A Case Study on Reducing Auto Insurance Attrition with Econometrics, Machine Learning, and A/B Testing

Abstract

Acquiring a new customer costs insurance companies several times more than what it costs to retain current ones, which is why reducing attrition (also known as churn) is so important. To retain their customers, insurance companies need to do three things: a) understand why clients churn, b) predict which clients will churn, and c) try to retain potential churning clients through interventions or campaigns that address the reasons for potential churning. However, companies struggle to execute these three tasks in a data-driven and integrated way. Therefore, we propose a data science and advanced analytics framework that integrates these three tasks using econometrics, machine learning, and A/B testing (i.e. randomized control trials). We argue that to take full advantage of data science, companies need to integrate both prediction and causation efforts. We provide a case study where we show how econometrics and machine learning can help design experimental studies to uncover causal effects (prediction enabling causation), and how econometric techniques can help inform and improve machine learning's feature engineering and selection tasks (quasi-causation improving prediction). We demonstrate its use and effectiveness in one of Latin America's largest insurance companies. Our framework provided the evidence that a phone call reminding clients of the benefits of their policy could reduce auto insurance churn by 6 percentage points, which after a year will represent an additional US$750,000 in revenue.