Modelling customer lifetime value in iGaming

Abstract

The shift of service-based products to web-based platforms for industries such as that of iGaming has led to a change in growth strategies adopted by companies in these industries. Rather than simply focusing on developing new products to obtain new customers, focus on customer value has been taking the spotlight, where higher value customers are targeted with the aim of retention. Traditionally, techniques such as Recency Frequency Monetary-value models and Pareto/Negative Binomial Distribution models have been used as a way to capture customer activity, consisting of financial data and engagement rates, in order to obtain predictions of customer lifetime value. With the emergence of big data, vast amounts of new datapoints are available on individual customers which are not used by these models. As such one is enticed to explore new techniques such as those from statistical machine learning. This dissertation explores three techniques from this field, namely Random Forests, Neural Networks and Neural Random Forests. With the use of K-foldcross-validation, the performance of the three models is evaluated on a dataset containing customers active with a local iGaming company.