The only survey about business banking that is not marketing

There is a document that costs nothing, carries no advertising, and answers the question most comparison sites only pretend to answer. It comes out twice a year. I had been running a business for six years before anybody told me it existed.

It exists because of a competition case. When the Competition and Markets Authority finished its retail banking investigation it ordered the largest providers to publish how their own customers rate them, on the same questions, at the same time, in a format nobody can dress up. The file is careful about whose idea it was: the results are published “at the request of the providers and the Competition and Markets Authority”. Ipsos runs the fieldwork. The banks pay for it and have no say in the result.

The document is almost aggressively boring.

Bar charts, a page of footnotes, and one paragraph making it clear that nobody involved is recommending anything to you:

“In providing this information, we are not giving you any advice or making any recommendation to you.”Ipsos, Business Banking Service Quality, Great Britain, August 2026

That sentence is the whole reason it is worth reading.

Nothing in it is trying to move you.

That is rarer than it sounds.

What the number actually measures

Every score in the survey answers one question and one question only: would you recommend this service to another business. Customers are asked about their own provider, not about banking in general, and the published figure is the share who answered “extremely likely” or “very likely”.

So a score of 60 does not mean six in ten customers are happy. It means six in ten were enthusiastic enough to recommend. Merely satisfied counts as a no. That is a hard bar. It is also why almost nothing on the list clears 85, and why a bank sitting on 48 is in worse trouble than the number looks.

The fieldwork behind the August 2026 round ran from July 2025 to June 2026. The published note is precise about the sample: “Approximately 1,200 customers a year are surveyed across Great Britain for each provider”, 20,400 in total. The definition of business is wide, and the footnote spells it out: “businesses, clubs, charities and societies” with turnover up to £25m. A one-person company and a £20m manufacturer answer the same question.

Five separate questions, not one

The mistake is to read the first chart and stop. My first attempt at this was to quote the overall ranking at a friend who runs a café, confidently, for about a month. Then I opened the rest of the file. The measures behave nothing like each other, and the advice I had given her was close to useless.

Ipsos for the CMA, Great Britain, published 17 August 2026. Top-ranked provider on each measure.

Two names take every top between them, and they are opposites in almost every respect: Monzo has no branches at all and wins on the app and on overall service, while Handelsbanken runs branches, gives every customer a named manager, tops both the branch measure and the relationship measure, leads on lending as well, and has an app its own customers rank fifteenth out of seventeen, which is the sort of contradiction a single ranking cannot hold.

One of them is a phone. The other is a person.

Handelsbanken scores 86 on the people who handle the account. Santander scores 45. Same question. Same survey. Forty one points apart, and the overall ranking hides all of it, because the overall figure is an average of things that have nothing to do with each other.

What a blank means

Some cells are empty. Those empty cells carry more information than most of the filled ones. I had assumed a blank meant a bad score somebody was hiding, and that is exactly backwards.

A provider is only ranked on a measure where enough of its own customers rated that specific service, and the threshold is printed in the file itself: a score appears only “where at least 100 customers have provided an eligible score for that service in the survey period”, which means a provider can be excellent at something, or catastrophic at it, and still show a blank simply because not enough businesses used that particular service during the twelve months the fieldwork covered.

A hundred is where a number stops being noise.

Blanks have two unrelated causes. Mix them up and you end up recommending a provider that cannot do the thing you need.

The service does not exist

Five names carry nothing at all under branches: Mettle, Monzo, Starling, Tide and Zempler. The document explains why in its own words, and the phrasing is worth keeping: they are excluded “as do not operate a branch network”.

That is not a criticism. It is a fact about what they are, and it cuts both ways: a business that never needs a counter loses nothing at all from the blank, while a business that banks notes every Friday has just been told everything it needs to know, no matter how good the app is.

Not enough customers answered

The lending measure carries ten providers instead of seventeen. The file explains the cut in one line: “The ranking does not include any brands where the eligible sample size is less than 100”. Monzo, Starling, the Co-operative Bank, Tide, TSB, Virgin Money and Zempler all fall out of it. Some do not lend to businesses at all. Others do, but not enough of their customers borrowed to produce a publishable number. There is even a footnote recording that “Starling Bank does not currently offer SME credit (overdraft and loan) services but did so during the survey period”, which is the sort of detail that never survives into a comparison table anywhere else.

So when you look at lending and see a short list, do not read it as ten banks competing. Read it as: these are the only providers where enough businesses borrowed to produce a publishable number.

Why the top of the table is the least useful part

The overall ranking in August 2026 goes Monzo, Starling, Mettle, Handelsbanken, then Tide and Zempler level. The first three are app-based.

It reads like a clean generational story.

Then you open the branch measure. The same three names are not on it.

This is the survey's most useful property and the one that gets edited out of every summary of it. The app-based providers score highly partly because they do fewer things. A provider that never has to run a counter, never has to staff a business centre and never has to decline an overdraft has fewer chances to disappoint anybody. That is not cheating. It is a genuinely lighter promise, and if the lighter promise covers what you need, it is the better product.

The moment your business needs the heavier promise, the ranking inverts. On branches, the best score in the country is 75 and the worst is 46. On the app, the best is 90 and the worst is 49. Both spreads are wide. They just contain different companies.

Reading it against your own week

Here is how I do it now. It takes about four minutes.

Write down what your business actually does with a bank account. Not aspirations. Last month. For a shop that might be taking card payments, banking notes twice a week, paying eleven suppliers. For a consultancy it is four invoices in, two people paid, something moved to a savings pot.

Now map each item onto a measure. Notes go to branches, anything done on a phone goes to online, and anything that needs a human being to make a decision, whether that is an overdraft, a limit increase or a payment stopped before it leaves, goes to relationship or lending depending on who would have to sign it off.

Then read only those columns. Read them as a spread, not a ranking, because the useful question is never who came first but how far apart the top and the bottom sit, and that distance is exactly what switching is worth to you in the only currency this survey deals in.

On account management that distance is forty one.

On the app it is the same, forty one. On branches, twenty nine. Those are the biggest arguments for moving that exist anywhere in public data, and not one of them appears in an advert.

Reading the movement between rounds

Two rounds are public for 2026. I printed both and laid them side by side on the floor, which is not sophisticated, and it took me most of an evening to type the pairs out, because neither file will let you copy a bank name.

The February round covered January to December 2025. The August round covered July 2025 to June 2026. Notice that they overlap by six months. Half the fieldwork is shared, which means scores move slowly on purpose, and a shift of one or two is mostly the sample turning over.

Against that background, here is what actually moved on overall service:

Overall service quality, Great Britain. Ipsos for the CMA, February and August 2026 rounds. Providers not listed shifted by two or less.

Santander shedding six is the largest single move on the board, and it took the provider from mid-table to last. I cannot tell you why. The survey publishes scores and no reasons, I have not found anything published that accounts for it, and I am not going to invent a story to fill the gap.

HSBC gaining five while Barclays gained four is the other thing worth noticing. Both are large branch banks. Both moved the same way at the same time, and when two competitors move together the cause is usually neither of them.

What you should take from this section is a piece of discipline. One round is a photograph. Two rounds tell you the direction. Three rounds, which arrive next February, will start to tell you whether a provider is actually changing or whether you are watching noise with a logo on it.

What the survey cannot tell you

It says nothing about price.

A provider can be loved and expensive at the same time, and several on this list manage it comfortably.

It says nothing about who is holding your money or what happens if they fail. That is a separate question with a separate answer, and it is lesson 3.

It says nothing about complaints either. For that you need the ombudsman, whose own quarterly note records that “The most complained about product was current accounts”, which is a different kind of unhappiness from the one this survey measures.

It cannot tell you about a provider too small to survey. Seventeen names appear in Great Britain. There are far more accounts on sale than that, and the small ones are invisible here whether they are good or awful.

And it is a photograph, not a film. The publication schedule is fixed: “Results are updated every six months, in August and February”. A score that moved four may mean something changed. It may also mean a different 1,200 people answered.