Deferred, not abandoned: the payments address migration turns into a patchwork

On 27 August 2026 Swift announced that it would defer every payments change planned for November. That includes the step the industry has been preparing for since 2023: from 14 November, payment messages in the ISO 20022 standard were to stop carrying addresses written purely as free text. The announcement put it plainly: “Swift will defer all payments changes.” For the address rule, Swift intends to consult banks, central banks and market infrastructures and to provide “an update by December, at the latest”. On the same day the Bank of England postponed its November standards release, including the messaging standards for CHAPS, and the Federal Reserve moved its Fedwire release to November 2027. Two weeks later the European Payments Council (EPC) dropped its end date of 15 November for the Single Euro Payments Area (SEPA).

Two readings suggest themselves, and neither holds up. One still treats 14 November as binding, relying on Swift’s own explainer pages, which carry the old date beneath a notice about the deferral. The other treats the deferral as an all-clear, because no date is pressing any more. The evidence supports a third reading: the target stands, the common date has gone. According to the Payments Market Practice Group (PMPG) overview dated 20 September, 17 of 27 payment systems follow Swift, nine take their own route and one has not committed itself. The range runs from an interim tolerance in T2 to a switch-off in Korea on the original date. By Swift’s own account at Sibos, the harder task lies with the data. In August, Swift measured that 54.8 per cent of debtor addresses in Swift payments still consisted of free text only. Nobody can say today when or how Swift will decide. What follows sets out what had been decided by early October and is not a forecast of the new date.

Swift has postponed the end of the free-text address, not just the move to fully structured data

According to Swift, the decision to remove unstructured postal addresses from payment messages was taken in 2023 as part of the annual Standards Release 2026. Only two formats were to remain. A fully structured address requires at least town name and country in dedicated fields, puts street, building number, postcode and other details into their own fields too, and may not contain a free address line. A hybrid address requires the same mandatory fields but allows up to two free lines of 70 characters each alongside them. In Swift’s own example, the hybrid form puts “LONDON” and “GB” into the two mandatory fields and the rest of the address, “55 MARK LANE, THE CORN EXCHANGE, 6TH FLOOR, EC3R 7NE”, into a single free-text line. The hybrid format has been permitted since November 2025. For agents, meaning the banks involved, Swift’s rules still accept the Business Identifier Code (BIC) on its own.

What Swift deferred on 27 August is the first step, the end of the purely free-text address. The announcement explains that “progress remains uneven with large parts of the industry across all regions still unable to meet the requirement”, and that several communities had formally asked Swift and their domestic payment systems, “who have similar deadlines”, for more time. On 21 September Swift added a date that has since been widely applied, wrongly, to addresses. The 12 June 2027 date covers the decoupled part of the release, namely MT messages outside Category 1, securities and funds messages, tracker messages, and exceptions and investigations. The address rule and the other payments changes are not on that list. The consultancy Projective Group was therefore right to call early coverage “incomplete” on 7 September, since more than an address deadline had moved: “All payments changes are deferred.”

The confusion is understandable, since Swift contributes to it. Its explainer page on the removal of unstructured addresses carries a notice about the “Revised Implementation of Standards Release 2026” at the top, while the text below still says that from 14 November unstructured messages may be “rejected or delayed”. The May announcement stating that “After 14 November 2026, unstructured addresses will be removed” appears in the related-content panel on the page about the deferral. Anyone quoting Swift should check the date on the page as well as its sender. For planning purposes, only the 27 August announcement and its update count, and they give no date for the address rule.

More than half of the addresses in Swift payments are still free text

Swift itself measures how far the industry is from the target. Its adoption data for August 2026, taken from customer credit transfers (pacs.008) under CBPR+, show 54.8 per cent of debtor addresses as unstructured, 24.1 per cent as fully structured and 16.5 per cent as hybrid, with 4.7 per cent carrying no address at all. On the creditor side the free-text share stood at 56.2 per cent, and 13.5 per cent of payments carried no creditor address. In April Swift had reported 61.2 and 62.9 per cent unstructured addresses, so the share fell by just over six percentage points in four months. Extend that pace in a straight line and zero would only be reached in a little under three years. That is a calculation of my own from two data points and not a forecast, but it shows how far the industry was from 14 November.

The statistics also only capture format. Swift itself notes that the figures say nothing about “the completeness of information supplied beyond the mandatory data elements”. A town field can be filled in and still be wrong. Anyone who has chased settlement failures in securities operations will recognise the pattern from another corner of the business: the message is formally correct, and the fault lies in the reference data behind it, in that case the standing settlement instruction. In payments, much of the reference data sits with corporate clients. That is why Swift asked corporates in May to maintain creditor addresses in their own ERP or treasury systems and to supply at least town and country, whether they submit via MT101, via pain.001 or through a bank portal. Those still using MT101 have to switch to field 59 with option F.

At Sibos in Miami at the end of September, Swift offered its own reading of the deferral. Nicolas Stuckens, who is responsible for ISO 20022 adoption and data quality at Swift, said, as reported by Finextra on 30 September, that the harder task is ensuring reliable and structured data “can be sourced, maintained, and carried through the payment chain”. Danielle Sharpe, who heads transaction banking at Standard Chartered in the Americas, described the effort as an ongoing operational task: “This is a data programme.” Frank Van Driessche, head of the ISO 20022 practice at the Federal Reserve Bank of New York, called the deferral “a loss” and at the same time “the right decision, albeit late during the process”. Finextra summarised the lesson drawn by Swift’s chief operating officer, Jerome Piens, as follows: a big-bang approach for 12,000 institutions on the network had been too ambitious.

The free AI tool that Swift has offered since late 2025 does little to change this. The model extracts town and country from the address fields of an MT103 message, “does not convert an unstructured address into a structured one”, as the description in its GitHub repository puts it, and is not built into any Swift product. That meets the minimum the hybrid format requires, but it does not build a structured reference database. The gap Swift measures will therefore close where addresses originate: in client onboarding, in the way corporates maintain creditor data, and in the channels through which instructions arrive. As Danielle Sharpe put it, data being somewhere in the bank does not mean it can be used.

Seventeen of 27 systems follow Swift, nine take their own route

The PMPG, a group of market practitioners that develops guidance on ISO 20022 in payments, has published an overview as at 20 September that sorts 27 payment systems by how they treat unstructured addresses after Swift’s decision. The picture is admittedly less chaotic than the word patchwork suggests. Seventeen systems align with Swift, which the table records as November 2027, among them CHAPS in the UK, Fedwire and CHIPS in the US, Lynx in Canada and RIX in Sweden. Swift has not confirmed November 2027 itself, so the attribution is the PMPG’s reading. The remaining ten, nine with a course of their own and one without a position, show how quickly a single common date has turned into a series of separate decisions.

Five systems will keep accepting unstructured addresses. T2 and EURO1, the two large euro systems, as well as SIC in Switzerland and CIPS in China, are going ahead with their 2026 release, while Singapore’s MEPS+ supports all three formats and is still assessing the impact on its release. On 4 September the European Central Bank (ECB) decided to move the TARGET releases back by only two weeks, to 28 November, and to introduce a measure in T2 that allows unstructured addresses “for a limited period”. The underlying change request, T2-0188-SYS, now part of the November release and installed in the test environment since 9 October, is narrower in scope but open-ended in time. It provides for the tolerance only in application-to-application (A2A) communication and for six message types, while the T2 graphical interface (U2A) will accept only hybrid or structured addresses from November. It also sets no end date, stating that the change will remain valid “until a new requirement will be raised via change request”. For participants this creates an obligation that rarely features in the debate about the deferral: according to the ECB, they must remain able to receive unstructured addresses while moving to structured ones themselves.

Four systems are removing unstructured addresses regardless, or have already done so according to their plans: BOK-Wire+ in Korea on the original 14 November, PhilPaSplus in the Philippines in November, Taiwan’s FISC as of 5 October according to the table, and RENTAS in Malaysia in February 2027. Japan has given no position. The PMPG attaches a footnote to these routes, warning that they “may create temporary interoperability considerations with other payment ecosystems”. In practice, a cross-border payment that Swift still allows with a free-text address may meet rules in the domestic leg of one of these countries that it no longer satisfies. No rejections from these corridors have been documented so far, and the footnote is a warning rather than a finding. Anyone paying in Korean won, Philippine pesos, Taiwan dollars or Malaysian ringgit should nevertheless take it seriously.

The mechanics differ as well. According to its own FAQ, Fedwire will support only structured and unstructured addresses for key parties until its November 2027 release, and the hybrid format only after that. The Bank of England expects the change to arrive with its November 2027 release, though explicitly “subject to confirmation”. Clearstream Europe will continue to accept unstructured addresses “until the revised industry implementation date is confirmed by Swift”. For the SEPA area, the EPC lifted the 15 November end date across all five of its rulebooks on 9 September and plans to set a new one in October. No new date had been published by 9 October. Deutsche Bank tells corporate clients that the final date has yet to be confirmed. According to a client notice from Hamburg Commercial Bank, the German Banking Industry Committee has suspended the planned withdrawal of legacy formats for the time being. J.P. Morgan is sticking with 14 November, but only as a readiness commitment: from then on it processes hybrid and structured addresses, while stating that unstructured addresses may continue to be used “until at least November 2027”.

Earlier deferrals came with a new date

Deferrals are nothing new in this migration. The start of ISO 20022 for cross-border payments had already been pushed back by a year, to the end of 2022. The next delay followed in October 2022, triggered by the ECB, which moved the launch of its new large-value system T2 from 21 November 2022 to 20 March 2023 because some users had not completed their testing. Swift followed on 27 October 2022 at the request of what it then called “an overwhelming majority of our global community” and set the start of MT and ISO 20022 coexistence for 20 March 2023. The same announcement carried a commitment: “the revised start date of 20 March 2023 will be definitive.” The end of coexistence on 22 November 2025 was never moved, and in March 2026 Swift reported, by its own account, an adoption rate of 97 per cent after that weekend.

The SEPA area offers a statutory variant. When it became clear in early 2014 that the 1 February end date could not be held, the EU adopted Regulation (EU) No 248/2014. It allowed legacy formats “until 1 August 2014”, with retroactive effect from 31 January. The recitals give the reasoning in figures: migration for credit transfers had only risen from 40 per cent in June 2013 to around 64 per cent in November 2013, and for direct debits it stood at 26 per cent. The extension was described as a “one-off additional transitional period of six months” and an exception to be kept as short as possible. Anyone who has watched from a trading floor how markets react to a postponed deadline will know the difference between these two cases: a new date gets priced in, an open date gets waited out.

In 2026 both are missing: the new date and a legislative backstop. The obligation to provide an address is law: since 30 December 2024, Article 4(1)(c) of the Transfer of Funds Regulation (EU) 2023/1113 has required “the payer’s address including the name of the country”. The word “structured”, however, appears nowhere in the regulation. The Financial Action Task Force (FATF), in its revised Recommendation 16 published in June 2025, requires only country and town for the beneficiary and expects the new requirements to be in place by the end of 2030. The Committee on Payments and Market Infrastructures (CPMI) recommends structured addresses with at least country and town by the end of 2027 and stresses that its requirements “are not, and should not be, regarded as regulatory requirements”. Address structure is therefore a market rule, set by Swift and the payment systems. That is why it could slip without any change in the law, and why the new date depends on industry readiness rather than on a deadline in the Official Journal. The CPMI’s end-2027 marker now sits roughly where the Bank of England expects its new date.

The time gained is best spent on the reference data

None of the bodies involved reads the deferral as a retreat from the target. Giorgio Andreoli, Director General of the EPC, said of the decision taken on 9 September: “Although the implementation timeline is being delayed, the direction of travel remains unchanged.” The Bank of England urges banks and end users to “not introduce further delays” and points out that hybrid and structured formats are already available. The ECB intends to argue for a return to the original roadmap “as soon as possible”. At the ECB’s advisory group on payment infrastructures in August, the PMPG said that more time would come with “appropriate accountability measures”, still to be defined. The PMPG wrote in September: “Additional time only creates value if it is used to improve readiness.”

The risk therefore lies in a bank’s own project planning. Projective Group described it this way: “Programmes that relied on an external date to win internal priority now need another argument”. Where an address clean-up programme justified its budget with 14 November, that argument has gone, and the temptation to let the work rest until the next date is considerable. Yet the data gap is the same as in August, the receiving side in T2 has to handle unstructured, hybrid and structured addresses from 28 November, and payments into the four countries with earlier switch-offs are subject to their rules. The banks that already send fully structured data are probably the least likely to experience the transition as a breathing space, because they still have to receive and process unstructured addresses. The deferral has postponed a date. It has not shortened the work on the reference data.

Recommendations

Glossary