Improving the diagnosis and prediction of customer churn: A heterogeneous hazard modeling approach

Abstract

The authors develop a hazard modeling approach to predict customer churn and to study the nature of the empirical link between customer churn and factors such as customer service experience, failure recovery, and payment equity. The approach uses a latent class Weibull hazard model with time-varying covariates. The model incorporates heterogeneity in both baseline hazard probabilities and in response parameters. The authors apply the model to the churn prediction problem at a continuous service provider, a direct-to-home satellite television firm based in a South American country. The empirical results show that the prediction of customer churn is significantly improved when heterogeneity is added to the customer churn rates and to the response parameters. Significant links are found between churn rates and variables capturing customer service experience, failure recovery efforts, and payment equity. Results also show important differences in the magnitude and significance of the response parameters across latent classes.