Executive Summary
Long before the concept of “customer lifetime value” (CLV) became a popular phrase among marketing practitioners and academics, database marketers were using simple summary statistics to assess the value of different customer groups on the basis of their prior behavioral patterns. The most popular framework classifies customers in terms of the recency, frequency, and monetary value (RFM) of prior transactions.
Today, both CLV and RFM have gained enormous popularity from a wide range of practitioners, and there is a great deal of interest in linking the simplicity of RFM (an effective “backward-looking” summary of prior behavior) with the allure of CLV (the net present value of all future cash flows associated with a customer). The goal of this article is to develop a formal model that requires nothing more than RFM inputs to make specific lifetime value projections for any given customer.
Central to the presentation of the results of this model is the notion of an “iso-value” curve, which enables the authors to group together individual customers who have different behavioral histories but similar future valuations. For example, consider two different customers: The first one has high frequency, medium recency, and low monetary value, and the second one has low frequency, high recency (i.e., just made a purchase), and medium monetary value. Despite the behavior differences underlying these buying patterns, both customers might have similar CLV estimates (and therefore might merit similar attention and treatment from the firm). Iso-value curves make it easy to visualize and summarize the main interactions and trade-offs among the RFM measures and CLV.
To obtain these iso-value curves, the authors combine two separate modeling components: one that captures the “flow” of transactions over time and one that focuses solely on revenue per transactions. Using data from an online retailer of music CDs (covering more than 20,000 customers across an 18-month time period), the authors estimate the complete model and evaluate the results. They validate each submodel at both the aggregate and the disaggregate level, primarily by examining forecasting performance in a longitudinal holdout period. The results of the tests are encouraging and provide the confidence to project customers’ future behavior and thus obtain valid and meaningful estimates of CLV.
A careful examination of the empirical iso-value curves reveals some surprising but believable results. For example, the authors identify the “increasing frequency paradox.” Under most circumstances, greater frequency for a given customer would be associated with higher levels of CLV, but for customers who have not purchased in a long time (i.e., low recency), higher levels of frequency indicate a greater likelihood that the customer is no longer active with the firm (and therefore has a lower CLV). The authors also demonstrate that the “zero class” (i.e., customers who have made no repeat purchases) constitutes an unexpectedly large portion of the overall future value of the entire customer base. Although each of these customers provides little value individually, the size of the group (roughly 50% of all customers) makes the collective CLV for this group much larger than that of most other RFM segments. After pointing out several substantive insights such as these, the authors discuss a set of broader managerial issues and opportunities associated with the use of this modeling framework and its applications in actual practice.
Biography
Peter S. Fader is Frances and Pei-Yuan Chia Professor of Marketing in the Whartson School at the University of Pennsylvania. When not evaluating the coolness of dollar bill serial numbers on his Web site (www.coolnumbers.com), he messes around other kinds of data. The article in this issue is a nice example, but as nice as it is, it is still not as cool as a number such as 17928161.
Bruce G.S. Hardie is Associate Professor of Marketing at London Business School. He holds a master’s degree and a doctoral degree from the University of Pennsylvania and a bachelor’s degree and a master’s degree from the University of Auckland (New Zealand). His primary research interest lies in the development of data-based models to support marketing analysts and decision makers. His current projects focus on the development of stochastic models for forecasting the sales of new products and for customer base analysis. His research has appeared in academic journals, such as Marketing Science, Journal of Marketing Research, Marketing Letters, European Journal of Operational Research, and Journal of Forecasting.
Ka Lok Lee is a dual-degree graduate from the University of Pennsylvania. He holds a bachelor’s degree in Mathematics from the School of Arts and Sciences and a bachelor’s degree in Economics from the Wharton School. Ka has consulting experiences in the consumer credit card market and the pharmaceutical industry. He has a special interest in building forecasting models for various marketing applications (customer base analysis, customer lifetime value models, and new product sales forecast) and making comparisons of different methods to solve the same statistical problems. Ka works for Catalina Health Resource as a market research analyst in analytical services.
J Marketing Research, Volume 42, Number 4, November 2005
View Table of Contents.