Increasing Retention in Insurtechs Through Churn Prediction

Abstract: Over the last decades, the Swedish insurance industry has seen decreased entry barriers due to deregulation and emerging new technologies, which have the potential to disturb the stagnated and consolidated competitive landscape of the industry. Initiated by newcomers like American insurance startup Lemonade, and later Swedish Hedvig among others, there is an increased push toward digitalization, transparency, and automation in the industry. This thesis examines how Insurtechs can increase retention by identifying customers at-risk of churning, as well as what actions they can take in order to make customers more likely to stay, with the digital insurance company Hedvig as a case study. Various machine learning methods for predicting churn are examined in a literature review, and a model is developed and proposed for Hedvig. Seven levers for increasing retention, 1) Understanding Churn, 2) Customer Intake, 3) Product Improvement, 4) Lock-in, 5) Targeting at-risk Churners, 6) Save Desk, and 7) Organizational Setup, are identified and presented with documented best practices from expert interviews. The conclusion is that churn could not be predicted accurately as the proposed model, a Gradient Boosted Tree model, achieved an ROC value of 62%, which is considered low, and an unsatisfactory precision and recall curve. In the discussion section, we propose that the reason behind this is that there is that there is not enough signal in the data, that the two classes are very homogeneous. In order to improve the predictive accuracy, more usage data from the customers, that have a stronger correlation with the outcome variable, churn, should be collected. Besides predicting churn, the thesis discusses some alternative ways to increase retention, based on discussions with industry professionals, and presents some company specific recommendations in the discussion chapter.