Abstract
Modeling customer purchases is a common but complex task. In non-contractual business settings, a popular approach to disentangle this complexity is to model three processes: customer attrition, transaction, and spending behavior. Typically, the former two are estimated separately from the latter. This disjointed modeling approach has regularly raised methodological concerns. However, marketers have widely adopted this approach. Its versatility, scalability, and accessibility have not been matched by alternatives that estimate all three processes simultaneously. Accounting for both rigor and applicability, this study introduces a novel model for customer base analysis. Generalizing previous work, this single probabilistic model captures all three processes with well-established statistical distributions. Optionally, covariates can be included for each process. Further, expressions to readily derive key managerial metrics are available. The model has a closed-form likelihood for scalable estimation and is widely accessible through an open-source implementation. To evaluate the model, a series of steps are taken: First, the identification of exogenous and endogenous covariate effects is validated through a simulation study and illustrated with a case study. Second, improvements in predictive performance of up to 35% are demonstrated for multiple real-world datasets. Given these results, the model enhances marketers' ability to conduct customer base analyses.