For as long as anyone has measured it, the trade finance story has had a reliable villain. Small and medium exporters get turned down; large corporates get funded. Fix the asymmetry and you close the gap.
The most recent survey data says the asymmetry has very nearly gone. The gap has not moved at all.
The gap that closed, and the one that did not.
The Asian Development Bank published its Global Trade Finance Gap Survey in December 2025, built on responses from more than a hundred and ten trade finance providers. Two findings sit in it that are rarely read together.
The first: for the first time in the life of the survey, the rejection rate for small and medium enterprises, at around forty-one per cent, has fallen to almost the same level as the rate for large and mid-cap corporates, at around forty per cent. The SME figure was about forty-five per cent in 2023. On the headline measure, the penalty for being small has nearly disappeared.
The second: the global trade finance gap in 2025 was about 2.5 trillion US dollars — unchanged from 2023. As a share of global trade it moved from roughly 10.6 per cent to roughly 10 per cent, which is a rounding difference in a number of that size.
So the discrimination in the approval decision has largely closed, and the quantity of trade that cannot get financed has not responded. That should be uncomfortable, because it means the approval decision was not the binding constraint. Something else is, and it is not being counted.
A rejection rate only counts the applications that got made.
A rejection rate is a ratio. It requires a numerator of refusals and a denominator of applications. It is silent about everyone who never entered the denominator.
That silence is where the interesting population lives. An exporter who has been turned down twice does not apply a third time. An exporter whose freight forwarder tells them the paperwork will take six weeks does not start. An exporter in a market where the last bank offering the product withdrew two years ago has nobody to apply to, and therefore cannot be rejected by anyone.
None of those three appear as a refusal. All three appear in the gap.
What the forty-one per cent leaves out
Read the two ADB numbers together and there is only one consistent reading. If rejection rates converge while the unmet demand stays flat, the composition of applicants must have changed — the marginal applicant who used to be refused is now not applying, or not reachable.
A bank that stops serving a segment entirely will report an excellent approval rate for that segment. It has no applications from it. The metric improves precisely as the service disappears. This is not a criticism of the banks in the survey; it is a property of the ratio.
The ADB is careful here, and it is worth repeating their caution rather than running past it: the survey says the convergence finding requires more research. It is a first observation, not a settled fact. But even taken at its weakest, it fails to explain a gap that did not move.
The banks left the corridor before the exporters did.
The other half of this sits in a dataset almost nobody in trade reads, maintained by the Committee on Payments and Market Infrastructures at the Bank for International Settlements. It counts correspondent banking relationships — the arrangements by which a bank in one country can settle and move money in another.
No correspondent relationship, no route. A trade finance product is a promise to pay across a border, and a promise that cannot be settled is not a product.
The BIS commentary published on 31 August 2020 reported that the number of active correspondent banks worldwide fell by about twenty-two per cent between 2011 and 2019, and by about three per cent in 2019 alone. A later commentary in the same series carries the title “the decline continues at a slower pace”. The direction has not reversed in a decade and a half.
Where the network thinned fastest
The average conceals the part that matters. Across that 2011 to 2019 period the decline ran from about thirteen per cent in Northern America to about thirty-four per cent in Latin America. Small island developing states lost about forty-one per cent of their active correspondents.
That distribution should be read next to the trade finance gap, because it is the same map. The places that lost the most banking connectivity are the places where the unmet demand concentrates. An exporter in a market that has shed forty per cent of its correspondent relationships is not facing a tougher credit committee. They are facing a shorter list of institutions that can reach their buyer at all.
There is a geographic overlap here that the difference in vintages tends to hide. The trade finance gap survey is run by the Asian Development Bank, an institution whose mandate is Asia and the Pacific, and the correspondent category that thinned most severely — small island developing states — sits substantially inside that same region. The two datasets are not describing unrelated worlds. They are, in significant part, describing the same exporters.
Volumes went up while relationships went down
The counter-intuitive part, and the one that keeps this from being a simple story of decline: over the same period the volume and value of cross-border payments kept growing. Both rose by roughly four to five per cent in 2019.
More money moving through fewer doors. The BIS reading is that flows are concentrating, and that some of the apparent loss reflects payments being routed through a third country rather than directly. Money is still getting there. It is getting there through a narrower set of intermediaries, and via longer chains.
Concentration is not access.
It is tempting to treat concentration as efficiency. Fewer, larger, better-supervised correspondents handling more volume is, in several respects, a healthier network than a long tail of thinly monitored ones.
For a large corporate that is broadly true. They have relationships with several global banks, and if a corridor closes at one they move the flow to another. Concentration costs them a negotiation.
One correspondent is a price, not a market
For a small exporter it is different in kind, not degree. When the number of institutions that can serve a corridor falls to one, the exporter has not lost some options. They have lost the ability to be quoted. Whatever that single institution charges, and whatever documentary standard it chooses to apply, is now the market.
This is the mechanism that converts a connectivity statistic into a financing gap. The application is approved. The pricing makes the trade uneconomic, or the documentary requirement takes longer than the shipment cycle allows. The exporter withdraws. No refusal is recorded anywhere, and the trade does not happen.
It is also why the gap can sit flat while approval rates improve. Both facts are consistent with a market that is getting easier to enter and harder to reach.
The asymmetry compounds, too. A large corporate that loses a corridor moves volume it already has to a bank it already knows. A small exporter that loses a corridor has to establish a new relationship from nothing, in a market where the remaining institution has no particular reason to want a low-value, document-heavy account. The same event costs one party a phone call and the other party the trade.
What digitisation actually moves.
Most of what our industry builds, ours included, attacks the cost of establishing facts. Electronic documents, verifiable transaction records, structured data in place of scanned paper, a counterparty history that can be checked rather than asserted.
That work is real and it does something specific: it lowers the cost of underwriting a given transaction. When the evidence is cheap and machine-readable, a file that was uneconomic to assess becomes economic to assess. The ADB survey is consistent with this — eighty-four per cent of the banks polled already use artificial intelligence for fraud prevention and risk analysis, and fifty-seven per cent are exploring whether it can expand financing capacity rather than merely police it.
Evidence gets cheaper. Presence does not.
But cheaper evidence only helps an applicant who has somewhere to apply. No amount of document quality creates a correspondent relationship. No verifiable trade record persuades a bank to re-enter a jurisdiction it exited for reasons of compliance cost, sanctions exposure or simple scale.
These are different problems with different owners. The first is an information problem, and technology is genuinely good at it. The second is a network and regulatory problem, and it is largely decided by supervisors, by the economics of compliance, and by whether anyone will carry the residual risk of a thin corridor.
Conflating the two is the standard error in digital trade marketing, and it produces the familiar disappointment: a corridor gets excellent documents and still cannot get paid.
What we are building, and what it does not fix.
We build the evidence layer. Our work is to make a transaction between two parties who do not know each other legible enough to be financed — documents that are native rather than scanned, a record of performance that a third party can verify, and enough structure that assessing a small shipment does not cost what assessing a large one does.
We are honest about the boundary. That reduces the cost to underwrite. It does not put a bank into a corridor that has none, and it does not restore a correspondent relationship that a compliance committee ended. When we work in markets where the banking network is thin, the constraint we hit is not document quality. It is that the number of counterparties able to settle the leg is small, and we have to build around that rather than pretend we have solved it.
The honest counterweights.
Several things cut against the argument above, and they should be stated rather than buried.
The datasets are different vintages. The correspondent banking figures quoted here run to 2019; the trade finance gap figures are from the 2025 survey. Trends from one period do not automatically explain a number from another, and a careful version of this argument would want the two series aligned.
The rejection-rate data is self-reported by providers, and a bank’s definition of an application is not standardised. Some count an enquiry, some only count a completed file. Convergence could partly be a definitional artefact.
Correspondent decline is not purely withdrawal. Consolidation, the retirement of duplicate relationships and migration to more efficient routing all reduce the count without reducing access. The BIS itself allows for payments moving via third countries.
And the gap holding flat at 2.5 trillion dollars while global trade grew means it shrank slightly as a share, from about 10.6 to about 10 per cent. That is not nothing. It is simply far less than the convergence in approval rates would have led you to expect.
The number to ask for.
If you run exports, your bank will tell you its approval rate, and that number will probably be improving. It measures the applications it agreed to look at.
Ask for two different numbers instead. First: how many institutions could actually settle a transaction with a buyer in your main destination market — not how many have a branch there, how many could take the leg. Second: how that count has changed in the last five years.
If the answer to the first is one, you do not have a financing problem that better documents will fix. You have a single supplier, and the price you are quoted is not a market price. If the answer to the second is that the count has fallen, then the trend that determines your cost of trade is running in a direction no approval rate will show you.
Most exporters have never asked either question, and most relationship managers have never been asked. Which is worth sitting with, because the industry has spent a decade getting measurably fairer at saying yes, and the amount of trade that goes unfinanced has not noticed.
