Q: I set my fraud-score block at the recommended level and now I'm killing good traffic. What's the mistake?

Q: I set my fraud-score block at the recommended level and now I'm killing good traffic. What's the mistake?

A: Treating a fraud score as a hard yes/no instead of a probability. A score of, say, 75 doesn't mean 'fraud' — it means 'riskier than average for this provider's model'. Block everything above an arbitrary line and you cut honest users who happen to share risk signals.

Where people overcut:

— You block the whole high range instead of routing it. Mobile, prepaid-card, and traveling users often score high without being fraudulent.
— You use the provider's default threshold without calibrating it against YOUR own conversion data.
— You block on a single score instead of combining it with behavior (did they actually convert and stick?).

Fix: pull a month of traffic, sort by fraud score, and look at real downstream value per score band. You'll usually find the line where value drops off — and it's rarely the default. Route the gray middle to extra verification instead of an outright block.

Short version: a fraud score is a dial, not a switch. Calibrate the threshold against your own outcomes.

Still stuck? Drop your case in the comments.