SWIFT and Correspondent Banking

41.0 What this chapter gives you#

1. You will be able to explain why SWIFT has never moved a penny, and say precisely what it does move instead.
2. You will be able to draw the correspondent chain for a dollar payment from Sheffield to Izmir and identify the point at which a dollar left the United States, which is never.
3. You will be able to account for eighty-five missing dollars on a USD 214,000 payment by naming the charge basis that entitled each bank to take its cut.
4. You will be able to separate the four causes of a three-day international payment, cut-off times, business calendars, value dating and compliance screening, and say which of them a faster network would fix.
5. You will be able to say which side of which balance sheet a nostro sits on and which side a vostro sits on, and why one account carries two names.
6. You will be able to map an MT 103 onto a pacs.008 and explain why translating back the other way loses data that cannot be recovered.
7. You will be able to say what changed on 22 November 2025, and what it now costs an institution that has not migrated.
8. You will be able to explain why SWIFT’s “nearly 60 per cent within 30 minutes” and the Financial Stability Board’s “50.6 per cent within one hour” are both honest figures.
9. You will be able to tell a customer what a UETR is, where it lives in the message, and what the gpi tracker will and will not tell them.
10. You will be able to argue that a correctly structured beneficiary address is a speed feature rather than a compliance chore, and show the mechanism by which it is one.

On a Tuesday morning a finance manager at an engineering firm in Sheffield instructs a payment of USD 214,000 to a steel mill outside Izmir. She has the mill’s account number, the mill’s bank, and a BIC. She types it into her bank’s business portal at 09:40, approves it on a second device, and gets a reference number. The screen says the payment has been sent.

On Friday afternoon the mill emails to confirm receipt. USD 213,915 has arrived. Eighty-five dollars are missing, nobody in Sheffield authorised anyone to take them, and three working days have gone by. Her bank’s helpline tells her the payment “left us on Tuesday”, which is true, and that “it’s with the correspondent”, which is also true, and neither statement helps.

Nothing went wrong. That is the disorienting part. Every institution in the chain did precisely what it was designed to do, the money arrived, and the arithmetic is defensible line by line. An international payment is not a thing that travels. It is a sequence of separate domestic bookkeeping entries in different countries, stitched together by messages, and this chapter is about the messages, the bookkeeping, and the gap between them.

All figures, dates and message versions here are accurate at the time of writing, August 2026. The standards in particular have just been through the largest change in their history.

The plain version#

Start with the single most common misunderstanding in all of payments. Ask most people what SWIFT is and they will say it is the system that sends money abroad. It is not. SWIFT sends no money at all. It has never sent a penny, a cent or a lira, and it never will.

SWIFT sends messages.

Two shops and two tins#

Imagine a corner shop in Sheffield and a corner shop in Istanbul. The two owners have known each other for years. Because their customers keep asking to send money to relatives in the other city, they set up an arrangement.

The Sheffield owner keeps a tin of Turkish lira at the Istanbul shop. The Istanbul owner keeps a tin of pounds at the Sheffield shop. Both tins are labelled, both are counted, and both live permanently in the shop where they sit. They do not travel.

Now a customer walks into the Sheffield shop and wants to send money to her brother in Istanbul. She hands over cash. The Sheffield owner picks up the phone and says: pay my customer’s brother out of my tin. The Istanbul owner opens the Sheffield tin, counts out the lira, hands them to the brother, and writes down that the tin is now that much lighter.

Money has moved from one person to another across two thousand miles. Nothing physical crossed a border. The cash the customer handed over is still in Sheffield. The lira the brother received were already in Istanbul. All that crossed the border was a phone call.

That is correspondent banking. The tins are called accounts, the shops are called banks, and the phone call is called SWIFT.

Why SWIFT had to be invented#

The phone call is the weak point. It can be misheard, it can be faked by someone who sounds convincing, and it cannot be proved afterwards. When there are not two shops but eleven thousand, in two hundred and twenty countries, all using different words for the same thing, “phone the other bank” collapses.

So in 1973 a group of banks built a shared, secure, standardised messaging network. They called it the Society for Worldwide Interbank Financial Telecommunication, which nobody says out loud, and it went live carrying real messages in 1977. It is a cooperative, owned by the banks that use it, based in Belgium.

It gives its members three things. A private network, so messages do not travel across the open internet. Cryptographic proof of who sent each message, so a fake instruction is very hard to inject. And, least glamorous but most important, a rigid vocabulary: a payment instruction has numbered fields, field 32A is always the value date, currency and amount, field 59 is always the beneficiary, and a bank in Japan and a bank in Chile fill in the same numbered boxes.

What SWIFT does not give its members is money. It has no view of anyone’s balance, it holds no accounts, and it does not know whether the payment it carried actually happened. It delivers the instruction. The banks do the rest.

Why there is usually a chain#

The two-shop story works because those two shopkeepers know each other and keep tins with each other. Real banks cannot do that with every other bank on earth. There are tens of thousands of banks. Maintaining an account with each one would mean tens of thousands of pots of idle money, tens of thousands of reconciliations, and tens of thousands of relationships to check for fraud and criminality.

So banks keep accounts with a small number of large banks, and those large banks keep accounts with each other. If you want to send dollars, the chain has to pass through a bank that has actual dollars in America, because dollars only really exist in America. If you want to send pounds, the chain has to pass through a bank that has actual pounds in Britain.

Here is the Sheffield firm’s payment, drawn as a chain.

Four institutions, three messages, and at no point did a dollar leave the United States. What travelled was a claim: at the end, the Turkish bank has more dollars in New York than it had before, and the UK bank has fewer.

Where the eighty-five dollars went, and the three days#

Each of those middle banks did work. Each one has a price list. Because the payment was sent on shared-charge terms, which is the default for most business payments, the sending firm paid its own bank’s fee up front and every bank downstream was entitled to take its cut out of the money itself.

Eighty-five dollars, all of it disclosed in somebody’s tariff, none of it visible on the Sheffield screen at the moment of sending.

The days are a different story and a more interesting one. The messages themselves were delivered in seconds. What consumed Tuesday, Wednesday and Thursday was everything around the messages: a bank’s daily cut-off time for dollar payments, after which your instruction is tomorrow’s work; the fact that New York is five hours behind Sheffield and its dollar clearing system has its own opening hours; and, in this case, a compliance check. The steel mill’s name resembled a name on a sanctions list closely enough for an automated filter to stop the payment and put it in a queue for a human being. The human being asked for the mill’s full street address, the question travelled back up the chain to Sheffield, the answer travelled back down, and by the time the payment was released it had missed the next cut-off.

The parcel tracking number#

For decades this was invisible. Once a payment left, the sending bank genuinely did not know where it was. It could ask, by sending another message, and wait.

Then in 2017 the banks agreed to attach a tracking number to every payment: a long, random, unique string generated when the payment is created and carried unchanged all the way down the chain. Every bank that touches the payment reports back to a central tracker: I received it at this time, I passed it on at this time, I took this much in fees, I used this exchange rate. The sending bank can now watch the payment move, the way you watch a parcel. That programme is called SWIFT gpi. It did not make payments faster by itself. It made them visible, which turns out to be most of what people wanted.

Where the plain version stops being true#

The tin is not a tin#

The most important simplification above is the pot of money sitting in the other shop. There is no pot. When a UK bank “holds dollars in New York”, what it actually holds is a claim on the New York bank: a promise, recorded as a number on the New York bank’s ledger, that the New York bank owes it that many dollars. It is a deposit, which means it is an unsecured loan to that institution. If the New York bank fails, the UK bank does not walk in and collect its dollars. It joins the queue of creditors.

This changes the shape of the whole business. It is why banks care intensely about which correspondents they use, why a correspondent relationship takes months of due diligence, and why a bank will keep balances with two or three correspondents in the same currency rather than one. It is also why the balance is not a passive store. It is a working balance, run deliberately thin, topped up during the day, drawn down by outgoing payments, and frequently supported by an intraday credit line from the correspondent. “Trapped liquidity” in nostro accounts means exactly this: money the bank cannot use for anything else, parked in other people’s institutions in other people’s jurisdictions, purely to be able to pay.

And because those balances sit inside another country’s legal system, they are subject to that country’s law. Dollar balances in New York are within reach of United States authorities, whoever owns them. That is the mechanism by which one country’s sanctions policy becomes binding on banks with no other connection to it, and it is the single largest reason correspondent banking has become as cautious as it has.

The messages do not always follow the money#

The chain drawn above implies one flow: the instruction walks down the chain alongside the value. That is one of the two ways it is done, and often not the one used.

In the serial method the payment instruction really does pass from bank to bank down the chain, each one debiting and crediting the accounts it holds and then forwarding a fresh instruction to the next.

In the cover method there are two separate flows. The originating bank sends the customer payment instruction directly to the beneficiary’s bank, so the beneficiary’s bank knows immediately that money is coming and for whom. Separately, it sends a bank-to-bank instruction into the correspondent chain to move the actual funds, the “cover”. The customer message and the cover message travel different routes and can arrive at different times.

This matters more than it sounds. Under the cover method the intermediary banks moving the money historically saw only a bank-to-bank instruction with no detail of the underlying customer. That gap was exploited: institutions stripped out the names of sanctioned parties, and intermediaries screened a message that no longer contained anything to find. The standards were changed in 2009 to force the underlying customer details into cover messages, and the enforcement actions that followed are among the largest financial penalties ever imposed. The simple chain diagram hides all of this.

A day per hop is a symptom, not a mechanism#

It is tempting to read the three-day journey as three banks each taking a day. Nothing in the system works like that. Delivery of the message is measured in seconds. The delay is made of four separate things, worth separating because they respond to different fixes.

The first is cut-off times. Every bank publishes a deadline for each currency, after which an instruction becomes next business day’s work. Miss it by a minute and you have lost a day. These are bank-specific and published in each bank’s own tariff and service schedules; read the ones that apply to you rather than any general claim about them.

The second is business calendars. A payment in dollars needs the American clearing system open, a payment in euros the European one. A chain crossing three currencies must satisfy three calendars, and every weekend and public holiday in any of them stops the clock.

The third is value dating, which is bookkeeping rather than movement. A payment can be credited today with a value date of tomorrow, so the recipient sees it but cannot use it.

The fourth is compliance screening, in practice the biggest single source of unexplained delay in the whole system. It gets its own treatment below.

None of these is fixed by making the network faster. The network was never the bottleneck.

SWIFT is not the only way, and not every hop takes a fee#

Two further corrections. First, plenty of cross-border value moves without any SWIFT message. A bank moving money between its own branches in two countries does it on its own internal ledger. Card networks run their own messaging and settlement. The euro area’s retail schemes run over their own clearing infrastructure. China and Russia have built their own interbank messaging arrangements. Money transfer operators run closed-loop networks where both ends are theirs and the “cross-border” element is an internal transfer plus local funding on each side. SWIFT is dominant in interbank correspondent messaging. It is not a synonym for international payment.

Second, the fee-per-hop is a consequence of a choice recorded in the message, not an automatic toll. Payments can be sent on terms where the sender bears every charge in the chain and the beneficiary receives the full amount, on terms where charges are shared, or on terms where the beneficiary bears everything. Within the United Kingdom and the European Economic Area, the shared-charge basis is effectively mandated by payment services regulation for payments in the relevant currencies, which is why intra-European transfers rarely arrive short and transfers to Türkiye or Nigeria frequently do.

The technical version#

What SWIFT actually is#

The legal entity is a cooperative society under Belgian law, headquartered in La Hulpe, owned and controlled by its member institutions, governed by a board of twenty-five directors elected by shareholders. It was founded in 1973 and carried its first live traffic in 1977.

The scale, on SWIFT’s own published figures for 2024: more than 11,500 connected institutions, more than 220 countries and territories, an average of more than 53 million FIN messages per day, and FIN availability of 99.99 per cent. Note carefully what that message count is and is not. FIN traffic includes securities, treasury, foreign exchange, trade finance and reporting messages as well as payments. It is not a count of payments, and each payment in a chain generates several messages.

The services are worth naming precisely, because practitioners use the names as shorthand.

Security of the endpoints, as opposed to the network, has been SWIFT’s largest operational preoccupation since 2016, when attackers who had compromised a member’s own systems used entirely valid credentials to issue fraudulent instructions from the Bangladesh Bank. The network worked perfectly; the customer’s premises did not. The response was the Customer Security Programme and its Customer Security Controls Framework, against which members must attest annually. This is the correct frame for SWIFT security generally: the network is extremely hard to break, and every significant incident has been an endpoint compromise.

Governance and oversight#

SWIFT is systemically important without being a financial institution, which makes it awkward to regulate. The arrangement the central banks settled on is cooperative oversight. The National Bank of Belgium is lead overseer, because SWIFT is incorporated in Belgium, supported by the G-10 central banks: Banca d’Italia, Bank of Canada, Bank of England, Bank of Japan, Banque de France, De Nederlandsche Bank, Deutsche Bundesbank, the European Central Bank, Sveriges Riksbank, the Swiss National Bank, and the Federal Reserve System, represented by the Federal Reserve Bank of New York and the Board of Governors. A wider SWIFT Oversight Forum brings in central banks from other major economies.

The oversight is conducted against a set of High Level Expectations covering risk identification and management, information security, reliability and resilience, technology planning, and communication with users. What it is not is prudential supervision, and the reason is stated plainly by SWIFT itself: it is “neither a payment nor a settlement system”, and in its own words, “we don’t actually move money.” There is no settlement risk to supervise because there is no settlement, only operational and security risk, and that is exactly what the oversight covers.

Nostro and vostro#

The vocabulary is Italian and, once you see the trick, trivial. The same account has two names depending on whose books you are reading it in.

Nostro means “ours”. A nostro account is our account, held with you, denominated in your currency. On our balance sheet it is an asset: money owed to us.

Vostro means “yours”. A vostro account is your account, held with us, denominated in our currency. On our balance sheet it is a liability: money we owe you. Older documentation sometimes uses loro, “theirs”, for an account a bank services on behalf of a third party.

A UK bank’s dollar account at a New York bank is that UK bank’s nostro and the New York bank’s vostro. One account, one balance, two labels, two sides of two balance sheets. The confusion in practice comes from people using “nostro” as a generic term for correspondent accounts regardless of viewpoint, which is harmless until it appears in a reconciliation specification.

Reconciling nostro accounts is a substantial operational function in its own right. The account servicer reports movements and balances, and the account holder matches them against its own expected entries. In ISO 20022 terms that is a camt.053 end-of-day statement, a camt.052 intraday report, and camt.054 debit and credit notifications, which replaced the MT 940 and MT 950 statements, the MT 942 interim report, and the MT 900 and MT 910 confirmations respectively. Aged unmatched items are a regulatory and audit concern, because an unexplained entry in a correspondent account is exactly what a fraud or a misdirected payment looks like.

The shrinking network#

The correspondent network has been contracting for well over a decade, and the data is unusually good because it is derived from the messages themselves. The Committee on Payments and Market Infrastructures, working with SWIFT, published an annual quantitative review built on MT 103 and MT 202 traffic, excluding MT 202 COV, covering more than two hundred jurisdictions.

The headline is stark. The number of active correspondents worldwide fell by about 22 per cent between 2011 and 2019, with a further roughly 3 per cent fall in 2019 alone, while the number of active corridors fell by about 12 per cent over the same period. The decline is universal but wildly uneven: over 2011 to 2022 the fall was about 19.4 per cent in Northern America and about 47.1 per cent in the Americas excluding Northern America, with Oceania down about 37.8 per cent. Small island developing states were hit hardest of all. Meanwhile the volume and value of payments flowing through the shrinking network kept rising. Fewer banks, carrying more. Concentration in this business means a single institution’s decision to exit a country can leave that country materially harder to pay.

The currency picture from the same data, as at December 2022, explains why the chain so often runs through New York. Of the value transferred, the US dollar accounted for 50.9 per cent, the euro for 27.3 per cent and sterling for 4.3 per cent.

The drivers of the decline are mostly regulatory in origin. A correspondent must satisfy itself about its respondent’s anti-money-laundering controls, and to some degree about the respondent’s own customers. That due diligence is expensive, the penalties for getting it wrong are enormous, and the revenue from a small respondent in a high-risk jurisdiction is not. The rational response for a large bank is to exit, and it is called de-risking. Attempts to reduce the cost side of that calculation include the Wolfsberg Group’s Correspondent Banking Due Diligence Questionnaire, which standardises what is asked, and SWIFT’s KYC Registry, which standardises where the answers are kept.

The last mile is always domestic#

An international payment always ends as a domestic settlement in the currency’s home system, and this is the single most useful thing to hold in your head when diagnosing one. Dollars settle across the Federal Reserve’s Fedwire Funds Service or across CHIPS. Euros settle across the Eurosystem’s T2 or across EURO1. Sterling settles across CHAPS, inside the Bank of England’s RTGS system, and for lower-value onward legs across Faster Payments. When a payment “arrives in the UK”, what has happened is that a UK institution’s sterling position at the Bank of England has increased, or a UK bank has credited a foreign bank’s sterling vostro account.

For a UK practitioner the consequence is that inbound international payments land on domestic rails and behave like domestic payments. An inbound sterling payment reaching a UK clearing bank for onward delivery will go out over CHAPS if it is large or time-critical, and CHAPS is same-day, irrevocable, real-time gross settlement with an early-afternoon cut-off for the same-day guarantee. Miss that cut-off and it is tomorrow’s payment. Alternatively the onward leg may go over Faster Payments, which settles in seconds and runs 24/7 with a maximum of GBP 1 million per transaction, subject to lower limits set by individual institutions. Neither rail knows or cares that the money came from abroad. They see a domestic instruction.

The domestic rails have themselves been converging on the same standard. The Fedwire Funds Service completed its migration to ISO 20022 on 14 July 2025. The Bank of England moved CHAPS to ISO 20022 and has been ratcheting up the data requirements, mandating Legal Entity Identifiers and purpose codes on certain CHAPS payments from 1 May 2025, with further enhanced data requirements consulted on for 2027. The point is that the same structured data now survives the whole journey instead of being flattened at each border.

MT and MX#

This is where the vocabulary trips people, so take it slowly.

MT stands for Message Type. It is SWIFT’s proprietary format, in service since 1977, carried over FIN. An MT message is a flat text structure of five blocks: block 1 the basic header, block 2 the application header, block 3 the user header, block 4 the text with the actual content, and block 5 the trailer. Inside block 4 the content sits in tagged fields, so field 20 is the sender’s reference, field 32A is the value date, currency and amount, field 50 identifies the ordering customer, field 57 the account-with institution, field 59 the beneficiary, field 70 the remittance information, and field 71A the details of charges.

Message types are numbered by category, and the first digit tells you the business.

So an MT 103 is a single customer credit transfer, an MT 202 is a general financial institution transfer, and an MT 940 is a customer statement. The messages ending in 92, 95, 96 and 99 in each category are the common group messages: cancellation requests, queries, answers and free-format.

The format’s constraints were reasonable in 1977 and became crippling. Names and addresses are unstructured free text in lines of thirty-five characters, typically four of them. Remittance information in field 70 is four lines of thirty-five characters: a hundred and forty characters to describe why the payment is being made. The permitted character set is restricted, which mangles accented names. And anything too long is truncated silently. A beneficiary address running to five lines loses its fifth line, and the sanctions filter downstream screens what remains.

MX is the informal name for the ISO 20022 XML messages that replace MT. ISO 20022 is not a message format so much as a modelling methodology with a shared dictionary of business concepts, from which messages are generated. Message identifiers follow a fixed pattern: a four-letter business area, a three-digit message number, a three-digit variant and a two-digit version. So pacs.008.001.08 is business area pacs, message 008, variant 001, version 08.

The business areas you will meet in payments are pacs for payments clearing and settlement, pain for payment initiation, camt for cash management, acmt for account management, and head for the Business Application Header that wraps the message.

The broad correspondence is as follows. Treat this as a map of intent, not a field-level mapping, because the two formats do not correspond one to one and that is the entire point of the change.

What ISO 20022 buys is data. Party names and addresses in discrete elements rather than free text, so a town is tagged as a town and a country as a country. Remittance information both structured and unstructured, and far longer. Purpose codes drawn from a controlled list. Room for Legal Entity Identifiers alongside BICs. An extended character set. And end-to-end identifiers that survive the journey.

The migration, and the date that mattered#

The cross-border implementation of ISO 20022 is governed by CBPR+, Cross-Border Payments and Reporting Plus: a set of usage guidelines defining exactly how the ISO 20022 messages are to be populated for cross-border traffic, developed with the Payments Market Practice Group and published through SWIFT’s MyStandards. CBPR+ messages travel over FINplus, not FIN. The guidelines were released in tranches, with the initial set covering the core payment and reporting messages, later additions covering cheques, direct debits, cancellations, and charges notifications and requests.

The coexistence period, during which institutions could send either MT or ISO 20022 for cross-border payments and reporting, ended on 22 November 2025. That is the date the industry had been working towards for years, and it is the reason a great deal of bank technology spending happened in 2024 and 2025.

What happens now, for an institution that has not migrated, is worth stating precisely because it is widely misreported. MT categories 1, 2 and 9 are no longer supported natively for cross-border payments and reporting. Institutions still sending MT are put through automatic conversion, with no opt-in, and institutions still needing to receive MT can use an in-flow translation service. From 1 January 2026, both the contingency processing for MT senders and the in-flow translation service are chargeable. So the MT world did not vanish on 22 November 2025; it became a paid fallback that degrades the data. Translation from ISO 20022 to MT is lossy by definition, because the structured fields have nowhere to go, and a bank running on translated MT is receiving less than it is being sent.

There is a second-order effect that shows up in the performance statistics. The Financial Stability Board’s key performance indicators recorded a fall in the share of wholesale cross-border payments credited within one hour between 2023 and 2024, and attributed part of it to the migration itself: message translation takes additional processing time. Standardisation is worth it, and it is not free while it is happening.

The UETR and SWIFT gpi#

The UETR, the Unique End-to-end Transaction Reference, is the identifier that makes tracking possible. It is a string of 36 characters, generated by the originator, and it must not be altered by anyone in the chain. In MT messages it lives in field 121 in block 3, the user header. In ISO 20022 it is carried in the Business Application Header and in the payment identification. It is required in the relevant payment instruction messages, including MT 103 and its variants and MT 202, MT 205 and their cover variants, for all SWIFT users originating payments, whether or not they are gpi members.

SWIFT gpi, the global payments innovation initiative, launched in 2017, is not a rail and not a message format. It is two things: a rulebook that participating banks sign up to, and a central tracker database. The rulebook commits members to same-day use of funds within the receiving bank’s business day, transparency of the fees deducted, unaltered transmission of remittance information, and end-to-end tracking. The tracker is a database, hosted by SWIFT, that every bank in the chain updates as the payment passes through, recording status, timestamps, charges taken and exchange rate applied. Because the UETR is unaltered and unique, all those updates from different banks in different countries attach to the same payment.

SWIFT’s own published performance figures are that nearly 60 per cent of gpi payments are credited to end beneficiaries within 30 minutes and almost 100 per cent within 24 hours, and that hundreds of the largest cash management banks send over USD 300 billion via gpi every day.

Around the tracker sits a set of services that turn visibility into action: case resolution, so investigations run as structured exchanges rather than free-format messages; stop and recall, so a payment can be chased down the chain if it was sent in error or under fraud; and pre-validation, so a beneficiary account can be checked before the payment is sent rather than bouncing days later.

Two sets of numbers that do not agree, and why#

Here is a genuine trap. SWIFT reports that almost 60 per cent of gpi payments are credited within 30 minutes. The Financial Stability Board reports, for 2024, that 50.6 per cent of cross-border wholesale payments were credited within one hour and 92 per cent within one business day, with retail at 33.5 per cent and 69 per cent and remittances at 53 per cent and 76.6 per cent. The 2025 progress report describes only a slight improvement at global level since the indicators were first calculated in 2023, with wholesale and remittance speed improving and cost remaining stubborn.

Both sets of numbers are honest. They measure different things. The gpi figure covers gpi payments, sent by banks that joined the scheme along corridors they serve, and the clock is the bank-to-bank leg. The FSB indicators are drawn from a wider population and are closer to end-to-end. If you are asked why your bank’s tracker says 30 minutes and the customer says three days, that gap is the answer, and it is mostly made of the domestic legs at each end, the cut-offs between them, and the compliance queue.

The G20 targets are the yardstick everyone is now measured against: by the end of 2027, 75 per cent of cross-border wholesale, retail and remittance payments credited or available within one hour and the remainder within one business day; a global average retail cross-border cost of no more than 1 per cent with no corridor above 3 per cent; and, restating the UN Sustainable Development Goal, a global average cost of sending a USD 200 remittance of no more than 3 per cent by 2030 with no corridor above 5 per cent. Against that last one, the World Bank’s Remittance Prices Worldwide put the global average cost of sending USD 200 at 6.49 per cent in the first quarter of 2025, with digital services at 4.85 per cent and non-digital at 7.16 per cent, and the global weighted average at 4.92 per cent. The distance between 6.49 and 3 is the whole of the reform agenda.

Sanctions screening, and where the days actually go#

Every institution in the chain screens every payment, and it screens the message, not the payment. This distinction is the source of most of the friction.

The lists screened against include, for a UK institution, the UK Sanctions List and the consolidated list of asset freeze targets maintained by the Office of Financial Sanctions Implementation; for anything touching dollars, the Specially Designated Nationals and Blocked Persons list and other lists maintained by the United States Office of Foreign Assets Control; the European Union’s consolidated list; and the United Nations Security Council consolidated list. Institutions add their own lists of high-risk entities, jurisdictions, and typologies on top.

The screening engine reads names, addresses, countries, BICs, and free-text fields, and applies fuzzy matching, because criminals do not helpfully spell their names the way the list does. Fuzzy matching generates false positives at a rate that would astonish anyone outside the industry, and every one of them is a stop. The payment is held, an analyst reviews it, and if the message does not contain enough information to clear the alert, the bank sends a request for information back up the chain. In ISO 20022 that is a camt.026 unable to apply or a camt.028 additional payment information, answered by camt.029; in the old world it was an MT 195 query and MT 196 answer. The round trip crosses time zones, and the answer often has to come from the originating customer, who is asleep.

Three consequences follow, and they are the real technical content of this section.

First, data quality is a speed feature. The single most effective way to make an international payment fast is to populate the beneficiary name and full structured address correctly. A truncated or free-text address is what turns a fifteen-second payment into a two-day investigation. This is the strongest practical argument for ISO 20022, and it is why the enhanced data mandates on CHAPS and elsewhere are framed around structured addresses, purpose codes and LEIs. Structured fields let the screening engine compare a town against a town rather than a hundred and forty characters of prose against everything.

Second, the cover method has a history. Plain MT 202 cover messages that omitted the underlying parties were used to defeat screening, deliberately, by stripping identifying data. MT 202 COV was introduced in the 2009 standards release specifically to force the underlying customer information into the cover message so that intermediaries screened the real parties. The enforcement that followed produced some of the largest financial penalties on record, including BNP Paribas’s guilty plea and approximately USD 8.9 billion penalty in 2014 for sanctions violations. The lesson institutionalised across the industry is that an intermediary is responsible for what passes through it, and cannot rely on the sender.

Third, the data requirements are about to get heavier. At its June 2025 plenary the Financial Action Task Force agreed revisions to Recommendation 16, the wire transfer rule, published on 18 June 2025 and intended to take effect by the end of 2030. They require enhanced and verified originator information and enhanced beneficiary information on cross-border payments. Above a threshold of USD or EUR 1,000, the message must carry the name and account number of both originator and beneficiary, or a unique transaction reference; for individuals, the originator’s address and the beneficiary’s country and town; and for legal entities, a BIC, a Legal Entity Identifier, or another unique official identifier. Card transactions for the purchase of goods and services remain outside the full requirements. Below the threshold the data set is smaller and need not be verified absent suspicion.

Read that list of required fields against the ISO 20022 message structure and the design intent becomes obvious. The standards migration and the transparency rules are the same project seen from two directions.

Sanctions as an instrument, and the network as a chokepoint#

Because SWIFT is a Belgian entity, it is bound by European Union law. When the EU has required it, SWIFT has disconnected named institutions from its network, most prominently a set of Russian banks from 2022. Be precise about what disconnection does. It does not freeze anyone’s money; balances in correspondent accounts are frozen, if at all, by asset-freeze measures applied by the banks holding them, not by a messaging network. What disconnection removes is the standard, automated way to instruct. A disconnected bank can still be paid by any institution willing to use telex, email, a bilateral link or another network, subject to whether doing so is lawful.

That is why alternative arrangements exist and are growing: China’s Cross-Border Interbank Payment System for renminbi, Russia’s domestic financial messaging system, regional systems such as Buna in the Arab region, and a lengthening set of direct links between national instant payment systems. None approaches SWIFT’s reach. All reduce the degree to which any single network is a chokepoint, which is precisely why some states are building them.

Where this is going#

The standards are converging. ISO 20022 is now the format for cross-border messaging, for Fedwire, for CHAPS, for T2 and for a growing list of domestic high-value systems. The same structured payment can, in principle, travel from a corporate ledger through a domestic rail, across a correspondent chain and out through another domestic rail without being flattened.

The service levels are hardening. SWIFT gpi set rules for wholesale payments; SWIFT Go extended a service level to low-value payments; and on 25 September 2025 SWIFT announced new rules for retail cross-border payments on its network, developed with over thirty banks from seventeen countries, committing to upfront transparency on costs, full value delivery, end-to-end visibility, and instant settlement where domestic infrastructure and regulation allow.

And the underlying model is being questioned. The Bank for International Settlements has framed the correspondent banking problem explicitly in terms of the cost and failure rate of the current chain, and Project Agorá, its work with central banks and private banks on tokenised deposits and central bank money on a shared programmable platform, is an attempt to collapse the sequence of separate ledgers into one. Whether that works is not yet knowable. What is knowable is that the friction it targets is real, measured, and has not improved much in a decade.

The Sheffield payment, explained#

The USD 214,000 never left the United States. Her bank’s dollar position sat as a claim on a New York correspondent, was debited there, moved across a domestic dollar clearing system to a second New York bank, and was credited to the Turkish bank’s dollar claim, also in New York. The steel mill was paid in Izmir out of the Turkish bank’s own balance sheet, against the increase in what New York owed it. SWIFT carried the instruction, in seconds, and knew nothing about any of the balances involved. It could not have told her where the money was, because at every instant the money was in New York and only the entitlement to it was moving.

The eighty-five dollars were three tariffed charges taken under a shared-charge instruction, each disclosed in a document she had not read, none of them visible at the point of sending. Had the payment been sent on terms where the sender bears all charges, the mill would have received the full amount and her own bank would have billed her more. Had the corridor been intra-European, the shared basis would have been mandatory and the deductions would not have happened.

The three days were not three banks taking a day each. They were one cut-off missed by hours, one time zone, and one sanctions filter that stopped on a fuzzy name match and needed a full structured address the original message had never carried, because thirty-five characters was not enough.

Every part of that is fixable. Most of it is being fixed, and the fix is not a faster network. It is better data in the message, one identifier that survives the whole journey, and an agreement among the banks about what they owe the customer. That is what ISO 20022, the UETR and the gpi rulebook respectively are, and it is why a chapter about a messaging network turns out to be a chapter about bookkeeping, deadlines and paperwork.

41.98 Common wrong ideas#

Wrong: SWIFT is the system that sends money abroad. Right: SWIFT sends messages, has no view of anyone’s balance, holds no accounts, and does not know whether the payment it carried actually happened.

Wrong: a bank’s dollars in New York sit in a pot of money that belongs to it. Right: they are an unsecured claim on the New York bank, recorded on that bank’s ledger, and if the New York bank fails its correspondent joins the queue of creditors.

Wrong: a payment that takes three days means three banks each took a day. Right: message delivery is measured in seconds, and the days are made of cut-off times, business calendars, value dating and a compliance queue.

Wrong: every hop in the chain automatically takes a fee out of the money. Right: the charge basis is a choice recorded in the message, and within the United Kingdom and the European Economic Area the shared basis is effectively mandated for the relevant currencies.

Wrong: the payment instruction always travels alongside the funds. Right: under the cover method the customer message goes straight to the beneficiary’s bank while a separate cover message moves the funds down the correspondent chain, and the two can arrive at different times.

Wrong: SWIFT is a synonym for international payment. Right: intragroup transfers, card networks, euro-area retail schemes, China’s and Russia’s own arrangements and money transfer operators’ closed loops all move cross-border value without a SWIFT message.

Wrong: disconnecting a bank from SWIFT freezes its money. Right: disconnection removes the standard automated way to instruct; balances are frozen, if at all, by asset-freeze measures applied by the banks that hold them.

Wrong: MT messages ceased to exist on 22 November 2025. Right: the coexistence period ended, and MT became a chargeable fallback from 1 January 2026, delivered through automatic conversion and lossy in-flow translation.

Wrong: the tracker says thirty minutes, so a customer complaining of three days must be mistaken. Right: the gpi figure covers the bank-to-bank leg of gpi payments, while the wider indicators are closer to end to end, and the gap is the domestic legs, the cut-offs and the compliance queue.

Wrong: SWIFT’s security record means the network has been broken into. Right: every significant incident, including the Bangladesh Bank fraud, has been an endpoint compromise using entirely valid credentials, which is why the Customer Security Programme is aimed at members’ own premises.

41.99 Chapter summary in 20 lines#

1. SWIFT is a Belgian messaging cooperative owned by its members, not a payment system: it has never moved money and holds no accounts.
2. An international payment is a sequence of separate domestic bookkeeping entries in different countries, stitched together by messages.
3. Correspondent banking means two banks holding accounts with each other, so that value is paid locally at each end while only the instruction crosses the border.
4. No bank can hold an account with every other bank, so payments run down a chain through a small number of large correspondents.
5. Dollars only really exist in America, so a dollar chain must pass through New York, and the dollars themselves never leave it.
6. What a bank “holds” at a correspondent is an unsecured claim on that institution, sitting inside another country’s legal system and reachable by that country’s authorities.
7. That is the mechanism by which one country’s sanctions policy binds banks with no other connection to it, and the largest single reason correspondent banking has become so cautious.
8. The eighty-five dollars missing from the Sheffield payment were three tariffed charges, each disclosed somewhere, taken lawfully under a shared-charge instruction.
9. Within the United Kingdom and the European Economic Area the shared basis is effectively mandated for the relevant currencies, which is why intra-European transfers rarely arrive short.
10. The three days were not three banks taking a day each; the messages were delivered in seconds.
11. Delay is made of cut-off times, business calendars, value dating and compliance screening, and none of them is cured by a faster network.
12. Screening reads the message rather than the payment, so truncated or unstructured data is what turns a fifteen-second payment into a two-day investigation.
13. The cover method splits the customer message from the funds message, a gap that was exploited by stripping names until MT 202 COV forced the underlying parties back into the cover.
14. MT is the 1977 flat-text format with thirty-five-character lines and silent truncation; ISO 20022 replaces it with structured, tagged and far longer data.
15. CBPR+ defines how ISO 20022 is populated for cross-border traffic over FINplus, and the coexistence period ended on 22 November 2025, leaving MT as a chargeable, lossy fallback.
16. The UETR is a thirty-six-character identifier that nobody in the chain may alter, which is what lets updates from banks in different countries attach to one payment.
17. SWIFT gpi is a rulebook plus a central tracker, not a rail: it made payments visible rather than fast, which turned out to be most of what customers wanted.
18. SWIFT’s thirty-minute figure and the Financial Stability Board’s one-hour figure are both honest, because they measure different populations over different legs.
19. The correspondent network has been shrinking for over a decade through de-risking while carrying ever more traffic, which concentrates the system and can leave a whole country harder to pay.
20. Every international payment ends as a domestic settlement in the currency’s home system, and the fix for its friction is better data in the message, one identifier that survives the journey, and an agreement about what the banks owe the customer.

Sources: Swift, on FIN traffic figures, organisation, governance and oversight, ISO 20022 and CBPR+ implementation, the UETR, gpi, and the retail cross-border rules announced 25 September 2025; National Bank of Belgium and Deutsche Bundesbank on the cooperative oversight of Swift; BIS Committee on Payments and Market Infrastructures correspondent banking data commentaries and chartpacks, and BIS Bulletin No 87; Financial Stability Board G20 targets and the 2024 and 2025 progress reports; World Bank Remittance Prices Worldwide, Issue 53, March 2025; Federal Reserve Financial Services on the Fedwire Funds Service ISO 20022 migration of 14 July 2025; Bank of England on ISO 20022 and CHAPS enhanced data; FATF, update to Recommendation 16, June 2025; ISO 20022 Registration Authority message catalogue.