Screening and monitoring.
WHERE IT HELPS
Transaction monitoring at volume Screening against sanctions and watch lists Identifying unusual patterns Summarising documentation for review Drafting reports
THE ALERT VOLUME PROBLEM
Traditional rule-based monitoring produces enormous numbers of false alerts.
Investigators spend most of their time clearing alerts that were never suspicious.
AI can reduce that volume, which is its main value here.
WHAT MUST NOT CHANGE
A trained person decides whether something is suspicious and whether to report.
That judgement is regulated and cannot be automated away.
WHAT TO BE CAREFUL WITH
Reducing alerts by learning from past clearances, which teaches the system to miss what was previously missed.
WHAT REGULATORS EXPECT
Understanding of how the system works Documentation of tuning decisions Evidence that reductions did not reduce effectiveness Human decision-making on reports
FOR NAME SCREENING
Matching across scripts, transliterations and spellings is a genuine strength.
Verify matches. False matches on common names are frequent.
