A top-five commercial bank faced recurring incidents of current-account mule fraud. Accounts that appeared legitimate at onboarding were later used to channel illicit funds. Many mule accounts showed clean or limited bureau histories, normal early behavior, and transaction patterns deliberately engineered to avoid suspicion. As fraud tactics adapted, manual screening and reactive investigations became increasingly strained.
Mule fraud is difficult precisely because it mimics legitimacy. The challenge was not finding more anomalies, but identifying structural signals that fraud networks cannot sustain at scale. Decisions needed to surface risk early, control false positives, and reduce operational load, while earning shared confidence across fraud, risk, audit, and investigations teams.
Dhurin designed a layered detection framework built around two complementary lenses. One examined onboarding behavior, bureau attributes, and aggregated transaction patterns to surface inconsistencies invisible in isolated views. The other analyzed phone, device, and behavioral connectivity to detect coordination patterns that fraud rings depend on but struggle to conceal. Both streams fed a prioritized reject-and-review framework embedded directly into screening workflows, supported by lifecycle validation and drift monitoring.
The bank achieved approximately 12% decision-stage rejects, capturing a significant share of high-probability mule accounts, while reducing manual screening effort by around 50%. Fraud teams moved from reactive investigation to earlier, defensible intervention, supported by an auditable system designed to adapt as adversarial behavior evolved.
