PAYMENT TECHNOLOGYPAYMENT PROCESSING
Pay by bank in 2026: what it actually costs a merchant, and what you give up
P
Payment Review Editorial Team
Payment Review Editorial Team
Every few years someone announces that merchants are about to stop paying interchange. The current version is pay by bank: the customer pays from their bank account at checkout, the money arrives without a card network in the middle, and the merchant keeps the two or three percent. Walmart has been building one with Fiserv. The pitch is not wrong. It is just far less general than it sounds.
The problem is that pay by bank is not one product. It is a label attached to at least three different mechanisms, running on different rails, priced differently, and carrying completely different dispute rules. A merchant who signs up for one expecting the economics of another will be unpleasantly surprised, usually about four months in.
Sorting them out first makes every later question answerable.
The first is old plumbing with a new front end. The second is genuinely new. The third is a commercial wrapper around either of the first two, and the wrapper is where the money goes.
Bank debit is the part where the savings are real and easy to verify. Stripe prices US ACH Direct Debit at 0.8% capped at $5, against 2.9% plus 30 cents for a domestic online card. On a $1,000 invoice that is $5 instead of $29.30. GoCardless, which does bank debit and nothing else, publishes 0.5% plus 5 cents capped at $5 on its Standard plan, rising to 0.9% plus 5 cents capped at $7 on Pro. Moov charges a flat 25 cents for next-day ACH and 40 cents same-day, with a $500 monthly minimum that tells you what size of business it is for.
Notice what the cap does. Percentage pricing on cards punishes large tickets; a per-transaction cap inverts that entirely. The $5 ceiling means a $200 payment costs the same as a $20,000 one, which is why bank debit takes hold in wholesale, tuition, rent, insurance and professional services long before it touches retail.
The instant and open-banking versions are where the pricing stops being obviously cheaper. Stripe lists its instant bank payments product at 2.6% plus 30 cents — a card-like price for a non-card rail. Trustly, among the largest pay-by-bank providers globally by its own account, publishes no US merchant rates at all; pricing comes out of an enterprise sales conversation and varies with the guarantee tier you buy, and the cheaper tiers exclude fraud losses. Dwolla, which reaches ACH, RTP and FedNow through one integration, stopped publishing prices altogether and now quotes per customer. When a rail is sold as radically cheaper and the price is not published, that is information.
In July 2025 a Federal Reserve Board researcher, Byoung Hwa Hwang, published a FEDS Note on pay by bank and the merchant payments use case. It is short, it is neutral, and it is the most useful thing written on the subject, because it takes the industry's own numbers seriously and then adds up everything the numbers leave out.
The note observes that estimates of 40 to 85 percent savings against credit card acceptance circulate widely, and concludes that a blanket claim of significant cost savings should be interpreted with care: setup fees, per-transaction fees, provider charges and bank service fees all sit on top of the interchange the merchant avoided. It also puts adoption in perspective — roughly 11 percent of US adults made an open banking payment in a year, against cards used for the large majority of retail transactions — and notes that more than half of consumers report security concerns about connecting their bank account to a third party.
None of that makes pay by bank a bad idea. It makes it a pricing question rather than a structural escape, which is the same question you would ask of any processor.
Card acceptance comes with a dispute system that is expensive and, from the merchant's side, at least legible. Bank rails do not have one. They have two very different things, depending on which rail you chose.
Under the Nacha rules an unauthorised debit to a consumer account can be returned up to 60 calendar days after the settlement date. A debit to a business account under a corporate entry carries roughly two banking days. There is no representment process to argue back with — the return is the end of it. If your customers are consumers and your product is disputable, a 60-day no-questions window is a materially worse position than a card chargeback, not a better one. The rules around ACH origination also tightened in 2026: every originator now has to monitor its own payments for fraud, which is a separate obligation worth understanding before you start pulling funds.
An RTP or FedNow credit transfer is final when it settles. There is no clearing window, no return file the next morning and no chargeback right. For the merchant that is the single best feature of the rail: the money is yours in seconds and cannot be pulled back. For the customer it means a mistaken or fraudulently induced payment has no automatic remedy, which is exactly why banks are cautious about the consumer experience and why the Fed note flags unclear liability arrangements as the main open risk. Refunds happen as a new payment in the opposite direction, on your initiative, from your balance.
Coverage is no longer the binding constraint it was two years ago. The FedNow Service lists more than 1,800 participating financial institutions. The Clearing House's RTP network set a single-day record on 1 May 2026 of 2.27 million transactions worth $8.62 billion, carries a $10 million per-transaction limit, and says it handles over 98 percent of US bank-to-bank instant payments. Walmart's pay-by-bank work with Fiserv runs over Fiserv's NOW Network, which connects to both RTP and FedNow — a proof of concept inside the Walmart app was completed and reported before the wider rollout.
What is still constrained is the consumer end. Reach depends on both sides' banks participating and on the customer being willing to authenticate into their bank at your checkout, which is more friction than a saved card and always will be. Conversion, not availability, is what decides whether the saving materialises.
The arithmetic is straightforward once you stop thinking of it as a card replacement. Bank debit wins where the ticket is large, the relationship is recurring and the customer is not shopping on impulse.
And run the comparison honestly. Take a real month of transactions, apply the bank-rail rate card and the fee schedule for failures — GoCardless charges $5 for a failed payment, which is not trivial if your customers' balances are — and set that against your actual card cost for the same month, which means your effective rate rather than your quoted one. A pay-by-bank pitch measured against a headline card rate is measured against a number you never paid.
