Your Card Fee Is Three Fees Pretending to Be One

Most small sellers know one number about card acceptance. It is on the offer letter, it looks like 1.9 percent plus 20 pence, and it is the number they compare when someone offers 1.7. That single figure is a package holiday price. It covers three different fees paid to three different parties, and the reason it is presented as one number is that one number is harder to argue with.

This piece is about opening it up. Not to grind out a tenth of a percent, though that is available, but because the parts behave differently. One of them is fixed by law, one of them rises on a schedule you do not control, and one of them is your provider's margin, which is the only part a conversation can move.

What is actually inside a card processing fee?

Three layers, plus two smaller ones that hide in the totals. Interchange goes to the bank that issued the customer's card. Scheme fees go to the network that carried the transaction. What remains is the acquirer's margin, which is the business you are actually a customer of.

The layers matter because they are not negotiable in the same way, or at all. Interchange is set by the card networks and, in several markets, capped by regulation. Scheme fees are set by the networks and move when the networks decide. The margin is set by your provider in a document you signed.

The two smaller layers are worth naming because they are where surprises live. Authorisation and gateway charges are per attempt rather than per sale, which means declines cost you money, and a high decline rate is therefore a fee problem as well as a revenue problem. That connection is the reason false declines at checkout are more expensive than the lost basket alone suggests. The other is the non qualified or standard rate surcharge, applied when a transaction does not meet the conditions for the rate you were quoted, and the conditions are usually things like a commercial card, a foreign card, or a card entered by hand rather than tapped.

Read the leverage column and the strategy becomes obvious. People spend their energy on the layer they cannot change and never ask about the one they can.

Are you entitled to see the breakdown?

In the UK and the EU, yes, and this is the most underused right in small business payments. The rule has a name that tells you what it does: unblending.

Article 9 of the interchange fee regulation requires acquirers to offer and charge merchant service fees specified individually for different card categories, brands and interchange fee levels. It requires them to provide the merchant with individually specified amounts for the merchant service charge, the interchange fee and the scheme fees, per card category and brand. There is one exception, and it is the important part: this applies unless the merchant has requested blended charges in writing.

Most small merchants are on blended pricing. Some of them requested it, in the sense that they signed a form containing that request. So the first action from this article is a short email to your provider asking for interchange plus pricing, or at minimum for the unblended breakdown the regulation describes. The answer tells you two things: what you are actually paying each party, and how your provider reacts to an informed customer.

The same regulation caps consumer interchange. Article 4 sets the credit card cap at 0.3 percent of transaction value, and the debit equivalent sits at 0.2 percent. If your blended rate is 1.9 percent on a customer base paying mostly by consumer debit card, you now know roughly what share of that is interchange, and it is small.

Why does the same card cost a different amount?

Because the fee follows the card's issuer and category rather than the purchase. The clearest published illustration of this comes from US debit data, where the cap applies to some issuers and not others.

The Federal Reserve's standard is precise: an issuer subject to the interchange fee standard may not receive more than 0.21 dollars plus 0.05 percent of the transaction value, plus a 0.01 dollar fraud prevention adjustment. Issuers with total worldwide banking and nonbanking assets below 10 billion dollars are exempt, along with certain government programme and prepaid cards.

Now the numbers that follow from that split, for 2024 across all networks. Covered transactions averaged 0.23 dollars each, which is 0.47 percent of value. Exempt transactions averaged 0.51 dollars, which is 1.21 percent of value. Exempt cards were 39.3 percent of the total number of transactions.

Sit with that third figure. More than a third of US debit volume carries an average interchange more than double the capped rate, and nothing about your business determines which card a customer pulls out. A shop whose customers bank with community banks and credit unions pays materially more than an identical shop across the street, and no negotiation touches it. The practical consequence is that comparing your effective rate against another merchant's is close to meaningless unless the card mix matches.

What has been happening to the layer nobody can negotiate?

It has been rising, and a regulator has now put a number on it. The UK Payment Systems Regulator's final report on card scheme and processing fees, published 6 March 2025, found the two networks had raised their core scheme and processing fees to acquirers by at least 25 percent since 2017. The regulator put the extra cost to businesses at a minimum of 170 million pounds every year.

The same report found a shortage of transparent fee detail imposing costs on both acquirers and merchants, with small retailers particularly affected. That finding is the part worth holding onto, because it validates something every small merchant suspects and cannot prove: the complexity is not incidental. When a regulator concludes that the information provided is complex or incomplete, a merchant who cannot reconcile their own statement is not being slow.

What is the machine actually good for here?

Extraction and reconciliation, which sounds modest and is where the money is. A processing statement is a nested document with subtotals that do not obviously roll up, which is precisely the case document AI research treats as hard.

The ReceiptBench work on receipt understanding organises this kind of extraction into four sub tasks, and the ordering maps onto a processing statement almost exactly: basic perception for raw text, format normalisation for following standardisation instructions, semantic reasoning for inferring implicit attributes from context, and structure parsing for handling nested line items. Your statement is the fourth case. The fee lines are nested under card brands, which are nested under transaction types, and the totals are computed at a level the document does not always show.

So the useful instruction to a tool is narrow and checkable. Extract every fee line with its amount, its card brand and its transaction type. Compute the effective rate per card type rather than overall. Total the authorisation charges separately from the percentage fees. List every distinct fee name that appears, because an unexplained fee name is the single highest yield finding in this whole exercise.

What the tool should not do is tell you whether a fee is fair. That requires knowing what your provider agreed to, which lives in your contract, and what the caps are in your market, which is a legal question. A model asked whether your rate is competitive will produce a confident paragraph from general web content about average rates, and averages across a mixed card base are the least useful number in this field.

What about the fees that are not on the statement?

There are three, and they are the ones a rate comparison never captures because they arrive by other routes. Each is worth a number before you decide any provider is cheaper.

The first is the payout delay. Money held for two days rather than settled next day is working capital you do not have, and for a business buying stock weekly that has a real cost even though it appears nowhere as a fee. Compare settlement timing alongside the percentage, and treat a longer delay as a price.

The second is the rolling reserve, which is not a fee at all but behaves like one. A provider holding a percentage of your volume for 90 days against future disputes has taken that money out of your business for a quarter. New accounts and higher risk categories see this most, and it is negotiable in a way the headline rate often is not, because it is set by an underwriting judgement rather than a rate card.

The third is currency conversion. If you sell in one currency and settle in another, the conversion spread sits inside the exchange rate rather than in a fee line, which makes it invisible to any statement audit that looks only at fee names. The way to find it is to compare the rate you were given against the interbank rate on that date. A spread of one or two percent on cross border sales dwarfs the basis points people argue about, and it is the single most common reason a merchant's real cost of acceptance is well above their quoted rate.

Does taking payments differently change the fee?

Yes, and this is where a small merchant has more control than they expect. The fee follows how the transaction was presented as much as what was bought.

Four mechanics move it in most rate cards. A card physically tapped or inserted qualifies for a better rate than one keyed in by hand, because the fraud risk is lower and the network prices accordingly. An online transaction with full address and verification data qualifies better than one without. A recurring payment flagged as recurring is treated differently from the same amount taken as a fresh purchase. And a transaction authorised once and captured once costs less than one authorised, reversed and reattempted.

None of those requires a new provider. They require your checkout and your terminal to be configured to send the data the network wants, which is usually a setting rather than a project. Ask your provider which of your transactions fell to a higher rate last month and why, and the answer either names a mechanical cause you can fix or reveals that nobody at your provider can explain your own pricing, which is itself useful information.

Which checks are worth running on one month?

Five, and they take an evening with a statement and a spreadsheet. Each one has a definite answer rather than an opinion.

Compute the effective rate per card type, not overall. Total fees divided by total volume tells you almost nothing, because the mix moves it. The per type figure is what you compare month to month and what you take to a renegotiation.

Count authorisation charges against successful sales. A gap means you are paying for attempts that failed, and the fix is upstream in your checkout rather than in your pricing. This is where retry behaviour on failed payments quietly adds cost, because a naive retry schedule buys you both a fee and an annoyed customer.

Find the non qualified or standard rate volume. If a meaningful share of your transactions is falling to a higher rate, the cause is usually mechanical: keyed entries, missing address data, commercial cards, or a terminal configuration. Mechanical causes have mechanical fixes.

Check refunds. On many rate cards the percentage fee on the original sale is not returned when you refund, so a high return rate carries a fee cost that never appears in your margin calculation. Establish whether yours are returned, because the answer varies by provider and it changes how you price returns.

Separate chargeback and dispute admin fees from the disputed amounts. They are different problems with different fixes, and merging them hides which one is growing. Providers also differ on whether the admin fee is returned when you win, and the reserve and holds side of that relationship is covered in how a merchant risk score drives reserves and payout holds.

What can a small merchant realistically renegotiate?

The margin, the authorisation charge, and occasionally the terms around refunds and disputes. Not interchange, not scheme fees, and no provider can give you those even when the sales conversation implies otherwise.

The leverage is smaller than a large retailer's and it is not zero. Three things move a margin conversation. Twelve months of volume history, because that turns your business from a projection into a fact. A competing written quote, because it makes the alternative concrete. And an unblended statement, because it proves you know which part you are asking about. Providers respond differently to a merchant who asks for a better rate and one who asks for 15 basis points off the margin on consumer debit.

Timing matters too. Contract renewal is the obvious moment and the worst one, because switching is a real cost and both sides know it. The better moment is after a volume increase, when your own numbers have changed and theirs have not.

One honest note on switching, since it is the implicit threat behind any negotiation. The cost of moving provider is rarely the fee difference. It is the integration work, the settlement gap during the changeover, and the risk of a new provider applying a rolling reserve to an account with no history with them. Price those before using them as leverage. For comparison, the business model we run ourselves publishes what a build costs rather than quoting per account, and the reasoning behind our own published pricing is the same argument this article makes about yours: a number you can read beats a number you have to request.

The part that compounds

Doing this once recovers a little money. Doing it monthly changes what you know about your own business, because the effective rate per card type is a leading indicator of things that have nothing to do with payments. A rising share of manually keyed transactions means something changed in how orders arrive. A rising share of commercial cards means your customer base is shifting toward businesses, which is a pricing and terms question rather than a fee question.

That is the argument for automating the extraction rather than the judgement. A statement parsed into the same spreadsheet every month gives you a series, and a series answers questions a single statement cannot. The tool does the reading. The decision about what a trend means stays with the person who knows why last March was strange.

Start with the email asking for the unblended breakdown. It costs nothing, the regulation is on your side in the UK and the EU, and until you have it every other number in this exercise is an estimate built on a package price.