A model of customer lifetime value and ex poste value-based segmentation: An application in a financial multi-services retailer based on probabilistic and data mining models

Abstract

Customer lifetime value (CLV) analysis is a generic term for methodologies that study the net present value of benefits associated with each customer, once he or she has been acquired, after subtracting incremental costs associated with each customer (e.g., marketing, selling, production and service), over his or her entire lifetime with the company. In particular, considering our selected context (a Spanish financial service provider) and the data available (a panel data of customers) described in the following Chapters, we define CLV as the present value of each customer’s current and future purchases of banking products. More specifically, CLV is the net present value of the sum of the current and future contribution margins from the customers of the company, which depends on length, depth and breadth of the relationship with each customer, over their lifetimes of operation with the company, taking into account the time value of money using a discount rate to adjust back the predictions about the future to the present. From a general marketing perspective, this research is motivated by the fact that some recent trends in banking context (e.g., better customer management, focused on strengthening relationships with existing customers through excellence in service quality or develop a customer-centric banking) encourage banks to achieve important challenges in order to stay competitive, especially since the advent of the current financial crisis. Therefore, in order to help to overcome the current economic difficulties we want to offer a tool to manage banking customers. This tool is going to impulse both sides: banks (our CLV model can be considered as a Customer Relationship Management (CRM) tool, which offers an economic assessment of customers) and customers (because this tool can help the bank to understand customer’s behaviours and anticipate their needs, which facilitates the relationship and exchange between bank and customers). From a theoretical point of view, customers and customer relationships have been considered intangible and valuable firm assets since decades. If we link this theoretical proposition with an empirical goal, the result is in line with the analysis of historical records of interactions between the customer and the company in such a way that companies will be able to obtain valuable information that will help them to understand customer’s behaviours and anticipate his/her needs, as we have noted previously. This way to proceed ultimately will impact on business performance and in the customer satisfaction with the offer. As firms increasingly see customers as important assets, methods for estimating CLV have been developed as an important strategic marketing tool. For the reasons previously mentioned, we have decided to deal with CLV in a banking context, in order to calculate the value from customers as the base for an ex poste segmentation scheme. In this way, we present a new research design to obtain a richer customer segmentation taking into account the CLV that each customer brings to the firm (and other sociodemographic information). While traditional customer segmentation is focused on identifying customer groups only using demographics and other attributes (such as attitude and psychological profiles), CLV allows us to undertake customer segmentation in a different way: a value-based segmentation approach. From a methodological perspective, the objective of this dissertation is to present a new mixture of statistical techniques to model CLV. Customer valuation and customer segmentation problems in marketing have been tackled previously, although in this research we propose a new empirical design that solves both problems in a different and particular way, that is, (1) firstly, estimating ܮܥܸ௜, where i refers to each customer, and (2) secondly, segmenting customers according to this individual value and other customer characteristics, such as sociodemographic information (i.e., age, gender and income). For these tasks, we have selected certain components and drivers of CLV considered essential in the customer-company relationship. In particular, regarding components of CLV we refer to retention (length dimension), product ownership (breadth dimension), product usage (depth dimension) (for more details about these three dimensions see Bolton et al., 2004), contribution margin and also a discount rate to adjust back the predictions about the future to the present. At the same time, we have analysed the underlying behaviours (drivers of CLV) that define length, breadth and depth dimensions that jointly predict CLV (in the following Chapters we give more details about these predictors). Therefore, using monthly data from a database of 1.357 customers of a Spanish financial services company (a multi-service or multi-product retailer), we present a probability model, in particular a hierarchical Bayesian model, used (1) firstly, to discover those customer characteristics with more potential to predict retention (length dimension), product ownership (breadth dimension) and product usage (depth dimension) and also contribution margin, and (2) secondly, to predict these quantities (using the drivers of CLV) that jointly help us to calculate lifetime value of each customer in our sample (ܮܥܸ௜, where i refers to each customer). Once we identify the most significant predictors of value and we predict length, breadth, depth dimensions and contribution margin, we get the ܮܥܸ) ௜using the formulas enclosed in Chapter 5). As we have mentioned in the previous paragraph, this amount and some socio-demographic information are inputs of an ex poste segmentation stage (a value-based segmentation). For this second task, a data mining technique is used, in particular a regression tree. This segmentation is performed in order to identify those groups of customers that are more/less valuable. Our aim pursues to propose the implementation of different strategies to manage different types of customers of the bank. Therefore, both analyses (hierarchical Bayesian models and regression trees) provide excellent opportunities to design a framework that takes into account their interdependencies. Ultimately, these models allow a careful assessment of the contribution of each customer within his/her entire lifetime of operation with the bank and provide a potentially powerful CRM tool to the bank.