What is anti-money laundering?
Anti-money laundering (AML) is the body of law and the set of internal controls that require banks, credit unions, money services businesses, broker-dealers, and other covered firms to detect and report attempts to move criminal proceeds through the financial system. The core obligations are customer due diligence, transaction monitoring, investigation of suspicious activity, and filing of suspicious activity reports with the national financial intelligence unit. Know your customer (KYC) checks at onboarding are the front end of the same program.
Transaction monitoring is where AI has changed the work. Traditional systems apply fixed rules, such as flagging cash deposits just under a reporting threshold or rapid movement of funds through a new account. Machine learning models add behavioral scoring that compares each customer’s activity against their own history and against peers, and network analysis that finds rings of accounts moving money in circles.
Why AML matters
Regulators fine institutions heavily for weak AML programs, and the fines are public. The operational problem is volume. Rule-based systems generate alerts of which only a small fraction turn out to be reportable, and every alert must be reviewed by an investigator. Models that rank alerts by likelihood of being genuine let a compliance team clear the queue faster and spend time on the cases that matter. Models that find patterns rules miss, such as structuring spread across many accounts, catch activity that would otherwise go unreported.
AML in practice
A community bank with a small compliance team was spending most of its hours on alerts that closed as false positives. The bank added a scoring model trained on three years of alert outcomes. The model ranks each new alert, attaches the features that drove the score, and drafts the investigation summary from the account history. Investigators work from the top of the list down, and every alert still receives a human decision. The model’s reasoning is retained with each case file for examiners.
What to watch for
- Regulators expect an explanation for every model-driven decision, so prefer approaches that can state which behaviors raised a score.
- Never let a model close alerts on its own. Suppression of alerts without review is a finding examiners look for.
- Validate the model against known cases and retrain as typologies change, since launderers adapt to whatever the system catches.
- Document the model under your model risk management policy from the first version.