Payment processor comparison: the four terms that cost more than the rate

I have recommended processors on the basis of 20 basis points, and I had it wrong to do so.

The rate is real and the saving is real. What none of my comparison work ever touched is the clause that lets a processor impose a rolling reserve, and a reserve appears in no rate table anywhere, so this is the corrected version of advice I have given in public more than once.

The number

One hundred and forty four thousand dollars, and I never used to look for it.

Take a business settling 300,000 dollars a month, put it on an 8 per cent rolling reserve with a 180 day hold, and run it to steady state. That is the arithmetic and it produces 144,000 dollars withheld. It is not a fee and the money does come back eventually. It simply cannot be used, nobody lent it, and no line in the accounts explains where it went.

I think the shape of the 144,000 is worse than the size, and the shape is what nobody warns you about.

The reserve builds slowly and that is the trap in it. I keep thinking about how well designed that is, whether anybody designed it or not, because a cost that arrives at 8 per cent a month for 6 months never triggers the alarm that the same total would trigger on day 1. Month 1 of that model costs 24,000 dollars. By month 3 the model is down 72,000, which is still small enough to read as a rounding issue on a 300,000 a month business. Month 6 is where it stops growing and starts hurting, because by then the working capital the plan assumed is simply absent.

What twenty basis points is actually worth

I want to put my own error in proportion, because the arithmetic makes it look sillier than it felt.

Twenty basis points on 300,000 dollars a month is 600 dollars, and I would guess most people reading this have optimised for a number that size. Over a full year that comes to 7,200 dollars, against 144,000 sitting in a reserve. That is what I optimised for.

The reserve takes 144,000 out of circulation. Funding that gap with a facility at 10 per cent costs 14,400 dollars a year, which is twice the saving. A facility is also not something anybody arranges quickly, because the lender wants to see balances that are no longer under the borrower's control.

So I measured a thing worth 7,200 and ignored a thing worth double that. It still irritates me, mostly because the clause was right there in the agreement and I read the pricing schedule instead.

What a payment processor comparison should actually compare

I rank these by how much damage each can do in a quarter, and the rate does not come first.

Reserve terms come first, and I think they should sit at the top of every comparison table in this industry, which is not where anybody puts them. Ask whether one applies today, at what percentage, over what hold period, and whether the processor can impose or raise one after you sign, which matters more than the other 3 answers. Almost every agreement I have read says it can, usually in a clause about changed risk profile, and that clause is negotiable more often than people assume. Fight for the notice period. Thirty days of warning turns a crisis into a plan, and a reserve imposed with no notice at all lands on a business in the same week it lands on the cash flow forecast, which is the difference between a hard quarter and a missed payroll.

Settlement timing comes second and it behaves like a fee. Nobody bills you for it. It never appears on a statement, and that is precisely why it survives every comparison anybody makes. Two business days against seven leaves five extra days of revenue in transit, permanently. On 300,000 a month that is roughly 50,000 dollars of working capital that exists on paper and not in the account, and funding that gap at ten per cent would cost 5,000 dollars a year. So a processor twenty basis points cheaper and five days slower costs money rather than saving it, and I have never seen a comparison table that says so.

Termination rights come third, and they are the clause people skip because they cannot imagine the relationship ending badly on the day they are signing up for it. They close you with thirty days notice and hold settled funds for a hundred and eighty afterwards. Find that out in week 1.

Approval rates come fourth and beat the headline rate outright, which is the least intuitive thing here and the one I would stake most on. A processor charging 2.7 per cent that approves 94 per cent of legitimate traffic beats one charging 2.5 that approves 91, because three points of declined good customers costs far more than twenty basis points. Nobody publishes any of these figures. I could not find a single processor stating a reserve percentage or a hold period on a public page, across 6 providers and an afternoon, and it strikes me as the strangest gap in an industry that measures everything else to four decimal places.

How to get the answers they will not volunteer

Always put these in an email rather than on a call. A call gives you reassurance and nothing you can quote back. An email produces a document you can hold up later.

Here are the ones I now send, and I have watched every one get answered in writing. Does a reserve apply now, and under what conditions can one be introduced later. What notice do I get before settlement timing changes. What happens to funds in flight if either side terminates. What is the chargeback fee, and do you charge it when I win. Does a refund return the original processing fee. And what is your approval rate for merchants in my category.

The last one usually gets refused. I would ask anyway, because the manner of the refusal tells you something, and because two out of nine processors I asked did answer with a number.

High risk, which you can join by accident

High risk says nothing about your honesty. It describes something narrower and colder than the phrase suggests. The label tracks the likelihood of chargebacks, and plenty of ordinary businesses sit inside it: subscriptions, free trials, travel, ticketing, supplements, coaching, and anything where delivery happens long after payment.

The thresholds are published by the networks, and I had them wrong here until this week, which is uncomfortable given how confidently the old version was written. The Visa figure of 0.9 per cent that still circulates in every guide is retired. Visa now runs one programme, and its own fact sheet describes the metric as a “single, count-based ratio (VAMP ratio) that includes key components of fraud and disputes on card-not- present VisaNet transactions”, defined as “Count of [Fraud (TC40) + Disputes (TC15)] ÷ Count of Settled Transactions (TC05)”.

Here is the part I had backwards. The 0.5 and 0.7 per cent lines everybody quotes are not merchant lines at all. In Visa's wording, “an acquirer's portfolio is identified as Above Standard if its VAMP ratio is ≥50bps and as Excessive if ≥70bps”. The merchant line sits much higher and moved recently: the Excessive Merchant threshold “reduced to >=150bps in AP, Canada, EU, and U.S. regions on 1 April 2026”, against a monthly count of at least 1,500 fraud and dispute events.

That count floor matters more than the percentage for most readers. Below 1,500 events a month you cannot be identified by Visa at all, and it would be a mistake to relax about that, because your acquirer watches the same ratio with no floor underneath it and sets its own line wherever it likes.

Mastercard runs its own programme with different arithmetic. Its published criteria put the Excessive Chargeback Merchant tier at 100 to 299 chargebacks in a month together with a rate of 1.5 to 2.99 per cent, and the High Excessive tier at 300 chargebacks and 3 per cent. Both conditions have to be met in the same month. Exit is slow: months in the programme “do not reset until the MID has achieved 3 consecutive months below thresholds”, so one bad month becomes a bad quarter with paperwork attached.

The fines escalate on a published ladder and the shape of it surprised me. The first month in the programme costs nothing at all. Month 2 is 1,000 dollars, month 3 is another 1,000, months 4 to 6 are 5,000 each, months 7 to 11 are 25,000 each, months 12 to 18 are 50,000 each, and anything past month 19 is 100,000 a month. The High Excessive column runs at double those figures throughout. Read the first two months as a nuisance and the ladder reaches 15,000 dollars by month 6 and 140,000 by month 11, on a problem that started as a hundred disputes.

Now the trap, and it is built into the arithmetic rather than into anybody's behaviour. Mastercard computes the rate as the current month's chargeback count divided by the preceding month's captured transactions. Read that again, because I did not read it properly the first time and wrote a whole paragraph on the wrong month.

A seasonal business whose volume halves in February does not cross a threshold in February. February's chargebacks are divided by January's larger volume, so the ratio looks fine. The damage lands in March, when March's chargebacks meet February's shrunken denominator, and by then everybody has moved on and nobody connects the breach to a quiet month six weeks earlier, which is the sort of lag that turns a bookkeeping curiosity into a real misdiagnosis, because the month that looks wrong on the report is not the month that caused it and nothing in the notification says otherwise. The numbers get read as a fraud problem. They are a calendar problem.

Watch the ratio weekly if you sit anywhere near a monitoring level. And watch the denominator rather than the numerator, which is advice I have given four times this year and which nobody has taken yet.

What the complaint record says about the rails

While I had the file open I ran the comparison the industry avoids, which is how the payment rails behave when somebody escalates.

The Consumer Financial Protection Bureau logged 37,516 complaints about money transfer and virtual currency in the twelve months to July 2026, and closed 2,699 of them with money going back, which is 7.2 per cent. Checking and savings ran at 12.2 per cent across 84,177 complaints, which is the closest thing to a market baseline anybody publishes. Prepaid cards managed 746 out of 6,294, or 11.9 per cent, near enough the same.

Deadlines tell the same story from another angle, and I find them the more honest column of the two. In banking, 615 responses out of 84,177 arrived late, so 0.7 per cent. Money transfer ran at 947 out of 37,516, which is 2.5, and prepaid came in worst at 274 out of 6,294, or 4.4 per cent, six times the banking figure and the widest spread anywhere in the file I pulled.

None of that measures your own processor. The Bureau records complaints from consumers rather than merchants, so you barely appear in the file at all. What it measures is how much apparatus sits behind an answer. A chartered bank has a compliance function built around an examiner eventually reading its files. A money services business has a smaller one, and under load the first thing that gives is not honesty. What gives under load is speed.

Switching, and what it actually costs

People describe a switch as a plumbing change and I find that framing genuinely dangerous. Budget for it properly before you sign, because 8 per cent of 300,000 dollars a month is 24,000 dollars leaving the working capital line every month until the hold period turns over.

Then there is the integration work, a re run of recurring billing tokens, which is the step that goes wrong most often, two to six weeks where approval rates sit below where they used to be while the new acquirer learns your traffic, and the loss of your dispute history, which resets the context in which your ratios get read.

Against all that, twenty or thirty basis points is not a reason to move. I know, because I moved somebody for twenty and I am still slightly embarrassed about it. A reserve you cannot live with, settlement timing that breaks payroll, or approval rates costing you real customers are reasons to move. The distinction is whether the problem is a price or a constraint.

Run both in parallel for a full billing cycle at minimum, and preferably for 2, because the second cycle is the one where the monthly minimum, the statement fee and anything annual finally show themselves. Keep the old one live. Nearly everyone skips that and nearly everyone who skips it finds something in week two.

The bit I could not find out

I asked nine processors for approval rate figures I could compare and got seven refusals, five of them polite. The two that answered gave numbers I cannot verify, over periods they chose, for categories they defined.

So I hold two data points that do not compare with each other and that nobody can check, which annoys me more than a flat refusal would have done, because it looks like evidence and behaves like noise.

I also could not establish how often reserves land on merchants who never breached a threshold. Three people have told me it happened after one heavy refund month. One month. No processor will describe how that decision gets made, and my guess is that a model produces a score and a human signs it off, which is a guess with nothing behind it.

A short warning to anyone about to cite this

The complaint figures came from the Bureau's public database for the twelve months to July 2026. The Visa wording is from its own VAMP fact sheet, where the threshold for merchants moved on 1 April 2026, recently enough that half the guides you will find still quote 2.2 per cent or 0.9. The Mastercard figures come from the network's published programme criteria. All of these get revised, and the revision that caught me was six weeks old when I found it.

Everything I say about reserves and settlement comes from agreements I have read rather than from any published survey, so treat it as the shape of the market and not as a measurement of it.

I am not your lawyer. None of this is advice on a contract sitting in front of you right now, and a reserve clause in particular rewards an hour of somebody who reads these professionally rather than an hour of me.

A reserve does get released in full and on schedule, and that is the part people forget when they read the number. What I cannot tell you is whether any given reserve was ever justified by the underlying risk, because a processor is not obliged to explain its reasoning, and across the 6 agreements I have read, none of them does.