Best business bank account 2026: we ranked it on 84,177 complaints instead of fees

My first attempt at this ranked banks by how many complaints they get, which is the obvious thing to do and is also useless, and it took me most of a day to work out why.

Chase drew 9,571 complaints about checking and savings in the year to July, and Wells Fargo drew 9,731. Both numbers look damning until you remember that both banks hold deposit relationships in the tens of millions, at which point the only thing the ranking measures is how many customers a bank has. I had built a size table and labelled it a service table. So do not rank anybody on complaint counts, and I say that as the person who published one and had to take it down again.

What rescued it was a column I had not paid attention to. When the Bureau forwards a complaint to a company, the company has to say how it closed the matter, and one of the options is that money went back to the person complaining. That is a ratio rather than a count, so it survives the problem, and it turns out to vary far more than I expected.

Across the whole year, 10,271 of 84,177 complaints closed with money. Call it 12.2 per cent as a baseline. Then look at the individual names and the baseline stops meaning very much.

Bank of America closed 2,848 of 8,353 with money, which is 34.1 per cent, nearly three times the market. Citibank came second at 925 out of 3,651. Wells Fargo, with more complaints than anybody, sits at 12.0, and Chase comes in at 9.7. Then there is a long flat middle where most of the industry lives, and at the bottom Capital One at 100 out of 3,919, and Block, which closed 3,753 complaints and gave money back in none of them.

Zero is a strange number to find in a column like that. The market rate would have predicted something over four hundred.

I do not know what it means and I want to be careful here, because there are two readings and the file cannot choose between them. Either a company is refusing to pay, or the complaints it attracts are genuinely the kind with no money attached, arguments about accounts being closed rather than funds disappearing. Both are consistent with a zero. A company that settles quietly before anything ever reaches the Bureau would leave the same trace, and so would a company whose customers happen to argue about different things than everybody else's customers argue about, and there is no field anywhere in the published record that lets me separate those three stories from one another. What I can tell you is that a company answered 3,753 people about their money and found nothing owed in any single case, and that is a fact worth carrying into a sales conversation.

An aside about the postal complaints

While I had the file open I looked at how complaints arrive, which has nothing to do with any of this.

75,856 came through the web form, and another 3,189 were phoned in. And 723 people printed a complaint about their bank account and posted it to a federal agency. Seven hundred and twenty three stamps, in a year, about money sitting in a bank. Some of those will be older customers, some will have wanted a paper trail, and some will have started on the website and given up. The web form is the default route and everything else is a deliberate choice to do it the harder way.

It is not relevant to which bank you should use and I keep thinking about it anyway. Right, that is the digression finished, back to the ratios.

The part you can act on

If you are choosing between two providers, the useful question is not what the fee is. Ask what share of escalated complaints they close with money, and if they will not tell you, the public file will, for any company large enough to appear in it. It is a question most salespeople have never been asked, and the pause before the answer tells you something on its own.

That is the whole practical content of this article and it is one paragraph long. Everything else here is me explaining why the more obvious comparisons do not work, and one thing I discovered late about which banks the file is even allowed to describe.

The 155 banks this file can see

Size is the only thing that decides whether a name can appear here at all. I had been treating the complaint database as a picture of American banking, and it is not one. The Bureau says so on the page that explains what it publishes, in a sentence I had read several times without taking in: “We do not publish complaints referred to other regulators, such as complaints about depository institutions with less than $10 billion in assets.”

Ten billion is a very high line. I went to the FDIC register on 3 August to see how high, using the same BankFind endpoint the rest of this page is built on, and the index dated 31 July 2026 holds 4,254 active insured institutions. Of those, 155 report assets of 10 billion dollars or more. A thousand and forty two report at least one billion.

So the ranking above covers 155 institutions out of 4,254, which is 3.6 per cent of the banks in the country and very close to all of the deposits. Every community bank on your shortlist, and every credit union that is not Navy Federal, sits outside this file entirely. Not because they are clean. Because their complaints go to a different regulator and never reach the public database at all.

I spent a fortnight before I noticed this quietly reading absence as a good sign, and the Bureau warns about exactly that habit: “The lack of complaints or a relatively low number of complaints published in the database about a product, issue, or company does not necessarily mean there is little or no consumer harm.” If your final two are a large national bank and a bank down the road with two billion in assets, this article can rank one of them and is silent about the other, and the silence carries no information in either direction.

What people are actually complaining about

The issue breakdown is more stable than the company breakdown and, for a business, probably more useful.

Managing an existing account accounts for 48,251 complaints out of 84,177. Then a lender or another company charging your account, 11,869. Then closing an account, 11,225. Then opening one, at 7,412 complaints. Then problems caused by low funds, 4,733.

So fifty seven per cent of everything is one category, and inside that category the recurring events are holds, freezes and closures rather than fees. Fees are what every business bank account comparison ranks on, including the ones I used to read before I started doing this, and fees are not where the trouble is.

The pair I keep returning to is closing against opening. Eleven thousand two hundred and twenty five complaints about leaving, seven thousand four hundred and twelve about arriving. Every guide written for business owners is about arriving.

Where I would not push this

There are limits on this file, and they are not small print, they are the shape of the thing.

The database cannot see business accounts separately. The Bureau's product taxonomy is checking and savings, and it does not ask whether the name on the account is a person or an LLC, so a plumbing company and a retired teacher land in the same bucket. My instinct is that business disputes skew towards holds and consumer disputes towards fees, and that is an instinct, not a finding.

The Bureau is blunter about this than most people who quote it. Its own page says the file “is not a statistical sample of consumers' experiences in the marketplace and these complaints are not necessarily representative of all consumers' experiences with a financial product or company”, and it adds that complaints are “not 'information' for purposes of the Information Quality Act”. On the narratives, the wording is “We do not adopt their views or verify that their experiences are accurate or unbiased”. I quote all of that at length because I have seen these counts presented as findings, including once by me.

Publication has a clock on it as well. Complaints are “only published after the company responds, confirming a commercial relationship or after 15 days, whichever comes first”, which means the most recent weeks in any pull are thinner than the older ones and nobody should read a trend off the tail end.

And every complaint in here is a case where something had already gone wrong. Nobody files with a federal regulator about a Tuesday when the transfer cleared. So the file describes the shape of failure at each company and says nothing at all about how often failure happens, which is the number people actually want and which nobody has.

This is our reading of a public file rather than advice about where to keep your money. A table like this cannot see your balances, your borrowing or what your bookkeeper has to work with, and if you are seriously weighing a move it is worth asking your accountant, who can see the parts we cannot.

The number that stopped me

615.

That is how many of the 84,177 complaints got a late response, against 83,562 answered inside the deadline. I went in expecting that to be bad and it is not bad at all. Companies treat a Bureau referral as something that gets answered on time.

Which leaves me with something I cannot resolve into a tidy sentence. If 99.3 per cent of complaints are answered promptly and only 12.2 per cent end with any money, then the standard outcome of escalating to a federal regulator is a punctual, well written explanation of why nothing will happen. That is either a process working exactly as designed or a process performing, and after four months in this file I still switch between the two depending on which company I am looking at.

One last thing, and it is the sort of detail that decides how much weight any of this can carry. When I ran the identical query again on 3 August, the same closed twelve month window returned 84,217 complaints rather than 84,177, and 10,380 of them had closed with money rather than 10,271. The Bureau backfills, so a window that ended a month ago is still moving. The percentage barely shifts, from 12.2 to 12.3, and the counts on this page are the 29 July pull throughout, which is why they will not match yours to the digit if you go and check. I would rather say that than have somebody assume the difference is an error.