The Results from the Lifetime Value and Customer Equity Modeling Competition

Customer lifetime value (CLV) is the present value of all the future cash flows attributed to a customer relationship and customer equity (CE) is the sum of CLV values across a set of customers. They are central concepts in direct and interactive marketing, and publishing original research on these subjects has long been a focus of Journal of Interactive Marketing. As an indication of the importance that JIM places on CLV and CE, the recent 10th anniversary issues included four review articles on different aspects of the subject (Fader and Hardie 2009; Kumar et al. 2009; Blattberg, Malthouse and Neslin 2009; Gupta 2009), and past two years we published other articles (Reinartz, Thomas and Bascoul 2008; Dreze and Bonfrer 2008; Boehn 2008; Fassnacht and Kose 2007; Johnson 2007). Organizations use these concepts for many purposes. One of the earliest was to determine howmuch can be spent to acquire a customer. CLV enables an organization to acquire customers whose acquisition costs exceed the profit from the first order because they know that the aggregate profits from the customer will eventually exceed the cost of acquisition. Others suggest using CLV to determine the amount of marketing resources that can be “invested” in certain customers (e.g., see Kumar et al. 2009). The word invested is central to the CLV approach: marketing activities are to be thought of as investments in customer assets rather than expenses. Another application is to use CLVand CE estimates to place a financial value on having a database of customers (e.g., see Gupta 2009). Many firms such as Amazon have justified their stock prices in part by the size of their customer databases. While there has been a large amount of research done on estimating CLVand CE with various methods, we do not know which work best. There have been other competitions focused on predicting response to an offer (e.g., KDD98) and customer retention (e.g., Duke Teradata Churn modeling competition), but no contests have focused on CLVand CE to our knowledge. The purpose of this contest was to compare models. A secondary goal was to provide the research community with a large, high-quality, publicly available data set that spans a long time. Such data sets are important so that different researchers can compare their models on common data. We are pleased that this dataset has already been analyzed in detail as part of a master's thesis (Platzer 2008), which also gives a thorough discussion of the modeling issues. To achieve these goals, the Journal of Interactive Marketing (JIM) the Direct Marketing Educational Foundation (DMEF) and the DMA Non-profit Federation organized this competition. A leading US nonprofit organization provided detailed transaction and contact histories for 1 million donors over 15 years (1992–2006). This data set is now available from the DMEF for academic research. Participants in the contest analyzed a sample of 21,166 donors acquired during the first half of 2002. They were provided with transaction and contact history through August 31, 2006 and asked to predict the behavior of donors during the two-year “target period” from September 1, 2006–August 31, 2008. The following criteria were used to judge the models: