Game Over: An Application of Customer Churn Prediction using Survival Analysis Modelling in Automobile Insurance

Abstract

Customer churn poses a big challenge in financial institutions including the field of insurance. Hence, retention analyses are designed to predict time to churn and identify which among the customers are most likely to leave the company so that retention campaigns and strategies may be done to address the issue. By using 6-year data of individual insurance policyholders from an automobile insurance company in Asia, the study utilized Kaplan-Meier, Log-rank tests and Cox-Proportional Hazard model to understand the risk of churning and to identify significant factors affecting the said event of interest. Demographic variables, including age, marital status, region, as well as policy and claim details, including mortgagee and claim indicator, were found to be significant in the hazard model. Furthermore, results of K-Prototypes, a clustering algorithm used for handling mixed data types, show that customers can be grouped into three (3) clusters and that strategies must be geared towards retaining the young, single and working adults' segment.