Section 02
The rails
A cross-border payment is assembled from parts, and the parts are bought in different markets at wildly different prices. This section prices each of them. It is the evidence floor everything later stands on, and it is where a sceptical reader should attack first.
What we mean by a rail
A rail is a system that actually moves money between accounts. Some are national and run by a central bank — Mexico's SPEI, Brazil's PIX, the euro area's TIPS, the UK's Faster Payments. Some are commercial networks that connect banks across borders, of which the correspondent banking network reached over SWIFT is by far the largest. Some are card networks running payments backwards, from business to consumer. And one is a public blockchain carrying a token that claims a fixed value against a currency.
They are not substitutes for one another. Each does part of the job, and the practical question for anyone assembling a payment is which combination to use. This section takes each in turn and asks the same three questions: what does it cost, how long does it really take, and what breaks it.
That figure contains the single most important fact in this research, and it is worth stating before any of the detail. The four jobs do not cost remotely similar amounts. Collecting and paying out are close to free. Moving value across the border is cheap. And the currency conversion costs almost nothing to produce — while being the thing customers are charged for. The last two rows of that chart are the same job, priced at wholesale and priced at retail.
Correspondent banking: half the reputation is now wrong
Correspondent banking is the default. A payment leaves your bank, passes through one or more banks that hold accounts with each other, and arrives at the recipient's bank. It is the oldest mechanism here and the one everything else is compared against — usually on the assumption that it is slow and expensive.
Half of that assumption no longer survives: The premise is half dead and half alive, and the live half is precisely locatable. Anyone still quoting a three-to-five-day baseline for a major corridor will be corrected in the room, because the bank-to-bank portion has genuinely been fixed.
But the published figures measure something narrower than they appear to, and the gap is not subtle.
The headline speed numbers are measured at a boundary that excludes the problem — and the same source publishes both numbers. Ninety per cent of payments reach the destination bank within an hour. Forty-three per cent reach the destination account. The difference is the last mile — beneficiary bank processing, compliance review, and the recipient's own bank deciding when to post the credit — and it is roughly four fifths of the journey.
There is a second exclusion, and it is structural rather than presentational. Every published speed statistic in this domain excludes weekends and public holidays by construction. The clock is measured in business time. A payment entering the Easter window arrives five calendar days later and still counts inside the same statistic, because the intervening days are not counted at all. For anyone selling predictability, this is the most useful fact in the section: the industry's own numbers do not measure the thing that goes wrong.
Domestic instant rails: the last mile is already solved
Most countries now run a national instant payment system — money moves between domestic accounts in seconds, at any hour, for a fraction of a cent. Brazil's PIX, India's UPI, the UK's Faster Payments, the euro area's SEPA Instant, and Mexico's SPEI are all of this kind.
Domestic instant rails cost fractions of a cent and settle in seconds, and no stablecoin construction beats that once the off-ramp is included. This closes off a whole category of argument. Whatever route a payment takes across the border, it finishes on one of these systems — and no alternative mechanism can undercut a rail that charges two-thousandths of a euro and completes in under five seconds.
So the territory genuinely available to any alternative is the crossing and the conversion. Never the domestic delivery at either end.
The distinction that breaks naive routing
These systems run continuously. The money that funds them does not. An instant payment system settles from a pre-positioned balance, and that balance is topped up through the central bank's settlement system — which observes business days like any other bank.
A payment engine that treats "the rail is up" as "the rail is fundable" will strand payments at six o'clock on a Friday. Over Easter, the euro settlement system closes for four consecutive days.
One correction worth carrying into any conversation, because getting it wrong is checkable and expensive: The always-on claim is half-true in a specific and exploitable way. SEPA Instant genuinely is continuous at scheme level. Claiming to be always-on in contrast to it is simply false, and any informed counterparty will know.
The conversion: where the money is
Converting one currency into another is a wholesale market of enormous depth. The cost of doing it, at institutional size in a major currency pair, is negligible.
Manufacturing a major-currency conversion costs under one basis point at wholesale while customers pay between 52 and 400. That is a range of roughly two hundred to one. It is not explained by cost, by risk, or by technology — all three are close to identical across the range. It is explained by distribution: who the customer is, what they know, and what else they could realistically do.
Which produces the finding that reorganises the whole problem: Routing across liquidity providers is a feature; routing across rails and counterparties is the product. Choosing between banks offering to convert euros into dollars saves a fraction of a basis point, because that market is efficient. Choosing between the firms who will deliver pesos in Mexico saves considerably more, because that market is not.
How much more — and how we know
The method matters here, so it is worth stating exactly rather than impressively. Nobody approached providers and asked them to quote. We read Wise's public comparison API, which returns a self-curated set of competitors and their pricing — pulled across 27 corridors at five ticket sizes on one morning, with the fee and the exchange-rate markup reported separately.
On that basis: The dispersion is real, it is bigger than expected, and it is now evidenced corridor by corridor — and, more usefully, Almost all of that dispersion is hidden in the exchange rate rather than the payout fee. A provider advertising a 0.3% fee can be several hundred basis points more expensive than one advertising 1%, because the fee is not where the money is taken.
Three limitations follow from that method, and they are not small.
Wise curates the list, and Wise is on it. It appears in every corridor it compares. The set is not obviously slanted — in several Caribbean corridors Wise is the most expensive non-bank shown — but it remains one competitor's account of the others, which is a secondary source however live the prices are.
How many firms Wise lists is not how many firms exist. Nigeria licenses 108 international money transfer operators. Three appeared. The prices observed are real; the supplier counts understate the market, in that case by a factor of thirty-six.
And it lists firms in places they do not serve. A later audit found roughly 7% of the corridor rows named a provider that its own published coverage list excludes — with five of the six errors being Wise's own rows — an error that inflates apparent competition rather than deflating it. The price spreads survived that correction; the supplier counts became an upper bound; and one corridor turned out to have no non-bank supplier at all. Section 04 records what changed.
The chain-settled route
A stablecoin is a token on a public blockchain that claims a fixed value against a currency, usually the dollar. The proposition is that value can be moved across a border in seconds, at any hour, for cents — bypassing correspondent banking entirely.
The crossing itself does exactly what it claims: The chain hop is genuinely free and genuinely fast — and it is also the part nobody was charging for. That is the difficulty. The expensive parts of a cross-border payment are the conversions at each end, and those still have to happen — euros into the token, the token into pesos.
For a business starting in euros, the arithmetic is worse still, because the deep token markets are dollar-denominated: The euro leg is the binding constraint, and it is far worse than anyone admits. Measured live, the euro-denominated token market is economically capped at one to two million per transaction and physically exhausts around five — with no deeper venue behind it.
Which forces a specific architecture: The correct architecture is therefore dollar-stablecoin-only, with the euro conversion done in the fiat market. The token route does not remove the currency conversion. It relocates it — and for a euro-origin business it can mean crossing the euro-dollar pair twice.
Card payout: ruled out by the rulebook
The card networks run a mechanism for pushing money to a card rather than taking it — used for insurance settlements, gig-economy earnings and marketplace payouts. It reaches recipients who have a card but no reachable bank account, which is a genuine advantage in some markets.
For business-size payments it is disqualified twice over. Push-to-card is a consumer-disbursement rail with a hard scheme-imposed ceiling of tens of thousands per transaction — against instant payment systems that carry ten million. And The business exclusion is not only about size — it is written into the acceptance rule: the card type a business recipient is most likely to hold is the one no issuer is obliged to accept a payout on.
It retains one genuine architectural advantage, and it is the only rail here that has it: The one genuine architectural advantage is that push-to-card does not require prefunding. Every other route requires money positioned in advance. This one settles between the institutions afterwards — which means the working-capital problem that dominates the rest of this research is a property of particular rails rather than of cross-border payments as such.