What a declined card transaction actually costs, built on Visa's decline categories, the reattempt limits the network rules set, and why most published decline-cost figures do not survive inspection.

A declined card transaction costs more than the sale attached to it. The larger costs are structural. The network rules cap what you are allowed to do next, the decline reason is routinely misread by the systems downstream of it, and the part that matters most to revenue is invisible to the merchant who lost it.
Most figures quoted about decline costs have no traceable source. A few do.
PYMNTS Intelligence, in a study produced with Nuvei, surveyed 300 heads of payments and fraud at eCommerce firms with annual revenue above $100 million, in the field between 10 and 31 August 2023. Merchants with a heavy cross-border focus reported an average failed payment rate of 11.4%, against 10.1% for those selling mostly at home. Across the sample, 72% said their cross-border failure rate ran higher than their domestic one. PYMNTS put the resulting loss to US merchants at upwards of $3.8 billion in 2023.
Two things are worth noting about those numbers. They are the best available, and they are self-reported by payments executives about their own books, which is a different thing from a measurement. The gap between those two statements is the subject of the rest of this piece.
Visa sorts declined transactions into categories that determine what a merchant may legitimately do next, and those categories, rather than the merchant's appetite for retrying, set the ceiling on what a decline can still be worth.
Under Visa's rules for declined transaction resubmission, a Category 1 decline means the issuer will never approve the transaction and no reattempt is permitted at all. A Category 2 decline means the issuer cannot approve it at this time, and the transaction may be reattempted up to 15 times in 30 days. A Category 3 decline is a data quality problem, where the correct action is to fix the message rather than resend the same one.
The first real cost of a decline is therefore a classification error. Treating a Category 2 timing decline as permanent discards a transaction the issuer was willing to approve on a later attempt. Treating a Category 1 decline as temporary spends attempts, fees and customer patience on an authorization that can never succeed. Both mistakes look identical on a dashboard that counts declines as a single number.
Reattempt limits are rules, not guidance. The response codes and the merchant advice codes returned alongside a decline exist precisely so the merchant can tell which category applies, and reading them correctly is most of the job.
Mastercard's merchant advice codes are explicit about it. Braintree's reference table lists code 03 as "Do not try again" and code 21 as an instruction to stop the recurring payment entirely. A book that retries through those codes is not being persistent, it is ignoring an instruction. We set out how to read the full set in Merchant Advice Codes in Card Payments.
Some response codes carry the same flat prohibition. Response Code 14, invalid account number, must never be resubmitted with the same account number. The account number is wrong, and sending it again cannot make it right.
The part nobody can price is what the customer did next.
A merchant sees the decline, and in a subscription book, the eventual churn. What the merchant does not see is whether that same cardholder completed the same purchase somewhere else five minutes later, on the same card. No feed returns that.
This is why decline-cost statistics circulate widely and rarely survive inspection. The denominator is unobservable from inside a single merchant's data, so any figure for the true revenue lost to declines rests on an assumption rather than a measurement. That is as true of a flattering number as an alarming one.
What is observable is narrower and more useful. A decline that returned a timing-related response code, on a credential that authorizes successfully later in your own book, was an approvable transaction your systems let go. That population you can count, and it is the only one worth building around.
Some of what gets filed under decline cost is a credential problem wearing a disguise.
When a stored card has expired, been reissued, or been replaced after loss, no retry schedule will help, because the number the merchant holds no longer points anywhere. Visa Account Updater exists for exactly this, and network tokens solve it from the other direction. We worked through where each one applies in Network Tokenization vs Account Updater.
A second group is about identification rather than the credential. The same charge, on the same card, can get opposite answers depending on whether it was flagged as cardholder-initiated or merchant-initiated, which we cover in What Is a Merchant-Initiated Transaction.
Counting these as retry failures inflates the apparent cost of declines and points the fix in the wrong direction.
A single decline percentage hides the only thing that matters, which is the mix. Two merchants with identical decline rates can carry completely different exposure depending on how much of that rate is Category 2. Splitting the number is the exercise, and it starts with defining the authorization rate consistently in the first place.
Four things are worth instrumenting before anything else:
Better is a machine learning based approval layer for real-time payment recovery. It scores each decline as it happens and re-submits only the ones likely to approve, across both cardholder-initiated and merchant-initiated transactions.
That is a deliberately narrow job. It acts on the second item in the list above, the timing-related population, and leaves the rest alone. Declines the issuer will never approve are not worth an attempt, and credential problems belong to the account updater and the token, not to a retry schedule.
Directly, the margin on the lost sale plus the cost of any attempts that follow it. Structurally, the larger cost is misclassification: retrying a decline the issuer will never approve, or abandoning one it would have approved later. The network category attached to the decline is what separates the two.
Under Visa's rules, a Category 2 decline may be reattempted up to 15 times in 30 days, and a Category 1 decline may not be reattempted at all. A Response Code 14 must never be resubmitted with the same account number, and Mastercard's merchant advice code 03 carries the same instruction. The code returned with the decline is what tells you which one you are holding.
Because the key number is unobservable. No merchant can see whether a declined cardholder completed the purchase elsewhere, so any figure for total revenue lost to declines rests on an assumption. Survey figures from payments executives, such as the PYMNTS and Nuvei study above, are the most traceable numbers available, and they are still self-reported. The category mix inside your own data is the more reliable thing to work from.
Not on its own. A book with a high rate concentrated in Category 1 declines has little available to it, because those transactions were never going to be approved. A book with a lower rate but a heavy Category 2 share is leaving more on the table. The mix matters more than the headline number.