Your payments run over a batch rail and the business wants to move to an instant one. What changes?
6 min readUpdated 2026-08-15
0
Answer
Short answer
Everything you did in the overnight window has to happen inside a few seconds, the payment becomes irrevocable so recall is a request rather than a right, an unanswered request leaves an unknown state you must resolve by enquiry, and liquidity plus availability become continuous obligations.
Was this answer helpful?
Likely follow-ups
→Your sanctions screening returns a possible match and the scheme timeout is about to expire. What happens to the payment?
→How do you deploy a change to this service when the rail has no maintenance window?
→What does a recall request actually oblige the receiving bank to do, and how does that differ from a card chargeback?
→How would you decide the funding level of the settlement account, and what happens overnight on a public holiday?
A batch rail gives you a window, and over the years a great deal moves into that window because it is free real estate. Fraud scoring that takes a few hundred milliseconds too long. Sanctions screening with a human adjudicating possible matches. A limit check that queries three systems. A funding decision made once, in the morning, by a treasury team looking at a file. A nightly reindex. A release at nine that quietly fixes yesterday's mapping bug before anyone outside sees it.
Instant payments remove the window, and the migration is mostly the work of finding everything that was living in it. The scheme gives you a timeout measured in seconds from instruction to a definitive answer, and within that budget you must decide, screen, check limits, reserve funds, and get a response back through the rail. Anything that cannot finish in the budget either moves earlier, moves to a different control, or stops the payment.
Moving earlier is the underrated answer. A beneficiary that was verified and screened when the customer added it does not need the same work at payment time; a limit that is maintained as a running figure does not need to be recomputed from history. The design pattern is to precompute what is stable about the parties and leave only what is genuinely per-payment inside the latency budget.
Irrevocability is the constraint, not the feature
On a batch rail, a mistake found before the cut-off is deleted from a file. On an instant rail, the credit to the beneficiary is final within seconds and there is nothing to delete. A recall is a message asking the receiving bank to co-operate, and whether the money comes back depends on the scheme's rules, the receiving bank's process, and whether the funds are still there — which, in fraud cases, they usually are not, because the whole point of speed for a fraudster is the onward hop.
That single property propagates through the design. Every validation that would previously have been caught by an operator scanning a file before release must be a hard pre-send control. Duplicate submission stops being a nuisance and becomes an unrecoverable loss, so idempotency on the outbound instruction is mandatory rather than good practice. Confirmation-of-payee style checking, where the rail or an adjacent service tells the payer whether the account name matches before they commit, exists because it is the last cheap moment to catch a mistake; several markets have made it a requirement precisely for this reason.
The state you must be able to be in
The three-outcome model — accepted, rejected, done — is wrong. There is a fourth state, and it is the one that generates every incident: you sent the instruction and did not get an answer inside the timeout. The payment may have been settled, may have been rejected, or may never have arrived.
Loading diagram…
The edge worth looking at is the gap between the lost response and the enquiry: in that interval your books and the beneficiary's books disagree, and the only safe action is to ask. Resending the instruction under a fresh reference in that state is how a single payment becomes two irreversible credits. Most schemes provide an investigation or status mechanism for exactly this, and your customer-facing state model needs a visible "we are confirming this" status rather than collapsing the unknown into either success or failure.
Liquidity and availability become continuous
Instant settlement is typically funded: the position that backs your outbound payments has to be there at the moment of settlement, and there is no netting cycle to smooth a spike. So funding stops being a morning decision and becomes a monitored, automated one, including at night, at weekends and on public holidays when the central bank's funding channel may itself be closed. The design question is what your service does when the position runs low — throttle, queue, prioritise by payment type, or reject — and that must be a stated policy, because the default of "reject whatever arrives next" is arbitrary and customer-visible.
Availability changes in the same way. A rail that runs continuously and publishes participant availability leaves you nowhere to hide a deployment, so you need changes that go out without downtime, a way to fail over that does not lose in-flight state, and a defined degraded mode. Scheme rules commonly attach availability obligations to participation, so an outage is a compliance matter and not only a customer one.
Where the fraud goes
Speed and finality do not create fraud, they change which kind wins. Card-style fraud is largely about unauthorised use of an instrument and is recoverable through the scheme. On an instant credit transfer the dominant pattern is a payment the customer authorised themselves after being deceived — an impersonated bank, a fake invoice, an intercepted conveyancing email. The instruction is genuine, the authentication is genuine, and no technical control was bypassed.
That shifts the controls towards behaviour and context rather than credentials: a first payment to a new beneficiary for an unusual amount, a device or session pattern that suggests coaching, a beneficiary account only days old on the receiving side.
It also shifts the money, because several markets have moved towards making the sending and receiving banks share reimbursement for such losses, which turns a customer's mistake into a balance-sheet exposure and justifies interventions that add friction — a warning the customer must read, a delay on first use of a beneficiary, a call-back on large amounts.
Deliberately reintroducing a small delay for a risky payment is a legitimate design and worth saying out loud, because it demonstrates you understand that "instant" is a scheme guarantee about settlement, not a promise that you must send immediately.
The message is not just faster, it is richer
Most instant rails are specified in ISO 20022, so the migration usually coincides with a move from a fixed-width or tag-based format to structured XML with defined party, purpose, remittance and identification elements. This is the part teams under-plan. Richer data is genuinely useful for screening, reconciliation and reporting, but only if it is populated and preserved end to end; a migration that maps the new format onto the old internal record throws away the reason for the change and leaves you re-solving the same reconciliation problems with a new schema. Practice varies by market on which elements are mandated and how strictly they are validated, so the profile of the scheme you are joining matters more than the standard in the abstract.
The overnight window was a design element, not dead time. Every control, decision and repair that quietly relied on it has to be relocated before the first instant payment goes out, because after that there is no file to amend and no morning to fix it in.