Sampling transaction alerts without drowning in false positives

Pulling consecutive alerts from the top of a queue biases your sample toward the noisiest rule. Stratify instead: high-value transfers, new corridors, politically exposed persons, and a random slice of low-score noise. That mix reveals whether investigators treat serious cases differently from junk.

For each sampled case, score three things: timely acknowledgment, quality of narrative, and whether the disposition matches the evidence. A closed-as-false-positive file that still contains unexplained third-party receipts is not a clean close.

Share the scoring rubric with your monitoring team before the audit week. Surprises during fieldwork waste goodwill; a published rubric turns the review into a coaching exercise as much as a judgment.