Designing Fraud Detection and Recovery Systems: Patterns and Trade-offs

This article, the second in a two-part series on payments-systems engineering, takes a systems-design view of fraud detection and recovery. It covers the core detection techniques (rules, machine-learning models, and anomaly detection), the layered responsibility across card network, issuer processor, and issuer, and the central trade-off between false positives and false negatives - including action bands and champion-challenger threshold tuning. It presents a practitioner's Detect-Decide-Recover-Learn (DDRL) loop, along with human-in-the-loop recovery workflows, real-time cardholder verification, case correlation, and investigation and recovery case modeling. The article is vendor-neutral and based solely on publicly available knowledge. A companion article covers credit card dispute resolution.