Subscription product in the EU. First payment went through 3DS, the card was tokenized, and every renewal after that was a plain POST to the charge endpoint with the token and an amount. It worked, so nobody touched it.
Then renewal approval sat at 71% while first payments sat at 90%. The decline codes were not 54 or 14. They were soft declines asking for authentication (Visa 1A), on cards that were fine.
The cause: from the issuer's side, our renewals were ordinary e-commerce transactions with no cardholder present and no 3DS. Nothing told them this was a merchant-initiated charge under an agreement the customer had already authenticated. We never sent a stored-credential indicator, and we never referenced the network transaction ID from the first payment.
The fix was three fields, not a rewrite:
- Mark the first payment as the customer-initiated one that stores the credential.
- Save the network transaction ID from its response, next to the token.
- Mark every renewal as merchant-initiated recurring and send that saved ID with it.
Renewal approval went from 71% to 88% in two billing cycles. We changed nothing else.
The annoying part is that the sandbox never complained. It approves everything with a token.
Does your gateway store the original transaction ID for you, or do you have to keep it yourself? We had to keep it ourselves, and I'd like to know how common that is.
Payneteasy handles this kind of payment flow for merchants and PSPs — see how it works.
Top comments (1)
The sandbox line is the finding here and it is the part most readers will skim. An environment that approves every tokenised charge is not a test of the thing you were about to get wrong. It tests that your request parses.
On your question I should be straight about what I do not run. We are open banking rather than cards, so I have never had to keep a network transaction ID, and I cannot tell you how common it is for gateways to hold it.
What I can offer is the structural difference, because it explains why the field is losable at all. For a card MIT the agreement is asserted per transaction out of fields you carried yourself. For a bank mandate the consent sits at the bank, the reference comes back from the bank, and a renewal cites a consent the bank already holds. You cannot lose it because you were never its custodian.
That is not an argument for switching rails. It is why I do not think 71 to 88 was really a gateway feature question. Any identifier that you hold, that the authorising party also holds, and that nobody reconciles, drifts eventually. Two custodians, one agreement, no check between them.
The thing I would add after the fix is a scheduled comparison of your stored IDs against what came back on the most recent authorisation for each token, rather than only finding out at charge time. Approval rate is a slow way to learn the chain broke.