Reducing False Positives in Financial Crime Monitoring
A challenger bank's transaction monitoring system was generating tens of thousands of low-quality alerts per month, overwhelming the financial crime team.
Headline outcome
70% Fewer alerts to review
The challenge
Investigators were drowning in false positives generated by rules-based monitoring, while genuinely suspicious activity sometimes slipped through unnoticed.
“Deployed an ML-based alert triage layer that scores legacy alerts and surfaces only the highest-risk cases to investigators.”
Our approach
-
01
Built supervised ML models on historical alert outcomes to score new alerts.
-
02
Retained the rules layer for regulatory comfort; ML acted as a triage layer on top.
-
03
Implemented full model risk governance and ongoing monitoring of detection rates.
The result
The financial crime team now focuses on the highest-risk activity, throughput more than doubled, and detection rates improved without breaching regulatory expectations.
70%
Fewer alerts to review
2.5x
More true positives caught
£1.4M
Annual savings
Be our next success story