ARTICLE IN PRESS

Journal of Retailing and Consumer Services 12 (2005) 431–441 www.elsevier.com/locate/jretconser

An empirical studyof dynamiccustomer relationship management

Chunqing Lia,b, Yinfeng Xuc, Hongyi Lid,Ã

aSchool of Economics and Management, Tsinghua University, Beijing 100084, China bSchool of Economics and Management, Xi’an Institute of Technology, Xi’an, Shaanxi 710032, China cManagement School, Xi’an Jiaotong University, Xi’an, Shaanxi 710049, China dDepartment of Decision Sciences and Managerial Economics, Chinese University of Hong Kong, Shatin, NT, Hong Kong

Abstract

We applythe Go¨ nu¨ l and Shi (1998) approach to the analysis of the optimal messaging and pricing policy mix by studying the past transaction patterns between a local supermarket and its consumers. We develop a dynamic customer relationship management model and investigate the relationship between customer utilityand purchasing frequencybymodifyingthe return function of the model discussed in Go¨ nu¨ l and Shi (1998). In particular, we extend the analysis to consider a messaging and pricing policy mix, and we use a genetic algorithm in our empirical estimation. When applied to some non-seasonal products in a local supermarket, we find that our model is suitable and far superior to the one-stage model commonlyused. Our dynamicmodel gives the optimal marketing mix strategies in different customer states and the results show that the firm could enjoya 22% increase in profit. r 2005 Elsevier Ltd. All rights reserved.

Keywords: Dynamic customer relationship management; Customer utility; Customer lifetime value; Markov-perfect equilibrium

1. Introduction effectivelyutilize the customer databases to manage the customer relationship. It is more important to capture The development of information and communication information about your customers than just to build up technologyprovides manycompanies with a convenient a database. environment to collect detailed data about their However, the potential difficultyof converting data individual customers. These companies can take advan- into profits is in how to obtain relevant information tage of customer data to improve their marketing from the data and customize the marketing mix policies policies to attract more customers and therebyincrease to satisfythe consumer’s wants and needs. As pointed their profit. However, while manyfirms have invested out in Accenture (2004), all customers are not created millions in the creation of massive customer information equal. Customer data are valuable onlyif it is translated databases, the goal of creating improved marketing into profitable interactions. Powerful analytical mathe- policies has often proved elusive (Hoffman, 2001; King, matics and statistics models are essential to identifythe 2001). While the trend continues of firms investing large most profitable customers and predict future profits amounts of moneyto collect and store customer data based on the past historyof customers’ data. Modeling (Accenture, 1998), there have recentlybeen a few high- of customer data has become an increasinglyimportant profit firms that have reversed this current and issue in customer relationship management (CRM). The abandoned their customer data collection efforts (Hunt, basic idea of CRM has been embraced and the potential 1999). Thus, the emphasis is now on the issue of how to benefits of relationship marketing based upon individual characteristics are generallyaccepted ( Peppers and ÃCorresponding author. Rogers, 1993). Over the past 10 years there has been a E-mail addresses: [email protected] (C. Li), yfxu@mail. rapid growth of research topics on modeling the xjtu.edu.cn (Y. Xu), [email protected] (H. Li). marketing mix as a function of customer relationship

0969-6989/$ - see front matter r 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.jretconser.2005.01.003 ARTICLE IN PRESS 432 C. Li et al. / Journal of Retailing and Consumer Services 12 (2005) 431–441 status, especiallytaking into consideration both the these tactics. In the new model, we use the continuous customer’s utilityand a firm’s profit. purchase times to replace the accumulative purchase Some important research considers the optimal times so as to reduce the customer state spaces and mailing policyin direct marketing that contributes to decrease the computational complexity. We modify and an increasinglylarge share of business sales. Bult and verifythe relationship between customer utilityand Wansbeek (1995) construct binarychoice stochastic continuous purchase time and modifythe profit function models to determine a mailing strategybased on the of the firm byapplying the concept of customer lifetime response behavior of customers. DeSarbo and Ramas- value (CLV) to obtain a better description of the actual wamy(1994) consider latent class models. Neural CRM situation of the local supermarket under con- networks are considered in Levin and Zahavi (1996). sideration. We estimate the customer discount factor as Several other papers applya Markov decision model for a parameter rather than taking it as a given constant direct mailing decisions, see Bitran and Mondschein which was commonlydone in the literature. In this way (1996) and Go¨ nu¨ l and Shi (1998), among others. In we can take the consumers’ heterogeneityinto account. particular, Go¨ nu¨ l and Shi (1998) were the first to Finally, we utilize a genetic algorithm to estimate the propose combining the customer’s utility-maximization model and test the hypothesis by applying the model to problem with a firm’s profit-maximization problem and some non-seasonal products data from the local super- thus solve the optimal timing and spacing of direct market. mailing of catalogs to customers. The paper is organized as follows. Section 2 presents Applying the Go¨ nu¨ l and Shi (1998) approach, we the model of dynamic customer relationship manage- develop a dynamic customer relationship management ment and the algorithm to solve the optimal marketing (DCRM) model and investigate how the marketing mix policy. Section 3 describes the data used in this policies can be improved byusing feedback from application and the genetic algorithm used for the customer transaction records. We contribute to the global optimum solution. Furthermore, the estimation literature in the development, estimation, and testing of results are presented. Section 4 analyzes the optimal an analytical DCRM model that deals with marketing marketing mix policyagainst the customer database. mix policies. Furthermore, we model customers’ pur- Finally, in Section 5 we offer our conclusions. chase behavior as a function of current and future marketing decisions in addition to the time elapsed since the last transaction and the number of continuous 2. Model of dynamic customer relationship management purchases. We also contribute in the use of a genetic algorithm when estimating the model. Bysolving the 2.1. Research background and estimation of OMP model model we arrive at new policies that are both utility- maximizing for the customer and profit-maximizing for The OMP model, combining a firm’s profit-maximiz- the firm over an infinite time horizon. ing problem and the customer’s utility-maximizing The Go¨ nu¨ l and Shi (1998) approach gave us the initial problem, was first presented by Go¨ nu¨ l and Shi (1998). idea of the DCRM. Theypresent a dynamicstructural The object of their research was the direct mail order model that estimates the consumer response to catalog industry. In the OMP model, the key factors affecting mailings, on the assumption that consumers act as the optimal mail policywere investigated bymaximizing dynamic optimizers. In this context, the assumption customer utilityand the direct-mail company’s profit. implies that a significant part of the motivation for This model utilized the estimable structural dynamic consumers to make purchases is to ensure that theywill programming (ESDP) technologyto model customer continue to be solicited. While this paper represents a utilityand the firm’s profit. Theyconsidered the simple type of customer management, the simple exchange between the customer and the firm as a decisions to mail or not to mail are insufficient for most stochastic game process and an equilibrium solution was settings. For example, marketing tactics such as pricing obtained. The main contribution of the model is that the and operational characteristics such as messaging are authors built, estimated, and tested one model that likelyto have significant effects on customer retention. manages the mail policy. The components the model This paper modifies the optimal mailing policy(OMP) considered were the recencyof the last purchase, the model suggested by Go¨ nu¨ l and Shi (1998). While the accumulative purchase time that describes the custo- same hypothesis is tested using the model, we propose a mer’s purchasing behavior, and the mail policy. Using dynamic customer relationship management model to the model, a firm can get the optimal mail policyto illustrate techniques for converting the information maximize customer utilityand its own profit in an contained in customer databases into meaningful knowl- infinite time horizon. edge and improved marketing policies. We take a mix of The basic assumption in the OMP model is that both pricing and messaging as the main marketing tactics and the firm and the customers consider not onlythe current measure the sensitivityof the customer’s response to benefit but also the future benefit of a decision. The