Streamline your transactions with advanced automation solutions that save time and reduce errors.
A cross-border payment settles by moving through a chain of banks that hold accounts with one another. No money crosses a border. Each institution in the chain debits and credits accounts on its own books, and the payment arrives when the last institution in the sequence credits the beneficiary.
That is how cross-border settlement works, and it explains most of what is wrong with it. Cost accumulates because every institution in the chain is a commercial party with its own fee schedule. Delay accumulates because each institution settles its leg independently — and, contrary to the usual explanation, most of the waiting happens at the final institution rather than in transit.
This post traces one payment through the chain, identifies the four places cost actually appears on an institution's P&L, and sets out the five routes available to an institution that wants a different arrangement.
A correspondent bank is a bank that holds an account and provides payment services for another bank located abroad. The Bank for International Settlements' Committee on Payments and Market Infrastructures defines it as a bank providing local account and payment services for banks based abroad, which collectively form the correspondent banking network.
The account is described from both sides. Your institution calls the balance it holds abroad a nostro account — "our account with you". The bank holding it calls the same balance a vostro account — "your account with us". Nothing else is required for a cross-border payment to work: two institutions, one account, and an agreed way to send instructions.
The difficulty is that no institution holds an account with every other institution. When your correspondent has no relationship with the beneficiary's bank, the payment must pass through one that does. That is where the chain forms.
Take a UK-licensed EMI paying a beneficiary at a commercial bank in Nigeria.
1. The instruction. The EMI does not hold naira. It holds a US dollar balance in a nostro account at its correspondent, and instructs that correspondent to make the payment.
2. The first leg. The correspondent debits the EMI's nostro balance. That debit is the settlement of the first leg. It happens on the correspondent's books, in London or New York, and the money has not moved anywhere.
3. The intermediary. If the correspondent holds no account relationship with the Nigerian bank, it routes the payment through a bank that does — typically a large US dollar clearer. A second leg settles, again as a book entry.
4. The destination institution. The intermediary credits the Nigerian bank's own dollar nostro account. At this point the funds are, in an accounting sense, in Nigeria's banking system. The beneficiary still has nothing.
5. The final leg. The Nigerian bank converts dollars to naira at its own rate and credits the beneficiary through the domestic payment system.
Five steps, three or four institutions, and each leg settled and screened independently. The Federal Reserve Bank of Cleveland describes the consequence plainly: each party must settle its leg in compliance with local law before instructing the next party in the chain, producing a sequential process in which fees are deducted and delays introduced at multiple points, and no single participant sees the whole path.
That last part matters more than it sounds. The CPMI's own dataset notes that payment chains cannot be identified from payment message data. If the international standard-setter cannot reconstruct the route from the messages, neither can you.
The standard explanation is that cross-border payments are slow because of the number of intermediary hops. The data suggests something more specific.
The Financial Stability Board's 2023 report on key performance indicators split wholesale payment time into two parts: the in-flight leg across the messaging network, and the beneficiary leg after the payment leaves it. It found that 53.8% of payments using Swift completed both legs within one hour and 92.7% within one day — with most of the elapsed time occurring in the beneficiary leg. Measured separately, only around 60% of wholesale payments were credited to customer accounts within one further hour after leaving the network.
The message is not what is slow. The stops are.
The reasons are operational rather than technological:
One qualification worth knowing, because it makes the published figures flattering. The FSB's speed measurement corrects for institutions' offline hours by excluding weekends and public holidays in the relevant country. Real elapsed time, as a client experiences it, is worse than the KPI.
The FSB's consolidated progress report for 2025 records that the global speed of wholesale payments has improved. It also states that the 2025 indicators show only slight improvement since 2023, and that it is unlikely satisfactory improvement will be achieved in line with the 2027 timetable.
Cost in this system is not one number. It appears in four places, and only the first is visible on a statement.
Each bank in the chain may deduct its charge from the principal rather than bill for it. This is a lifting fee — a charge taken out of the payment itself by an intermediary. The originating institution often learns the total only when the beneficiary reports a shortfall, which is why reconciliation queries are a recurring operational cost in their own right.
Whoever performs the conversion applies their own rate. This is typically the largest single component and the least visible, because it is expressed as a rate rather than a fee. Where the conversion happens at the destination institution, the originating institution has no control over it and frequently no advance sight of it.
To settle in a currency, an institution must hold a balance in that currency before any payment is made. Multiply that across corridors and currencies and a material share of working capital sits in accounts that exist only to make settlement possible. It earns little, it cannot be deployed, and it scales with the number of markets served rather than with revenue.
A caution here, since the figures circulate freely. A widely repeated estimate of the total capital held globally in nostro and vostro accounts is attributed inconsistently across published sources, with materially different numbers appearing under the same attribution. We have not been able to trace it to a primary source and so do not repeat it. The mechanism is real and quantifiable for your own institution; the industry aggregate is not reliably established.
Your correspondent is required to understand your business, your customers and in some cases your customers' customers. Meeting that requirement is a standing cost: due diligence questionnaires, periodic reviews, transaction monitoring responses, and the internal capacity to answer them at the speed the correspondent expects. The BIS has noted that the current structure produces duplication of effort in AML screening across institutional silos. Every institution in the chain screens the same payment, and every institution charges for the capability that lets it do so.
This is the cost line most often omitted from vendor comparisons, and for a mid-sized institution it is rarely the smallest.
Fewer correspondent relationships would be an improvement if it meant shorter chains. It has generally meant the opposite: fewer available routes, longer paths through third countries, and greater concentration.
The CPMI tracked this for several years using Swift message data. Its published commentary recorded a decline in active correspondent banking relationships of about 20% over seven years alongside a fall of roughly 10% in active corridors, with the contraction universal but uneven — the Americas excluding North America down around 30% since 2012 against roughly 10% for North America. The BIS later put the cumulative contraction at approximately 25% between 2011 and 2020, while transaction volume and value continued to rise. Fewer relationships carrying more traffic is the definition of concentration.
One point of precision, because it is widely got wrong. The CPMI's annual quantitative review concluded with the publication of the 2022 data, completing its commitment to the FSB. There is no official series running to the present. Figures presented elsewhere as current correspondent banking decline data are extrapolations from a dataset that stopped, and should be read as such.
An institution unhappy with its correspondent arrangement has five options. None is universally right.
Route
What changes
Honest trade-off
Renegotiate with your correspondent
Pricing and service levels on the existing relationship
Cheapest and fastest to attempt. Leverage depends entirely on your volume, and it does not shorten the chain or add a corridor
Marketplace or aggregator
One integration, many underlying providers, with comparison across them
You remain a buyer selecting suppliers. Coverage is inherited from the panel, and pricing improves through competition rather than through structure
Direct provider
A single counterparty handles the corridor end to end
Simple and often materially faster. Concentration risk sits with one provider, and some providers also serve your client type directly
Operate under another institution's licence
Market access without your own authorisation
Fastest route into a regulated market. You are a tenant: your permissions, and your continuity, are someone else's
Obtain your own licence
Direct access to the destination market's payment system
Highest control, highest cost. Capital requirements, local presence and an ongoing supervisory relationship, on a multi-year timeline
Join a settlement network
You settle directly with other member institutions rather than through a chain
Value depends on whether members already cover your corridors. A network with no member in your destination market is of no use to you there
Each of these deserves its own treatment, and we will publish one.
Some providers in the second and third categories settle over regulated stablecoin rails rather than through correspondent accounts. That is a change of plumbing, not of commercial position: the institution is still buying a service.
The five routes above share one feature. In each, the institution is a purchaser — of better pricing, of coverage, of permissions, of access. Money moves in one direction.
UNYX, the cross-border settlement network for regulated financial institutions, is built on the opposite proposition: that a licensed institution should be a correspondent within the system rather than a client of it. When another member settles a payment into your market, that payment routes through you, and you earn on volume you did not originate. UNYX is not a blockchain, a cryptocurrency or a crypto exchange; only licensed financial institutions are admitted, and every member is licensed and KYC/AML checked before it can settle a transaction.
A network is the wrong answer if your volume is concentrated in a single corridor where you already hold a direct, well-priced relationship. It is the wrong answer if your problem is a specific compliance finding rather than a structural one — no change of rail fixes an unresolved supervisory issue. And it is the wrong answer if the markets you need are not markets other members serve, because a settlement network is only as useful as its membership at the point you need it.
Copyright ©️ 2026 UNYX AI LTD. All rights reserved.
UNYX AI CA LTD (Ontario Corporation No. 1001253500), a wholly owned subsidiary of UNYX AI LTD (United Kingdom, Companies House No. 16786021), is registered with the Financial Transactions and Reports Analysis Centre of Canada (FINTRAC) as a Money Services Business under registration number C10001322 for foreign exchange dealing, money transferring, and dealing in virtual currency. UNYX AI CA LTD is also registered with the Bank of Canada as a Payment Service Provider under the Retail Payment Activities Act.
UNYX AI CA LTD is not a bank. It does not accept deposits and funds held are not insured by the Canada Deposit Insurance Corporation. Services are provided to regulated financial institutions and are not available to retail consumers.
Virtual assets are not legal tender in Canada, their value can fluctuate, and they are not covered by deposit insurance or investor protection schemes.
Registered office: 251 Consumers Road, Suite 1200, North York, Ontario, M2J 4R3, Canada.
UNYX AI operates a network of independently licensed and regulated financial institutions. UNYX AI itself is not a bank, e-money institution, or payment institution, and does not hold client funds. Products and services made available through the network are provided by member institutions under their own regulatory licenses in their respective jurisdictions and may not be available in all regions.
.svg)