Ranking customers for marketing actions with a two-stage Bayesian cluster and Pareto/NBD models

Abstract

Modelling customer behaviour to predict their future purchase frequency and value is crucial when selecting customers for marketing activities. The profitability of a customer and their risk of inactivity are two important factors in this selection process. These indicators can be obtained using the well-known Pareto/NBD model. Here we cluster customers based on their purchase frequency and value over a given period before applying the Pareto/NBD model to each cluster. This initial cluster model provides the customer purchase value and improves the predictive accuracy of the Pareto/NBD parameters by using similar individuals when fitting the data. Finally, taking the outputs from both models, the initial cluster and Pareto/NBD, we present some recommendations to classify customers into interpretable groups and facilitate their prioritisation for marketing activities. To illustrate the methodology, this paper uses a database with sales from a beauty products wholesaler.