SWIFT gpi and Cross-Border Payments: Speed, Tracking, Friction

SWIFT gpi added end-to-end tracking, fee transparency and same-day settlement to correspondent banking. Here is what improved, what still breaks, and what it costs.

2026-11-06 · Updated 2026-11-07 · 8 min read

Key takeaways

  • Every gpi payment carries a unique end-to-end transaction reference, giving treasury systems one key to follow a payment
  • Median end-to-end processing time on gpi is under two hours, but the average is far worse because a minority of routes take days
  • ISO 20022 replaced MT for cross-border payment instructions on 22 November 2025, and new tracking features now appear in ISO 20022 and API formats only
  • Lifting fees, FX spreads, cut-off times, compliance holds and beneficiary-side processing are the friction that tracking cannot remove

Why correspondent banking needed fixing

A cross-border payment rarely moves from bank A to bank B. It hops through a chain of correspondents: your bank sends a message, an intermediary debits its nostro account, a second intermediary screens the payment again, and a local clearing system in the beneficiary's country finally credits a domestic bank that pays the customer.

Every hop historically added a black box. The sending corporate knew the payment had left, but not where it was, when it would be credited, or how much each intermediary would deduct on the way. Investigations were run by phone and email, often for several days, and the answer frequently arrived after the beneficiary had already chased the payment.

SWIFT gpi, launched in 2017, was the industry's attempt to replace that opacity with a service level built on four principles: settlement within the same business day in the receiving time zone, near real-time status monitoring, transparency of fees charged along the chain, and unaltered remittance information reaching the beneficiary bank. The Bank for International Settlements' CPMI research on Swift gpi data summarises the design and measures whether it worked.

What SWIFT gpi actually changed

One reference for the whole journey: the UETR

Every gpi payment carries a unique end-to-end transaction reference (UETR) in the message header. It is not a bank's internal reference and it is not regenerated at each hop; it travels with the payment from the ordering bank to the beneficiary bank.

For treasury teams this is the single most practical change. Instead of reconciling three or four different bank references for the same payment, the UETR becomes the join key across your ERP, your bank portals and your reconciliation files. Any system that stores a payment identifier should store the UETR alongside it.

Tracking without calling the bank

The gpi Tracker receives status updates from each institution handling a payment and reports them on demand. Banks expose this through the gpi for Corporates API, which supports two connection models: a tracker-to-bank mode where SWIFT provides the link between the Tracker and the bank, and a bank-to-corporate mode where the bank relays tracking data through its own API platform. Corporates can query by UETR or by time window, and can subscribe to notifications for incoming payments as well as outgoing ones.

That inbound notification capability is underused. Many corporates use tracking to answer "where is my payment?" questions on outbound flows, while the bigger operational win is often automating receivables matching from structured inbound advice data.

Confirmation of credit

Tracking tells you where a message is. Confirmation of credit tells you the money arrived. Beneficiary banks on gpi confirm whether a payment has been credited, transferred to a domestic system, or placed on hold. Swift's own reporting, based on network data from a second-quarter snapshot, found that 92% of cross-border gpi payments were credited within 24 hours and 40% within 30 minutes, with 56% of payments by value confirmed to the beneficiary.

Fee transparency

Charge codes decide who pays. Under OUR the sender covers deductions and the beneficiary receives the full amount; under SHA the beneficiary absorbs intermediary deductions; under BEN the beneficiary pays all charges. gpi rulebooks require intermediaries to report deductions in the relevant charge field, so a payment sent under SHA can be reconciled after the fact. Transparency, however, is not the same as predictability: if you want a known amount to arrive, you usually need an OUR-style instruction or a no-deduct service.

Stop and recall

Using the UETR, a bank can route a stop-and-recall request through the Tracker to the institution currently holding the payment. Swift's Stop and Recall API documentation states that the receiving institution is expected to process the request within 24 hours, and that if funds have already been credited, the beneficiary returns them in line with existing market practice. It is a coordinated request process, not a reversal right.

How fast is fast, in numbers

Speed claims in cross-border payments are notoriously sensitive to what is being measured. The BIS analysis, based on roughly 20 million gpi payments with credit confirmations, separates processing time as follows.

Measure Value Source
Median end-to-end processing time 1 hour 38 minutes BIS CPMI gpi analysis
Average end-to-end processing time 8 hours 36 minutes BIS CPMI gpi analysis
Median time on the fastest corridors under 5 minutes BIS CPMI gpi analysis
Median time on the slowest corridors more than 2 days BIS CPMI gpi analysis
Payments credited within 30 minutes 40% Swift gpi report

The gap between the median and the average is the story. Most payments are genuinely fast; a minority of routes are slow enough to drag the average to over eight hours, and those routes are often the ones involving capital controls, batch-based domestic clearing or additional compliance stops. More recent network data points the same way: Swift reports that 75% of cross-border payments reach beneficiary banks within ten minutes. Note the different boundary, which is the point worth internalising: ten minutes to the beneficiary bank is not the same as ten minutes until the corporate's account is credited. The G20 target is defined on funds availability, not message delivery: 75% of payments available to the recipient within one hour by end-2027.

ISO 20022 changed the plumbing in November 2025

For most of gpi's life the underlying messages were MT formats: MT 103 for customer credit transfers, MT 202 COV for cover payments, MT 199 for tracking queries. That period ended. The coexistence window between MT and ISO 20022 for cross-border payment instructions closed on 22 November 2025, with Swift reporting that 97% of payment instructions were already being sent in ISO 20022 on the first business day after the switch. Remaining in-scope MT instructions are converted by a contingency service, and the community roadmap notes that charges and additional validation apply to that processing, with further message types moving in later waves.

For tracking specifically, MT 199 and MT 299 still function but are deprecated, and Swift's guidance is that new features will only appear in ISO 20022 tracking messages and APIs. That matters for anyone building now: a tracker integration written against MT 199 is a legacy implementation on day one.

The upside of structured data is not cosmetic. Structured addresses, purpose codes and richer party data reduce repair rates, improve the precision of sanctions screening, and make straight-through processing possible for reconciliation systems that previously had to parse free-text fields.

What still creates friction

  • Cut-off times and time zones. Even a payment that clears in seconds has to wait for the next window in a batch-based domestic system. A Friday afternoon instruction in Europe can sit until Monday in the beneficiary's market.
  • Beneficiary-side processing. The last mile is a bank deciding when to credit a customer account. Tracking stops at the edge of that decision.
  • Non-gpi participants at the end of the chain. One legacy agent without tracker connectivity breaks visibility for the whole route.
  • Compliance holds. Sanctions and AML screening can hold funds indefinitely, and value dates are lost while an investigation runs. This is the most common reason a "fast" payment is not fast.
  • Fees embedded in the FX rate. A payment can look free while the conversion margin carries most of the cost. Fee transparency under gpi covers correspondent deductions, not the FX spread applied by the sending or receiving bank.
  • Cost at the retail end. The FSB's 2025 progress report on the G20 roadmap puts the global average cost of sending a USD 200 transfer at about 6.5% of the amount sent, against a G20 target of 3% by 2030, with digital channels averaging around 4.7% versus 6.8% for cash-based services.
  • Recall limits. Stop and recall reduces the damage from fraud and error, but once funds are credited the outcome depends on the beneficiary bank and its customer.

What this means for treasury and fintech builders

Use the UETR as a primary reconciliation key, not a bank reference. Query status through APIs rather than portals where volume justifies it, and consume confirmation-of-credit events rather than treating dispatch as settlement. Be explicit in payment templates about charge codes, because OUR, SHA and BEN produce materially different amounts received.

For fintech developers, the practical work is in three places: ingesting tracker statuses and mapping the status codes correctly; building reconciliation around structured ISO 20022 fields while they are rich rather than after translation loss; and treating beneficiary account validation as a first-class control, since a returned payment costs more than a slow one. Where a payment value is under USD 10,000, low-value schemes such as Swift Go are designed as a simpler alternative to a full correspondent payment, with defined fee and timing conditions.

Orchestration decisions sit on top of all of this. Choosing between a correspondent route, a domestic instant rail, or a local collection account is a routing and cost question as much as a messaging question, and it is worth comparing rail economics before designing the integration — see our guides to payment rails and payment orchestration platforms.

Risks and costs

Transparency has a price, and it is not only the subscription. Institutions pay for gpi membership and tracker connectivity, for ISO 20022 programme work, for API integration and, after November 2025, potentially for contingency conversion of in-scope MT traffic. Corporates pay for bank implementation projects and for the operational change of wiring tracking into treasury systems.

The risks sit in several buckets. Operational risk: tracking changes expectations, and a message that is "in progress" may still be held for compliance, which team processes must be able to explain. Counterparty risk: stop and recall depends on another institution's cooperation and on a beneficiary customer's willingness to return funds. Data risk: structured ISO 20022 fields have to be populated correctly by upstream systems, or the benefit disappears into empty optional fields. Concentration risk: most cross-border messaging runs over one network, so access, pricing and outages are single-point dependencies for a treasury function. Regulatory risk: sanctions screening, capital controls and local reporting requirements differ by jurisdiction and cannot be assumed to work the same way in the EU, the US or an emerging-market corridor.

None of this is investment advice, and none of these products is a recommendation: they are infrastructure choices with different cost structures, availability by jurisdiction, and failure modes.

Bottom line

SWIFT gpi solved the information problem more completely than the speed problem. Payment status is now queryable, fees along the chain are reported, credit confirmation is standard, and the UETR gives treasury one identifier to reconcile on. What it cannot fix is everything outside the messaging layer: correspondent cut-off times, FX margins, compliance holds and the beneficiary bank's own processing decide when money actually lands. Corporate treasury teams should build reconciliation and cash forecasting on confirmation of credit, not on dispatch, and developers should assume the ISO 20022 and API world as the baseline rather than the migration target.

This article is for information and education only. It is not investment advice and does not take into account your personal situation. Read the full disclaimer.

Follow the shortlist

Following payments, banking and market-infrastructure change? Our global ETF and stock research watchlist is tracked daily. Premium members get the full list, the reasoning behind every entry and updates when entries change.

See premium access

Information and education only, not investment advice.

Related