
The institution's fraud and AML systems generated hundreds of alerts a day; up to 95% were false positives the team reviewed manually, while real cases could slip through.
60–80% fewer false positives while keeping detection capability, with alerts prioritized by real risk.