US Payment Rails Compared

A rail is a risk decision before it is a speed decision

Businesses usually choose a payment rail by asking how fast the money arrives and what the transfer costs. Those are the two least interesting questions. The one that determines whether a payment operation survives contact with fraud, disputes and reconciliation is different: once this payment has been made, who can undo it, on what grounds, and for how long?

The five divide cleanly on that question, and the division does not track speed. ACH is slow and highly reversible. Wires are fast and effectively final. RTP and FedNow are instant and final. Cards are fast to authorize, slow to settle, and reversible for months. A business that picks a rail on speed alone will eventually pick an irrevocable rail for a payment it needed to claw back, or a returnable rail for revenue it treated as banked.

One structural point applies to every bank rail: a business does not connect to ACH, RTP, FedNow or Fedwire directly. Access runs through a depository institution — an ODFI (Originating Depository Financial Institution) in ACH terms — or a provider that has arranged that access on its behalf. Dwolla states plainly that all funds transfers made using its platform are performed by a financial institution partner. The rail's rules bind the business through that intermediary, exactly as card network rules bind a merchant through its acquirer.

The five rails at a glance

Everything in this table is structural. Dollar limits, operating windows and fee levels move — Nacha, The Clearing House and the Federal Reserve all revise theirs — so treat any specific figure as something to confirm rather than rely on.

ACH: cheap, batched, and returnable

ACH (the Automated Clearing House network) is the workhorse of US business payments — payroll, supplier payments, insurance premiums, subscription collection. Two clearing operators run it: the Federal Reserve's FedACH, and the Electronic Payments Network operated by The Clearing House, which has run a private-sector ACH network since 1974 and says it handles roughly half of US commercial ACH volume. Both work under one rulebook, the Nacha Operating Rules.

ACH is a batch system. Files are exchanged at scheduled times, not continuously, and only on banking days. Same Day ACH added intraday settlement windows, letting a properly submitted entry settle on the day it was originated, subject to a per-transaction dollar cap that Nacha has raised over time. Same Day ACH does not make ACH real time; it makes it same business day, which is a different product.

The property that makes ACH so useful is that its cost is a flat amount per transaction, effectively unrelated to the amount transferred. A payment of $80 and a payment of $80,000 cost the originator roughly the same. That is why large-value B2B invoicing gravitates to ACH and away from cards.

ACH is the only mainstream US rail that can pull

ACH supports both credits (push) and debits (pull). A business with a valid authorization can debit a customer's account on a schedule without the customer acting each cycle. Every recurring bank-collection business depends on this: GoCardless built a company on mandate-based bank debit across ACH in the US, Bacs in the UK, SEPA in Europe, BECS in Australia and PAD in Canada. No US instant rail can do this, and that single fact determines where a recurring-billing business can move.

Returns are the defining risk

An ACH debit that settles is not final. The receiving bank can return it, and return codes fall into three groups: administrative returns for bad account data, insufficient-funds returns, and unauthorized returns. The windows differ sharply by account type. A business account generally has two banking days to return an entry as unauthorized. A consumer account gets far longer — Nacha provides a window measured in weeks after settlement for an unauthorized consumer debit, and Regulation E gives consumers separate error-resolution rights against their own bank on top. Confirm the current periods against the Nacha Operating Rules rather than a remembered number.

Nacha also polices originators on return performance, setting ceilings on unauthorized, administrative and overall return rates. An originator that breaches them triggers a mandatory inquiry through its ODFI and can face fines or termination of origination privileges. This is the ACH analogue of a card chargeback monitoring program, and it is why ODFIs and third-party senders underwrite ACH originators rather than simply switching them on. Nacha separately requires originators of internet-initiated consumer debits to validate the receiving account before the first debit, which is why account-verification products became standard rather than optional.

RTP and FedNow: instant, credit-push, and final

The United States now has two instant payment rails. The RTP network, operated by The Clearing House, launched in November 2017 and settles through a joint account at the Federal Reserve Bank of New York; any federally insured depository institution can join without owning a stake. The FedNow Service, the Federal Reserve's own instant rail, launched in July 2023. They are separate networks with separate participant lists, and a payment can travel over one only if the receiving institution participates in that one.

Structurally they are near-twins, and both differ from ACH in the same three ways.

• They run continuously. 24 hours, 7 days, including weekends and federal holidays. There is no cut-off time and no banking-day concept.
• They are credit push only. The payer's institution sends the money. There is no debit pull and no equivalent of a direct debit mandate. Both support a Request for Payment message, which asks the payer to send rather than moving funds — the payer still has to act.
• They are irrevocable. Once the receiving institution accepts the payment it is final: no return code for buyer's remorse, no dispute process, no chargeback. Both provide a message by which a sender can request the return of funds sent in error, but that is a request the receiving institution and its customer are free to decline.

Irrevocability is the single most important practical difference between ACH and the instant rails, and it is the one most business cases skip. It inverts where fraud loss lands. On ACH a bad debit can be returned and a disputed collection unwound, so the recovery mechanism absorbs some of the cost of imperfect screening. On RTP and FedNow, an authorized push payment induced by fraud is simply gone. Every control has to happen before the send — beneficiary verification, dual approval, velocity limits, hold periods on new payees — because there is nothing after it.

That property is also why these rails are valuable. A payout that cannot be reversed is one the recipient can spend immediately and the payer can reconcile with certainty, and both networks carry richer remittance data than ACH. Providers building on them treat instant rails as the destination: Modern Treasury reported that the large majority of its payment service provider volume in mid-2026 moved over real-time rails, and Moov, Dwolla and Melio all expose RTP or FedNow alongside ACH. The binding constraint is reach, not technology — a business can send over either network only if the counterparty banks somewhere that participates, so any instant payout product needs an ACH fallback and a rule for when it fires.

Wires: the rail for large value and the rail with no safety net

A domestic wire moves through Fedwire, the Federal Reserve's real-time gross settlement system, or through CHIPS, the private-sector large-value system operated by The Clearing House. The difference matters more to banks than to businesses: Fedwire settles each payment individually in central bank money, while CHIPS nets payments through an algorithm and settles net positions, which is how it clears around $2.2 trillion each business day across roughly 43 participating institutions. CHIPS is designated a systemically important financial market utility and supervised accordingly.

For a business, wires have three characteristics. They are expensive relative to every other rail, charged as a flat fee usually levied on both sides. They operate on banking days within a defined window rather than continuously — the Federal Reserve has approved an expansion of Fedwire's operating hours, so the current schedule is worth checking rather than assuming. And they are final.

Wire finality is not a technicality. It is why business email compromise fraud targets wires specifically. When a company is deceived into wiring funds to a fraudulent account there is no chargeback, no return code, and no rulebook obligation on the receiving bank to give the money back; recovery depends on speed, law enforcement, and the funds still being there. Any business making regular large payments should treat callback verification on changed bank details as a hard control rather than a policy document. It is the only control that exists.

Cards: the expensive rail you use for the dispute rights

The card networks are structurally different from every other rail here. A card transaction is a pull, initiated by the merchant against the cardholder's account under a rulebook the merchant never signed. Its cost is a percentage of transaction value plus a fixed amount, split between interchange (which goes to the card-issuing bank and is set by the network), network assessments, and the acquirer's markup. Cards are the only US rail where doubling the transaction size roughly doubles the fee.

What that percentage buys is not speed — authorization is instant, funding usually is not. It buys distribution and a consumer guarantee. The cardholder can dispute a transaction with the issuing bank for a period typically measured in months from the transaction or expected delivery date, with the merchant bearing the reversal and a fee unless it wins a representment. That right is a large part of why consumers will hand card details to a business they have never dealt with.

The consequence is a clean rule of thumb: cards are the right rail where the counterparty is a stranger and the ticket is small, and the wrong rail where the counterparty is a known trading partner and the ticket is large. A supplier invoicing $60,000 to a customer it has served for a decade gains nothing from the consumer dispute apparatus and pays a percentage for it.

What happens when something goes wrong

The failure modes are the part of a rails comparison worth memorizing.

• Wrong account, ACH credit. The entry may be returned if the account cannot be posted. If it posts to a real account belonging to someone else, the originator can transmit a reversal within a short window for a genuine error, but has no right to the funds. Usually recoverable, sometimes not.
• Wrong account, instant payment or wire. If the receiving institution accepts it, it is final. The sender can ask for a return; recovery depends entirely on the goodwill of a stranger and the speed of the bank's fraud team.
• Customer claims they never authorized it, ACH debit. Returned as unauthorized, within a window that is short for business accounts and long for consumer accounts. The originator loses the funds and the return counts toward its Nacha return-rate performance.
• Customer claims they never authorized it, card. A chargeback. Funds are debited, a fee is charged, and the merchant can contest it by representment on a defined timetable — with network monitoring programs waiting if the rate climbs.
• Customer received the goods and disputes anyway. Cards have a reason code for this and a process. Instant rails and wires have nothing: the money moved, and the dispute becomes a commercial or legal matter between the parties. ACH sits in between, depending on entry type and whether the account is consumer or business.
• Payment falls on a bank holiday. ACH and wires do not run and the payment queues. RTP and FedNow do run. This is a real operational difference for payroll and weekend payouts.

Choosing, and the mistakes worth avoiding

A workable default for a US business: collect recurring payments by ACH debit where the relationship supports a mandate, invoice large B2B amounts by ACH credit or wire, pay out on instant rails where the recipient values speed and the amount is verified, and accept cards where the counterparty is a consumer or an unknown business and convenience is the point.

The mistakes that recur:

• Treating settled ACH as cleared funds. Money in the account is not money kept. A business that ships the moment an ACH debit settles has ignored the return window entirely, and consumer entries carry the longest one. Fulfillment policy should be set against the return window, not the settlement date.
• Assuming instant rails can collect. RTP and FedNow are credit push only, so a subscription business cannot debit customers over them. Request for Payment shifts the action to the payer, which is not the same product as a mandate, and modeling it as one produces a badly wrong collection forecast.
• Assuming instant means instant reach. Both networks depend on the receiving institution participating. Coverage has grown steadily, but "instant payouts" as a product promise requires an ACH fallback and a clear trigger for it.
• Using wires for anything that might need to be undone. If there is a plausible scenario in which the payment should not have been made, a wire is the wrong instrument regardless of how convenient the speed is.
• Paying card percentages on large trusted invoices out of habit. The percentage on a large B2B invoice buys dispute rights neither party needs. Moving that volume to bank rails is often the single largest available reduction in a company's payment costs.
• Ignoring who holds the rail access. When ACH origination runs through a third-party sender or a payments API, the Nacha rules still apply, the ODFI still underwrites, and the return-rate performance still belongs to somebody. Ask which entity is the originator of record, and what happens to the account if return rates rise.

Frequently asked questions