The amount credited is smaller than the amount sent. Here is where it went.

An exporter sends fifty thousand euros. The supplier confirms receipt and mentions, without much interest, that they were credited slightly less.

Nobody chases it. On one invoice the difference costs less than the phone call.

Do that twelve times a year and it stops being a rounding error.

Who takes what, and where

The money is reduced in three places. They are not the same kind of charge, and only one of them is negotiable with the party you actually deal with.

The sending bank charges a fee you can see. It is on your statement, you agreed to it, and it is usually the smallest of the three.

The correspondent in the middle takes a charge you did not agree to, because you have no contract with it. Your bank holds an account with it, the payment passes through, and a deduction is applied on the way.

The exchange rate takes the rest, and this is the one people miss. The rate applied to the conversion is not the rate you read in the news. The difference between them is a margin, it is not itemised anywhere, and on most corridors it is larger than both fees put together.

What the published average actually is

The World Bank prices this quarterly in Remittance Prices Worldwide, sampling 367 corridors between 48 sending and 105 receiving countries. It measures total cost, fee plus exchange rate margin, on a transfer equivalent to two hundred dollars.

The measure is defined as the total cost of sending an amount "equivalent to 200 USD", fee and exchange rate margin together, which is worth holding on to when somebody quotes the number at you.

For the third quarter of 2025 the global average was 6.36%, down from 6.49% the quarter before.

Split by how the transfer is sent, the same report gives 4.59% for digital and 7.30% for non-digital.

That spread is the finding. Same corridors, same quarter, 2.71 percentage points of difference between sending a payment digitally and sending it the way it has always been sent.

Read the average honestly

Two caveats. We would rather print them than have you quote the number wrongly at a bank.

First, this series measures remittances, which are small personal transfers. Your commercial payment is larger, and the fixed part of the fee spreads across a bigger amount, so a percentage taken from this table will usually overstate what you pay. The direction of the finding survives that. The size of it does not.

Second, an average across 367 corridors hides the corridor you actually use. Some are close to the average. Some are nowhere near it. The number is useful as a benchmark to argue with, not as a prediction of your next invoice.

Use it the way I do. If your own arrival difference sits well above the digital figure, that is a question for your bank rather than a fact of life.

It has been getting cheaper, slowly

The same series stood at 9.67% in 2009. Sixteen years later it is 6.36%.

That is real progress and it is also slower than almost anyone expected in 2009. The cost has fallen by about a third in a decade and a half, while the cost of moving information across the same border fell to nothing.

The gap between those two curves is the whole subject of this site. Messages became free. Money did not.

The euro area is a different country

Inside the euro area the picture changes, because the payment stops being cross-border in the sense that matters.

The European Payments Council counts 2,792 providers signed up to instant euro credit transfers as of October 2025, covering 92% of providers in the euro area. Instant transfers now carry about 30.7% of the volume.

What that changes for an exporter is narrow but real. If both ends of your payment sit inside that scheme, the conversion disappears, the correspondent disappears, and the two largest deductions go with them.

It does nothing for a payment that leaves the currency area. Most corridors in the World Bank sample do exactly that.

Time behaves the same way. Swift reports that "90% of cross-border payments sent over the Swift network reach the destination bank within an hour", and in the same release that only "43% of all cross-border payments over Swift reaching the end customer's account within an hour". Cost and speed both improve on the international leg and both stall in the same place, which is the domestic one.

What I could not find out

I tried to get a corridor-level figure for commercial payments rather than remittances. I failed.

The banks I asked quote fees, which are the visible part, and describe the exchange rate as market-based without publishing the margin. That is not a refusal to answer. It is an answer nobody can check, and I am not going to convert it into a number that looks like data.

So: I know the shape of the deduction and the published average for small transfers. I do not have a verified series for commercial ones. Anyone who tells you they do should show you the sample.

What I got wrong

I used to quote the 6.36% figure at exporters as though it were their cost.

It is not. That series measures transfers equivalent to two hundred dollars, where a fixed fee lands heavily, and a commercial payment of fifty thousand spreads the same fee across two hundred and fifty times the amount. The direction of the finding holds. The size of it does not transfer, and I was overstating people's losses by quoting it the way I did.

Getting your own number

You do not need the industry average. You need yours. The data is already sitting in your bank statements.

Take one payment. Write down what left your account, in your currency. Write down what arrived, in theirs. Convert the sent amount at the mid-market rate published for the value date, and compare it against what arrived. The difference is your total cost, fees and margin together, on that corridor, on that day.

Do that for three payments across the same corridor and you have a rate you can take to a bank. One payment is an anecdote. Three is a pattern, and it is a pattern about you rather than about 367 corridors.