Enhancing customer-centricity in financial institutions through open banking : a machine learning approach to potential customer lifetime value

Abstract

The financial sector is undergoing significant transformation, driven by digitalization, regulatory changes like Open Banking (OB), and heightened competition, making customer-centric strategies more crucial than ever. This dissertation comprises three interconnected studies, all focused on enhancing these customer-centric strategies. The first study presents a predictive framework for customer churn, achieving high performance with a PR-AUC of 0.95, providing actionable insights for retention strategies. The second study builds on and improves the first by incorporating an additional layer of exploratory model analysis, allowing for a deeper interpretation of OB customer behavior. It develops two predictive models: one for inflow (data received from competitor banks) and another for outflow (data shared with competitor banks), achieving strong predictive performance with PR-AUC scores of 0.91. These models help explain the factors influencing customer data-sharing behavior, offering opportunities for targeted marketing to attract customers from competitors while addressing risks associated with potential losses in retention, products, and services. The third study explores Potential Customer Lifetime Value (PCLV), identifying 4.23% of customers who transmitted OB data as having a high potential for profitability growth. Although the dataset for this study depends on inflow data from OB, only a small fraction of customers currently engage in this process, emphasizing the importance of the second study in identifying customers most likely to utilize OB. The PCLV calculations also integrate retention probabilities from the first study's churn model, enhancing predictions of potential profitability and enabling the estimation of Total CLV, representing the total estimated profitability a customer holds within the market. This comprehensive framework, leveraging machine learning (ML) and OB data, was conducted in one of the largest banks in Brazil, with over 3 million customers, providing a robust approach to improving churn prediction, understanding customer data-sharing behavior, and refining CLV estimation.