False positives are a systems failure, not a risk win

Your fraud system didn’t “protect revenue.”
It interrupted a legitimate user flow.

From a developer perspective, false positives are unhandled edge cases showing up at scale.

Takeaway: If valid transactions are being blocked, the system is incomplete - no matter how strong the model looks on paper.

Framework: FP as a First-Class Metric
If you treat fraud like infrastructure, you should be tracking:

  • FP rate (legit tx blocked / legit tx total)
  • Resolution latency (block → user success)
  • Retry fallout (% users who never retry)
  • Support amplification (tickets per block)

Question for the community:
Which of these do you actively monitor today—and which do you wish you had better visibility into?

1 Like

We actively monitor FP rate and resolution latency; support amplification is approximated by tagging fraud-block tickets, though tagging is inconsistent. Retry fallout is the gap—session data doesn’t reliably link a blocked attempt to later success. If I could add one view, it’d be cohort-level retry fallout by block reason, so we can see which rules cause silent abandonment rather than a quick retry. That’s often more actionable than an aggregate FP rate.