Balancing Risk and Profit: Predicting the Performance of Potential New Customers in the Insurance Industry

Abstract

In the financial sector, insurance companies generate large volumes of data, including policy transactions, customer interactions, and risk assessments. These historical data on established customers provide opportunities to enhance decision-making processes and offer more customized services. However, data on potential new customers are often limited, due to a lack of historical records and to legal constraints on personal data collection. Despite these limitations, accurately predicting whether a potential new customer will generate benefits (high-performance) or incur losses (low-performance) is crucial for many service companies. This study used a real-world dataset of existing car insurance customers and introduced advanced machine learning models, to predict the performance of potential new customers for whom available data are limited. We developed and evaluated approaches based on traditional binary classification models and on more advanced boosting classification models. Our computational experiments show that accurately predicting the performance of potential new customers can significantly reduce operation costs and improve the customization of services for insurance companies.