Predictive modelling of customer lifetime value in online gaming

Abstract

As companies are increasingly collecting more information about their customers’ transactions on a granular level, it is becoming more popular for businesses to assign a value to each of their customers, or particular customer segments. This is commonly referred to as customer lifetime value (CLV). Several statistical models have been developed along the years in order to determine the structure of the customer base, as well as to predict the future activity of a customer. This is then linked to the customer lifetime value, and the top customer segments are identified, leading to a better allocation of marketing spend. A number of probabilistic models are reviewed in order to analyse the customer base of an online gaming company based in Malta, which is a continuous, non-contractual business setting. New acquired customers in 2016 are particularly considered, with the aim of using the first three months’ activity for modelling and then predicting the activity in the coming three months. The models employed cover the frequentist and the Bayesian approaches, as well as the parametric and non-parametric settings. These include the Pareto/Negative-Binomial model, hierarchical Bayes and Gaussian process regression.