Returning customers were failing their first payment after a card replacement, then churning. The fix lived entirely in the details.
What I walked into was a quiet leak: returning customers were failing their first payment after a card got replaced, then dropping out of the basket. The stored tokens didn't update when an issuer reissued a card, so a perfectly valid customer looked invalid at exactly the wrong moment.
The fix was moving from PSP tokens to network tokens (Visa's VTS, Mastercard's MDES), and the whole job lived in the details. A card gets replaced, the token can survive, but only if our vault stays in sync with the network's lifecycle stream, and only for issuers who participate. I mapped that with the card networks and our acquirer, then worked with engineering so lifecycle events flowed cleanly from network to acquirer to vault, with idempotency so a repeated event couldn't corrupt a token.
We migrated brand by brand, least-active cohort first, with a rollback path at every step and a weekly readout tracking authorisation rate by token type, by issuer, by market. A 0.3% slip on lifecycle handling would have damaged the exact metric we were trying to move. There was no room to approximate.
Credential continuity is one of the most under-recognised drivers of merchant economics. Getting the boring plumbing exactly right is what turns it into revenue.