The Payout Delay That Cost a Payments Firm 4,000 Merchants
Archi3 min read
Contents
The decision behind a one-day payout lag#
Payments companies tend to talk about uptime and fraud. Less often they talk about settlement timing, the gap between when a customer pays and when a merchant can touch the money. Priya Raghavan runs payment operations at a mid-sized processor that handles card and instant bank transfers for about 40,000 small merchants. Last year the company chose to move its settlement cutoff from an evening batch to an intraday cycle. The change went badly for two weeks. I asked her how the decision got made and what it cost.
Q: What forced the change?#
Raghavan: Merchant complaints, mostly from restaurants and small retailers. Our cutoff was 6 p.m. local time. Any transaction authorized after that landed in the next day's batch, which meant a Friday night dinner rush settled on Monday. Merchants saw the sales in their dashboard but not the cash. Some of them ran payroll and supplier payments off that number and got caught short.
We measured it. About 11 percent of weekly volume fell into that late window. For a merchant doing $30,000 a week, that is roughly $3,300 sitting an extra two or three days. Not fatal, but annoying enough that it showed up in churn conversations.
Q: What were the options?#
Raghavan: Three. Keep the batch and just move the cutoff later, say 11 p.m. Cheaper, but it fixes nothing on weekends. Add a second evening batch around 10 p.m. on weekdays. Or go to continuous settlement, running payouts several times a day against whatever cleared the card networks.
The engineering team priced each one. The second batch was maybe six weeks of work. Continuous settlement touched our ledger, our reconciliation with two acquiring banks, and the fraud hold logic. Call it five months and a real risk of double payouts during the transition.
Q: Who made the call?#
Raghavan: The CEO, after a meeting where finance and risk disagreed loudly. Finance wanted continuous settlement because it matched what larger competitors advertised. Risk wanted the later cutoff because it kept the hold window intact. I argued for the second batch. I lost. In hindsight the CEO was probably right on direction and wrong on timing.
Q: What broke?#
Raghavan: The hold logic. Our fraud rules held certain transactions for review, usually first-time buyers or unusual ticket sizes. Under the batch model, a held transaction simply missed that batch and got picked up the next day. Under continuous settlement, a held transaction could clear the hold after one payout run had already been calculated, and then get paid in the next run. That is fine. But our reconciliation report counted it twice on the day of release.
Accounting caught it on day three. About 4,000 merchants saw a payout line that did not match their sales total, and roughly 600 of them were paid twice for some transactions. We clawed back about $180,000 over the following week. Merchants hated that more than the original delay.
Q: What would you do differently?#
Raghavan: Run both systems in parallel for a month and compare totals daily, rather than flipping a switch. We did a two-week parallel run. It was not long enough to catch a rule that only fires on refunds, which is what tripped us. Refunds reversed against a settled payout are rare but not that rare. We also should have frozen fraud rule changes during the cutover. One analyst tightened a threshold mid-migration and nobody noticed for two days.
Q: What does the settlement schedule look like now?#
Raghavan: Four payout runs a day on weekdays, two on Saturday, one on Sunday. Average time from authorization to merchant availability is about nine hours, down from roughly 34. Chargebacks went up slightly, under a tenth of a percentage point, because merchants get money before disputes can land. We absorbed that. Support tickets about payout timing dropped by about 70 percent.
Q: Any advice for a team facing the same choice?#
The hard part is not the schedule. It is the reconciliation logic that assumed one payout per day per merchant. Map every report, alert, and downstream export that touches settlement before you change the cadence. We found four we had forgotten, including a partner feed that only ran nightly. That one sat stale for a week and nobody flagged it because nobody owned it.
Also, write down who can change a fraud rule during a migration. Ours was nobody, in theory. In practice it was whoever had the admin login.
More articles