Modeling customer relationships as Markov chains

Abstract

A major issue that dominates the current thinking of catalogers is how to send mail more efficiently in order to increase response rates and decrease wasteful mailing activity. Currently cataloguers assume that a mailing decision should be made when expected profit (or lifetime value) is positive or at best when expected profit exceeds a certain threshold. We offer a new perspective. We suggest that a customer be sent the current catalog only if the expected profit with the current catalog exceeds the expected profit without the current catalog. We develop, estimate, and test a hazard function model with unobserved heterogeneity to estimate the sensitivity of the customers’ purchase intensity to catalog drop and to other factors. We explicitly allow for customers to order from previous catalogs and control for the effects of promotional rest and wearout. We find that it is better to send (not too many but) a few catalogs with (not too long but) a moderate amount of time between them in order to encourage purchases.