When a good customer looks like fraud

The fraud stack is supposed to stop stolen cards. It also flags legitimate customers. When that happens, the default is to decline. You lose the order.

The work is dynamic friction. If a checkout looks suspicious, step the cardholder up with 3DS instead of killing the order. If the challenge succeeds and the network shifts liability, that is how the order that would have been lost comes back. $59.6M GMV recovered in seven months. 65.6% of those flagged orders authorized.

I do not take credit for the liability shift itself. That is an issuer and network outcome. I take credit for not declining the customer in the first place, and for the orders that came back.

Email OTP when we need a challenge that is not bound to a single processor. That layer recovered $64.6K in the first two weeks. 91.5% end-to-end completion. 99.1% submit-to-verify. 57% of issued challenges converted.

Two processors run the authorization. A retry is not a second charge. A second challenge completion is not a second order. An unresolved payment does not get a second dispatch.

$59.6M is GMV of card orders the fraud stack would have declined as false positives, subsequently authorized, over seven months. 65.6% of those flagged orders authorized. It is not profit. It is not a claim that every recovered order is a perfect incremental counterfactual. It is the money that was about to be a decline and became an authorization.

I still watch abandon on challenge, issuer timeout, soft declines after a passed challenge, and chargebacks on the recovered cohort versus baseline. Recovery that manufactures fraud is not recovery.