Online customer identification based on Bayesian model of interpurchase times and recency

Abstract

When it comes to customer segment, most researchers lay great emphasis on the aggregate view but neglect the individual view. In this article, the primary focus will be on both. The main purpose is to analyse how the activity of website browsing contributes to the segmentation of online customers (aggregate view) and how the current and prospective activities of the individual can be accurately determined and predicted (individual view). In aggregate view, we employ a finite mixture model on interpurchase times to segment customers into three states: super-active, active and inactive. In individual view, we apply the hierarchical Bayesian model of interpurchase times to achieve heterogeneity across customers. However, this heterogeneity is inadequate in the evaluation of the individual activity in the present or near future. Moreover, recency is one of the most effective behavioural indices on the relationship strength and one of the meaningful predictors for customer reconsumption. Specifically, in individual view, this article adopts not only the interpurchase times to evaluate customer status, but also adds a new variable or recency, to execute this evaluation. The procedure of customer identification via interpurchase times and recency are applied to the case study of an online company. The result of this case study shows that this approach provides a more genuine customer status than the procedure that merely involves interpurchase times. Further, inferences of mixture model and Bayesian model in our case study are solved by Winbugs, a software package of Markov Chain Monte Carlo method.

Keywords:

• hierarchical Bayes
• finite mixture model
• heterogeneity across customers
• behavioural relationship strength
• Markov Chain Monte Carlo method

Notes

Note

1. The formal definition of the natural conjugate prior is quoted form Robert (2001): A family F of probability distributions on Θ is said to be conjugate (or closed under sampling) for a likelihood function p(X|θ) if, for every π ∈ F, the posterior distribution π(θ|X) also belong to F.