This November, the Swift network was due to start rejecting cross-border payments based on how an address is written. The target was the free-text postal address, the ‘10 King Road, London’ format that has ridden along in bank messages since the 1970s. The network would refuse any payment still carrying one and bounce it back to the sender before it got anywhere near settlement. All the network asks for is a town and a country, typed into the right boxes. On 27 August, with eleven weeks to go, Swift pulled the date, though not the requirement. In its announcement, Swift said large parts of the industry, across all regions, were still unable to comply.
The last step is the hardest
The change is the last step in moving addresses to ISO 20022, the standard that replaced Swift’s decades-old message formats with structured data. Most of that work is done. But the address remains a sticking point: a bank can rebuild every payment engine it owns, yet if what sits in its customer records or in a corporate client’s payment file is still a long string of free text, the payment fails regardless, on whatever date the switch is finally thrown.
The network accepts three formats: fully structured, hybrid (structured town-and-country, with the rest left as text) and the old unstructured form. The plan was to switch off the last of these on 14 November. That switch-off is what has been deferred. Swift has taken the whole payments portion of its 2026 standards release off the calendar without setting a new date, and will spend the coming weeks consulting banks, central banks, market infrastructures, market practice groups and corporates on when, and in what form, to put it back. An update is promised by December at the latest. The securities and trade messaging changes due at the same time – a separate set of messages from the payments this article is about – have been split off and go live in the first quarter of 2027.
Swift’s own guidance is explicit that the requirement applies to all payments: corporate, securities, trade, foreign exchange, and funds. A letter of credit settlement travels the same rails as a salary run; so does the payment leg of a documentary collection and every drawdown under a supply chain finance programme. An otherwise flawless trade transaction – documents presented, checked, and taken up – can still fail at the last step because the beneficiary’s address was typed into the wrong box, possibly years ago, by someone who no longer works there.
Nobody can add the town later
What made this deadline different from the industry’s many others was that nothing could be done about a failure after the fact. Swift said in March that it was “not possible for Swift to develop a contingency solution” for institutions that fell behind, because the address has to be captured correctly where the payment starts – in the originating bank’s records, or in the corporate client’s systems before that. No amount of downstream cleverness can reconstruct information that was never structured in the first place. The contrast with the rest of the migration is instructive. As the old MT formats are retired, Swift offers a translation service to convert stragglers’ messages into the new standard – chargeable, but available. Even that service will not touch a message containing an unstructured address. Somebody downstream can fix everything else in a payment. The address, nobody can. Which is why, when several of its communities formally asked for more time, time was the one thing the network was in a position to give.
The corporate side has not been let off. A company sending payment instructions from its own treasury or ERP system (the software that generates its invoices and payment files) will still have to supply a structured town and country with every payment it sends, however it sends it, on whatever date the requirement finally lands.
The biggest banks are furthest behind
The most detailed public picture of the shortfall behind Swift’s decision comes from RedCompass Labs, a payments modernisation consultancy, which surveyed 308 senior payments professionals across Europe and North America, published in March. It found 44 per cent of banks not on track for November – and among those behind, only 40 per cent considered their projects recoverable. Twenty per cent of institutions with assets above $250bn described the deadline as outright unrealistic, compared with around five per cent of smaller banks.
“That’s counterintuitive until you realise the biggest banks carry the most legacy systems, the most customer records, the most interconnections to untangle,” said Pratiksha Pathak, partner and head of payments at RedCompass Labs, whose research it is. “That’s where the scale works against you.”
The survey’s systems data explains the pessimism: 60 per cent of core banking applications, 60 per cent of onboarding and know-your-customer platforms and 55 per cent of payment engines did not yet support fully structured addresses. Even payment engines, the furthest along, are not enough, because an engine can only work with the data it is fed from upstream. A bank that has modernised its payment engines but not its customer records has fixed the part that was never going to fail.
Then there is the half of the problem the banks do not control. RedCompass found that, on average, 32 per cent of an institution’s client base still needed to migrate their address data, rising to 48 per cent at the largest banks. Nearly a quarter of banks are pushing the task back onto clients by asking them to redo their know-your-customer checks, which carries a particular hazard, because the deadline, whenever it lands, binds the banks, and their clients are under no such obligation.
Swift’s own traffic data makes the scale plain. As of July, 58.3 per cent of payments on the network still carried unstructured debtor addresses, and 59.3 per cent unstructured creditor information – down from 61.2 and 62.9 per cent in April.
Ask Norwich
The requirement was never really about addresses. The free-text format now being retired dates from an era when payments, like letters, were read by people, and a person will make sense of almost anything typed in front of them. The Post Office tried taking the reading away first. In the summer of 1959, the Postmaster General launched a trial in Norwich that gave some 150,000 addresses a machine-readable code, on what was then a novel idea: that an envelope should make sense to a sorting machine and not just to the postman who knew the street. The rollout took until 1966 to begin and until 1974 to finish, when the programme’s closing act was to recode Norwich itself – a piece of civic history that Alan Partridge would be proud of.
Restructuring one country’s addresses took fifteen years of patience and persuasion, and in all that time nobody’s letter was refused for want of a postcode. The payments industry allowed itself a cut-off date instead, and then, with the date in sight, came round to the Post Office’s way of doing things, which is to take as long as it takes. Meanwhile, five decades of free-text latitude have filled the industry’s systems with abbreviations, transpositions, stale entries, and improvisation. Structured addresses are how the mess gets discovered. They also matter well beyond operations: a payment’s address data feeds sanctions screening, and global standards on payment transparency are moving in the same direction as Swift. The deadline was an audit: the network checking fifty years of its members’ customer data and bouncing whatever failed. The audit has been postponed, with the mess still in the files and the auditor still coming.
In November last year, Swift released a free, open-source artificial intelligence model whose sole job is to read a free-text address and infer a structured town and country from it – the network shipping AI tools to help its members meet the network’s own deadline. In its June update, Swift acknowledged the community still had a long way to go to reach the target of zero unstructured addresses. In August, it accepted the community’s request for more time and removed the date.
The tools exist
Harri Rantanen, a business developer at SEB who spent years inside the standards machinery of ISO 20022 itself, has built the grassroots version: a free address converter, released in March through the Standardised Trust community. Throughout June and July of 2026, it processed 1,000 transformation actions, against a network that moves tens of millions of payments a day. Speaking in a personal capacity, Rantanen said the difficulty is that such tooling is too “open source” for banks, whose regulators require formal approval before anything enters a payment workflow; corporates and their software vendors face no such wall. The free tooling is open to the parties whose systems generate free-text addresses and, in practice, closed to the banks that will suffer the rejections when they come.
“I like good regulation – there are many, even in the EU – but not when it goes beyond the common-sense rationale,” Rantanen said.
Trade finance does much of its business in places where an address is a district, a landmark, and a family name – and where nobody has ever asked the records to be tidier than that. A treasury team in Frankfurt can clean up its supplier records in a quarter; the question is what happens to the settlement legs flowing to and from the thousands of smaller counterparties who have never heard of ISO 20022 and will find out about it when their money comes back.
The problem is deeper than the budget
RedCompass puts the average spend on the 2026 requirements at around $20 million, rising beyond $30 million at the largest institutions, with 13 to 14 additional staff assigned on average – and the data and technology gaps persisting regardless. “This is emphatically not a story about underinvestment,” Pathak said. “It’s a story about the sheer depth of the problem. Money alone doesn’t undo decades of accumulated unstructured data. It surprises a lot of boards who assume a big budget buys a solved problem.”
What a miss looks like was not abstract when TTP first spoke to her in July. A payment initiated with a free-text address will fail somewhere along a chain of systems that all assume the data is right – and if it fails late, the customer has already been debited. Some banks, she said, were still working out which systems those are: “We’re in July, talking about November, and they’re still discussing how to map this information. Even backtracking the whole plan, you are already running at the risk of not meeting these deadlines.” By the end of August, Swift had reached the same conclusion about the industry as a whole. “Swift’s decision not to proceed with the entire Standards Release 2026 as planned, not just the unstructured address requirement, shows how much of a mountain this migration is,” she said.
Nobody, she was clear at the time, should plan on the date moving. European schemes, SEPA among them, had aligned their own structured-address timelines to Swift’s, so a slip in one would mean unwinding several. As of late July she had seen no indication of delay, though clients kept asking – less in hope, she suggested, than to check whether their competitors were hoping. Her overall reading stopped short of catastrophe: “a pressure, not yet a crisis.” That was the right advice, and it stayed right until the last week of August. What her clients were asking about was being asked, more formally, elsewhere: several Swift communities requested more time, and on 27 August the network granted it to everyone. The unwinding she described began within a day, when the Bank of England said it would defer its own November release for CHAPS in full, to stay aligned. Not every scheme will follow; she expects some will go ahead as planned.
In a written response after the announcement, she said: “Banks should not wait for a new deadline to keep migrating. Otherwise, they risk finding themselves back in exactly the same position when the revised timeline is confirmed. Standards Release 2027 is still coming, too.”
And the trade legs riding on the same rails? “I don’t think I’ve had that discussion directly,” she said. Banks are consumed by getting core payment flows over the line; settlement instruments are “a problem to solve later”. The letter of credit is not behind in the remediation queue. It is not in the queue.
The industry asked for this
The morning of 15 November will now be like any other. The deadline was not imposed from above; it was set in 2023 by the payments community itself, through its market practice group, and the major currency systems – dollar, euro, sterling among them – aligned their timelines to it. That is what gave November its force. It is also what made it movable, and the extension answers, in the title of Swift’s own announcement, a ‘community request’. The same word, both times. When the magazine you are holding was printed, no new date existed. Swift had promised further word at Sibos and a decision by December, and the industry had spent millions preparing for a cut-off it asked for, published years in advance, and was, in the end, excused.
For fifty years, the envelope did not matter. A cross-border payment moved on its account numbers and amounts while the address rode along as decoration, read by no one, checked by nothing. One day the network will read the envelope, and anything illegible will come back marked return to sender. For now, the sorting office keeps reading by hand, and the date on the notice has been left blank.
