Implementing an AI-Driven Risk-Management and Fraud-Prevention System for Real-Time Application-Fraud Detection in Buy-Now-Pay-Later Lending: A Simulation-Based Evaluation

Abstract

Background. Buy-now-pay-later (BNPL) lenders approve or decline loan applications in real time, and the fraud patterns those decisions must catch change as fraud rings adapt, genuine applicant behaviour evolves, and seasonal demand shifts. Guria et al. argue that AI-driven risk management and fraud prevention in digital finance is a catalyst for economic resilience, but their contribution is conceptual and offers no lender-specific quantitative evidence of what such a system delivers relative to a conventional, non-adaptive scoring approach. Methods. This study implements an AI-driven, continuously updated, SHAP-explainable application-fraud scoring system, following the risk-management and fraud-prevention approach set out by Guria et al., and compares it against a legacy static scoring policy over five independent replications of a 52-week, 166,400-application synthetic BNPL stream containing a sudden synthetic-identity application ring, a gradual shift in genuine thin-file applicant behaviour, and a seasonal checkout surge. Results. The legacy static policy's recall at a 1% alert budget collapsed from 12.4% at baseline to 7.5% during the synthetic-identity ring and never recovered; the AI-driven system held recall at 13.8% during the same event and delivered an overall average precision of 0.240 against the legacy policy's 0.174, a 38% relative improvement, while SHAP analysis traced the ring's detection specifically to synthetic_identity_score, a feature that carried negligible weight before the ring began. Conclusion. For a BNPL lender, this evaluation provides simulation-based quantitative support for Guria et al.'s argument that an AI-driven, continuously updated, explainable risk-management approach measurably strengthens resilience to emerging fraud patterns relative to a static legacy policy, at the operational cost of weekly retraining that a lender's model-governance programme must accommodate.