Personalizing debt collections: Combining reinforcement learning and field experiment

Artificial intelligence (AI) brings about opportunities to revolutionize financial services. We focus on the loan debt collection context wherein collectors usually leverage private-information-based actions and follow a strict sequential collection strategy. We apply reinforcement learning to optimize the collection strategy with fine-grained data. The optimized results suggest loan platforms generally use collection actions less (by 63.39%) and more cautiously. Further borrower profiling analyses underscore the importance of personalization in debt collections. Moreover, we demonstrate the vast economic value of personalization in debt collections as it not only improves loan recovery rate (by 8.11%), but also enables platforms to allocate limited resources to cover more delinquent loans. A field experiment helps validate and quantify the economic value of the optimization algorithm in a real-world context. This study contributes to the literature of AI in FinTech, debt collections, and privacy. The findings also offer concrete, actionable, and cost-effective practical and policy implications.