AI-Driven Business Intelligence for Real-Time Fraud Risk Monitoring in Digital Payment Platforms: A Framework-Based Review

Abstract

AI-driven business intelligence has become a critical organizational capability for identifying fraudulent activity within rapidly expanding digital payment environments. This study quantitatively evaluated the effectiveness of artificial intelligence, real-time data architecture, business-intelligence system quality, and organizational conditions in supporting fraud-risk monitoring across payment cards, online banking, mobile wallets, peer-to-peer transfers, payment gateways, electronic-commerce systems, and instant-payment networks. A systematic review and framework-based evidence synthesis were conducted using studies published between January 2010 and June 2026. Database searches identified 1,846 records, from which 72 studies were retained, 63 contributed to structured quantitative synthesis, and 49 provided comparable outcomes for random-effects meta-analysis. The included studies represented approximately 86.42 million transaction observations for the principal accuracy analysis. Pooled results showed accuracy of 96.1%, precision of 84.2%, recall of 84.9%, sensitivity of 85.1%, specificity of 97.2%, an F1 score of 84.1%, and an area-under-the-curve value of .954. Hybrid ensembles achieved the strongest descriptive performance, with an F1 score of 90.0%, while graph-based models produced the largest comparative effect for coordinated fraud-ring detection. Integrated AI-driven business intelligence increased fraud recall by 12.4 percentage points, improved pre-settlement detection by 17.6 points, reduced alert-response time by 31.8%, decreased investigation duration by 24.6%, and increased analyst productivity by 28.7%. Prevented fraud value rose by 21.3%, while false-positive rates and legitimate payment interruptions declined. AI capability demonstrated the strongest framework relationship with fraud-detection effectiveness, with a pooled correlation of .58. Class balancing, data integration, feature breadth, validation design, and methodological quality explained substantial variation in reported performance. High heterogeneity and modest publication asymmetry indicated that pooled estimates required interpretation alongside dataset origin, fraud prevalence, validation rigor, and operational context. The findings established that effective real-time fraud monitoring depended on coordinated data governance, adaptive analytics, rapid processing, intelligible reporting, skilled investigation, cybersecurity, and accountable managerial control across digital payment platforms.