Arm’s Length Lending for the Thin-Filed Using Artificial Intelligence

Abstract

This article reviews existing techniques and proposes new methods to evaluate credit risk in economies without formal credit systems but widespread mobile phone usage. Limited availability of formal financial data remains the primary drawback to credit access, yet AI-based analysis of digital behavior can help bridge this gap. Our review suggests that applying machine learning methods to classify behaviors such as gambling and alcohol-related spending as high risk may accurately capture credit risk, while expenditure on education, including school fee payments, may signal creditworthiness. Further, stable and long-term location patterns may serve as strong indicators of high credit quality. We propose using AI to discover hidden “non-linear” patterns, such as a combination of alcohol consumption and irregular phone charging, which may predict high credit risk but possibly escape human analysts. The proposed systems automate the search for these factors and create models that are robust to sparse data. Finally, we analyze the “Digital Utility Trap”, where the fear of losing essential mobile access motivates borrowers to pay, offering a safe and scalable path for lending to the unbanked and thin-filed.