A Business-Oriented Recommender System for Insurance Policy and Add-On Upselling Using Implicit Behavioral Data

Abstract

Recommender systems are now important for making personalised product offers, but using them in the insurance industry is not easy. This study aims to develop a recommender engine suitable for real-world businesses. It will focus on recommending insurance policies and additional coverages to support effective upselling. We look at and compare several recommendation methods based on how well they predict, how hard they are to model, and how much work it takes to tune them to find a solution that is both balanced and scalable. Because there are no clear customer ratings and insurance data is highly sensitive, the proposed system relies on implicit behavioural signals and next-item prediction techniques. Some of the other problems solved were a lack of item diversity, a lack of interaction data, and the difficulty of using customer metadata due to concerns about randomness and bias. The study shows how to address implicit feedback while reducing bias from favouring popular products over less popular ones. The results show that carefully chosen recommendation strategies can deliver real business value while adhering to the rules and limitations that apply to data in the insurance industry.