Calculator

AML False-Positive Rate Tuning Estimator

Published May 10, 2026

AML false-positive rates run 75-95% at most fintech firms. Each false positive is 15-30 minutes of analyst time. Tuning thresholds to bring FP rate down — without losing true-positive detection — requires synthetic typology cases that exercise the threshold without burning analysts. This calculator estimates the cost of the current FP rate and the synthetic-corpus size to support tuning.

What you walk away with

~50s · 6 inputs
  • Annual analyst-hours consumed by false positives.
  • Hours recoverable by tuning to target FP rate.
  • Synthetic typology archetypes required to support tuning without losing detection.

Inputs

alerts
%

Default source: ACAMS / industry surveys 2024

minutes

Default source: Median across firms — initial review + close

$/hr

Default source: AML analyst salary survey 2024

%
typologies
Annual cost of current false-positive rate
$1,471,360

Direct annual analyst cost of false-positive reviews. Doesn't include burnout, attrition, or quality-degradation costs.

Cost at target FP rate
$1,086,800
Annual savings from tuning
$384,560
Analyst hours recovered per year
4,048 hours
Synthetic typology archetypes required
96 archetypes

Approximately 12 archetypes per typology (true positives + true negatives spanning the threshold) is the floor for defensible tuning. Per-typology depth varies; this is the structural minimum.

Sensitivity

Annual savings under three target-FP-rate assumptions.

  • Aggressive — target 50% FP$635,360
  • Median — target as input$384,560
  • Conservative — target 75% FP$217,360
Talk to us
Calibration source: ACAMS Compliance Report 2024 + Robert Half Salary Guide 2024 + observed AML programsDirect savings only; doesn't model burnout-driven attrition, hiring/replacement cost, or quality degradation as analysts triage burnout. Each of those would amplify the savings number.

FAQ

Won't tuning lose true positives?

Without a synthetic corpus that exercises the threshold's TP cases, yes — that's the trap. The recommended typology-archetype count provides the TP test cases needed to verify each threshold change preserves detection. The synthetic corpus is what makes the tuning defensible.

Why not just hire more analysts?

It's the more expensive answer. The annual savings from tuning typically pays for the corpus + integration in the first quarter. Hiring also doesn't scale linearly — alert quality matters more than alert volume per analyst.