How would you tune an AML transaction-monitoring system that is generating an unmanageable volume of false-positive alerts?

First, don't simply raise thresholds to cut volume — that risks missing genuine suspicious activity and is a regulatory red flag. I'd analyze alert outcomes: which rules/scenarios produce alerts that almost never convert to SARs, and segment by customer risk so thresholds reflect expected behavior (a cash-intensive business differs from a salaried retail customer). Then tune scenario parameters with data, document the rationale and get model-governance sign-off, and run the change in parallel/back-test against historical SARs to confirm you're not suppressing true positives. Address the backlog with risk-based triage and consider a lookback if a gap existed. The principle: optimize precision without lowering true-positive coverage, all evidenced and governed.