Optimization of Long-Term Profit when Cross-Selling Insurance Policies

Insurance companies try to maximize profits from their existing customers by cross-selling other policies to them. These recommended policies may either be chosen by agents, who may try to maximize commissions, or the company who wants to maximize profit. However, in either case, these policies may not be the most suitable for the client and they may end up either taking the policy but then cancelling or, if it is obviously unsuitable, the customer may cancel all policies and switch to a competitor. In this paper, we aim at reducing these risks of cancellation and dissatisfaction while achieving an acceptable profit. To achieve this, we propose and apply benefit-based decision trees to predict customer satisfaction and their probability of cancellation if offered a given policy. These trees are trained with the goal of maximizing long term profits. Long term profits are achieved when the client purchases a profitable policy but does not cancel due to dissatisfaction. An overall score is calculated using suitability predictions, cancellation probability and long term benefits of the policies. The policy with the highest score is recommended to the client. We demonstrate the benefit of our approach using data from an insurance company.