When the naira redesign pushed an entire economy onto electronic rails overnight, the hardest engineering problem was not keeping systems up. It was keeping them honest. Senior engineer Al-Amin Ajao worked on that problem from the client side of a cloud banking platform.
By Al-Amin Ajao Olayiwola
In the last week of February 2023, a failed bank transfer was no longer the worst outcome for a Nigerian customer. The worst outcome was a transfer that failed ambiguously: a debit alert on one phone, silence on the other, a spinning wheel where a confirmation should be — and no physical cash anywhere in the country to fall back on. Multiply that moment by hundreds of millions of transactions, and you have the defining infrastructure story of Nigeria’s cashless crunch.
The story usually gets told from the queue outside the bank. This one is told from inside the systems — where a small number of engineers spent those months answering a question most payment platforms had been able to avoid at scale: when a transfer times out, did the money move or not?
A policy shock becomes a load test
The sequence that produced the crunch moved quickly. In late October 2022, the Central Bank of Nigeria was authorised to redesign the ₦200, ₦500 and ₦1,000 banknotes, with the new notes unveiled the following month and January 31, 2023 set as the deadline for the old ones to stop being legal tender. The new notes did not arrive in anything like sufficient volume.
The deadline slipped to February 10 amid what the presidency called systemic and human difficulties; on February 8, the Supreme Court restrained the government from enforcing that date after three states sued; and on March 3, 2023, the court ruled that the redesign had been implemented unconstitutionally and ordered the old notes to remain legal tender until the end of the year.
By then, the damage to daily commerce had already been done. The central bank’s own figures showed cash outside the banking system collapsing from roughly ₦2.57 trillion before the policy to about ₦788 billion by January 2023. For an economy where most retail transactions were still settled in physical cash — market trade, transport fares, the vast informal sector — this was not a nudge toward digital payments. It was a forced migration, compressed into weeks, in the run-up to a general election.
The migration showed up immediately in the data. NIBSS Instant Payment, the account-to-account rail that underpins Nigerian bank transfers, processed 541 million transactions in January 2023, up 55 percent from the same month a year earlier.
February volumes rose 121 percent year-on-year even as the value of transactions dipped — a signature of millions of first-time and small-ticket users pushed onto the rails, splitting daily life into ₦500 and ₦2,000 transfers.
March 2023 set an all-time record of 1.1 billion instant payment transactions in a single month, with electronic transaction value surging to ₦48.3 trillion, a spike NIBSS itself attributed directly to the cash scarcity. Across the full year, transaction volume on the platform nearly doubled, from 5.1 billion in 2022 to 9.7 billion.
No Nigerian bank had provisioned for that curve, because no reasonable capacity plan assumes the country’s cash supply will be switched off in a quarter.
What breaking looked like
The consumer-facing failures of February and March 2023 are well documented. Banking apps froze or refused to load; some institutions quietly disabled their mobile applications for days at a stretch to prevent total collapse under the traffic. USSD channels — the fallback for customers without smartphones — buckled under the same surge, aggravated by a long-running billing dispute in which banks’ debts to telecom operators on USSD sessions swelled past ₦100 billion. Point-of-sale agents, suddenly the only reliable source of physical cash, charged premiums that made a ₦10,000 withdrawal a considered purchase.
The deeper failure was what happened to transactions that entered the system and never cleanly left it. Reporting through the period described customers debited at ATMs that displayed dispense errors and never paid out, transfers that left one account and reached no other, and reversals that took weeks.
By March 2023, industry figures indicated that only around 60 percent of failed electronic payment transactions from the crisis months had been resolved — meaning roughly four in ten complaints, many involving money trapped mid-flight, remained open while the customers involved had no cash alternative.
KPMG’s banking industry survey for the year recorded the crunch as a period of significant downtime and deteriorating service levels, and raised the question of lasting damage to trust in the sector. The market answered part of that question on its own: mobile money operators whose infrastructure held up through the surge, notably OPay and PalmPay, became the most downloaded financial applications in the country by March and converted the crisis into durable market share at the banks’ expense.
The anatomy of a retry storm
A transfer over an instant payment rail is not one atomic event. The sending institution submits an instruction, the switch routes it, the receiving institution must accept and credit it, and a response code travels back along the chain. Under normal load, that round trip completes in seconds and the response is unambiguous.
Under the load of February 2023, every hop in the chain slowed, and the initiating system was left facing distributed computing’s oldest hard question: a request that times out has not failed — it has become unknowable. The money may have moved. It may not have. The timeout itself carries no information either way.
Systems that answer that question naively fail in one of two directions. Treat the timeout as a failure and retry, and you risk debiting a customer twice for a transfer that actually succeeded downstream. If you treat it as a success and do nothing, you risk a customer who has paid but received nothing.
At ordinary volumes, these are edge cases, absorbed by support desks and end-of-day reconciliation. At crisis volumes they became a dominant traffic pattern — and the resolution backlog that followed showed what happens when reconciliation processes built for the edge case meet the average case.
The failures then compound mechanically. Mobile apps, PoS terminals, merchant integrations and desperate human beings all respond to a failed request the same way: they try again, usually immediately, often repeatedly. Each retry lands on infrastructure that is already saturated, lengthening queues and producing more timeouts, which produce more retries.
Engineers call the result a retry storm, and it is the specific mechanism by which a congested payment platform tips over into a collapsing one. Adding servers does not fix it; a storm will fill whatever capacity exists. The discipline that addresses this issue has two components: every retry must be made safe, and every retrying client must be polite.
Why the work travels
Nigeria’s cash crunch amounted to an involuntary, country-scale stress test of instant payment infrastructure, and its findings are not local. Every market converging on real-time account-to-account payment as the default – the United Kingdom with Faster Payments and its New Payments Architecture programme, Brazil with Pix, India with UPI — inherits the same physics.
Once instant rails stop being an alternative and become the medium through which an economy breathes, ambiguous failure handling stops being an edge case and becomes the central reliability problem, with consequences measured in public trust rather than error budgets.
Nigeria’s episode demonstrated both halves of that proposition within a single quarter: the migration of a cash economy onto digital rails in weeks and the market-share consequences for institutions whose failure handling was not ready.
