org/abs/1912.07753.
    38. Yang, X. and Lian, L. (2014), A new data mining algorithm based on MapReduce and Hadoop, International Journal of Signal Processing, Image Processing and Pattern Recognition, 7(2), 131-142.