Modelling customer churn behaviour in the online gambling industry

Abstract

The online gambling industry is one of the most high-revenue generating branches of the entertainment business, resulting in fierce competition. With this competition, customers are churning from one company to another at an alarming rate. Current churn literature reveals the fact that it is cheaper to retain customers than acquire new ones. The ever-growing data make the tasks of performing, analysing and forecasting future trends more complicated. The solution lies in the use of statistical and probabilistic tools for modelling churn behaviour of the customers. In this dissertation, model-based clustering, continuous-time Markov chain and survival models have been used in attempt to construct an idealised customer churn behavioural model. A database containing 32,582 customers was used as a testing ground. Distributional fits for several variables concerning the betting history yielded various parameter estimates which were subjected to clustering techniques. Clusters were then 'profiled' based on the customer's value and comparative analysis of the survival functions was performed. The amount of money wagered suggested a stochastic setting. A continuous-time Markov chain setting was then created and fitted to the data. This leads naturally to survival estimates for the transition rates describing the random time taken for a Markov chain to reach its absorbing state - the churn state.