Business sweep account: what it costs, when you need one, and the cover you already have

I had it wrong in a Slack thread in March and somebody corrected me in front of about 40 people, which is the fastest way to learn something and not the one I would have chosen.

Somebody had asked whether four accounts at the same bank would cover two million dollars. I told him yes, confidently, in front of everybody. It does not, and the clause I had misread costs real companies real months, so I spent an afternoon working out what it actually takes to fix and this is what I found.

The number

The FDIC states it as "$250,000 per depositor, per FDIC-insured bank, per ownership category". I had read that as per account, and I had it wrong for years rather than for weeks. Four accounts at one bank in one category do not give you a million. They give you 250,000 dollars and nothing more.

Two million in a single bank leaves 1,750,000 uninsured, or 87.5 per cent of it. Five million needs twenty separate banks. Ten million needs forty of them. Twenty five million needs one hundred.

A business sweep account is the product most people reach for at this point, and I will come to it, but first a short digression about the phrase ownership category, because it is the part of the sentence people skip and it is where all the room is. It is not a synonym for account type. It means the legal capacity in which the money is held, so a sole proprietorship is the owner personally while a two member LLC is not, and 2 accounts in the same capacity share 1 limit no matter how differently they are named. Anyway, back to the arithmetic.

One hundred banks

That figure sat in my notes for three weeks and I could not make it feel normal, so I costed it.

Each new bank relationship needs the same stack: your formation certificate, the EIN letter, an operating agreement that matches both, and an identity file for every individual who owns 25 per cent or more plus one control person. For a three member company that means four files per bank. Then a login, a signatory list, a set of payment limits, and a statement somebody reconciles every month.

Call it four hours of setup and twenty minutes a month afterwards. At one hundred banks that lands at 400 hours of onboarding and 33 hours every month forever. Two weeks of somebody's year, permanently, to hold money you already own.

Nobody does this and nobody should.

What bothers me is the gap it leaves. Finance teams look at the hundred bank version, decide it is absurd, which it is, and then do nothing at all instead, so they sit on uninsured balances for months and hope. The fix was never one hundred banks. It is three, and it takes an afternoon.

How to get from one hundred banks to three

Three levers, and I rank them by how much protection each buys per hour you spend.

The first lever costs nothing and I am still faintly annoyed that nobody told me about it. Ownership category multiplies your limit at one bank, for free, on paperwork you have already filed.

A single member LLC shares its owner's limit, so the two of them together stop dead at 250,000 and no amount of extra account numbers moves that. A multi member LLC counts as its own depositor and adds another 250,000. So does a corporation.

The one that surprised me is the revocable trust. The FDIC gives its own worked example: an account "with one owner naming three unique beneficiaries can be insured up to $750,000", and the mechanism runs to 250,000 per beneficiary up to 5 of them, so a founder with a spouse and three children carries 1.25 million of cover at a single bank on a document most people set up for entirely unrelated reasons.

Run that through a real structure. An operating LLC plus a holding corporation already carries 500,000 dollars of cover at 1 institution, because they are 2 separate legal entities, and nobody at the bank has any reason to mention it. Add the owner's personal accounts and that trust and you clear 2 million without opening a single new relationship.

Count what you already own before you open anything.

The second lever is a sweep. You keep one account, agree a target balance, and the operator pushes everything above it into deposits at other banks in slices under 250,000. You reconcile one statement instead of a hundred, and coverage of 50 million or more is routine rather than exceptional.

Two things to pin down before signing. The cut off time, because same day return before it and next day after it are different products on a payroll Friday. And whether you can exclude specific banks, which matters if one of them already lends to you and you would rather not hand them your deposits as well. The catch sits in the pricing and I could not find it published anywhere, which I come back to below.

The third lever deletes the problem rather than managing it. Treasury bills carry no insurance because no third party exists to fail, so the 250,000 stops applying at any size.

Bills come at 4, 8, 13, 17, 26 and 52 weeks. Treasury auctions 4 and 8 week bills every Thursday, 13 and 26 week bills every Monday, and the 52 week roughly every four weeks, so a ladder built on the weekly maturities gives you something coming due every seven days. Minimum purchase through TreasuryDirect is 100 dollars, which surprises people who assume this is only for large balances.

What I would actually do with two million

Here is the split I use when people ask, and I offer it as my working method rather than as advice.

Three months of outgoings stay in the operating account. A company burning 120,000 a month keeps 360,000 there, which leaves 110,000 outside the limit, and I accept that exposure quite happily because chasing it would cost more attention than it is worth and attention is the scarcer thing.

The next six to nine months, so 720,000 to 1,080,000 on that burn, go into a second insured home. A plain second bank handles it if the amount fits under 250,000; above that you want a sweep. One extra relationship, four hours of onboarding, and roughly 85 per cent of the original exposure disappears.

Everything past that goes into bills or a fund holding them, laddered so something matures every four to thirteen weeks. Two accounts and a ladder, against one hundred banks and 400 hours.

What the exposure actually slows down

People treat this as a safety question. In practice it shows up as a speed question first, and that is the part nobody warns you about.

Payroll goes first, and it goes badly, because payroll is the one payment in the month where being 1 day late is a different category of problem from being 1 day late on anything else.

A finance lead who has scattered money across five banks to chase coverage now needs transfers between them before every run, and interbank transfers settle on their own schedule rather than yours. Audits go second, because every additional account means another confirmation letter and another reconciliation in the working papers. Board reporting goes third, since cash across five institutions never agrees on a single date without somebody chasing it.

Credit facilities go last and hurt most, because lenders write covenants against balances they can actually see, and a balance sitting at an institution the lender has no relationship with is a balance that may as well not exist for the purposes of the document you are about to sign. Money spread thin across banks the lender does not hold makes a company look worse on paper than it is, and I have watched a term sheet slow down over exactly that.

The comparison nobody runs

While I had the file open I compared the risk everybody insures against with the one nobody does.

Eleven insured institutions have failed since January 2023. Every one closed with insured depositors made whole, and each resolution took a weekend. Three were the cluster everybody pictures. First Republic held 212.6 billion dollars of assets, Silicon Valley 209.0 billion, Signature 110.4 billion.

The other eight came in far smaller, with a median around 200 million, and Elkhart in Kansas has under two thousand people living in it. I find that list oddly reassuring, and I suspect almost nobody reads past the first three names, which is how the whole subject ended up feeling more frightening than the record supports.

Against that sits a much bigger file. The Consumer Financial Protection Bureau logged 84,177 complaints about checking and savings in the twelve months to July 2026, and 48,251 of them concern an account somebody already held, where holds, freezes and closures dominate the reasons.

The pair I keep staring at is 11,225 against 7,412, which is closing an account against opening one, so leaving generates half as much trouble again as arriving and every guide written for founders covers only the arrival. And across that entire file of 84,177, exactly 10,271 ended with any money going back to the person who complained, which works out at 12.2 per cent.

One of those risks has a product sold against it. The other happened to 48,251 people last year and has nothing.

So run a second banking relationship even when your balance sits under the limit. It costs an afternoon and it solves the far more likely problem, which is that your own bank freezes something on a Tuesday.

The bit I could not find out

Sweep programmes charge through a spread. The network banks pay a rate, the operator keeps part of it, and you receive the rest.

I spent a week hunting for the size of that part. Three product pages, two custodial agreements, one of which defeated me, and two conversations with people who sell the product, and at the end of it I had nothing I could print. Operators negotiate the number per client, no public benchmark exists, and 2 companies of the same size on the same product can be receiving materially different yields without either of them ever finding out.

Set that beside the neighbours. A bank publishes its deposit rates on a page anybody can read. A money market fund publishes an expense ratio and a seven day yield. This product publishes neither, and my guess is simply that nobody ever wrote a rule requiring it, which is duller than a conspiracy and fits how disclosure obligations actually arrive in finance, one scandal at a time and never in advance. That is a guess with nothing behind it except the absence of a rule.

The workaround is 1 email and I think it is the most useful paragraph in this piece. Ask what yield you will receive, in writing, alongside the rate the network banks are paying that week. The gap between those 2 numbers is the spread. Then ask a second operator the same question and you have built, in an afternoon and for nothing, the benchmark that an entire product category has managed not to publish.

The card, briefly

A startup corporate card is usually a charge card rather than a credit card. It settles in full out of your operating account, with no revolving credit and no interest rate, which is why the product exists at all for companies that no lender would underwrite.

Underwriting runs against your balance rather than your history, which explains why a young company holds a limit a bank would refuse, and also why the limit drops when the balance drops. Ask them which day of the month it recalculates, because that date decides whether a spend down before a round closes costs you a limit or not. A company that spends down before a round closes can watch its corporate credit card startup limit fall in the week its costs peak, and the fix is simply to warn the issuer first.

Rewards come out of interchange. The networks publish those rates. On 40,000 dollars a month a 1.5 per cent rebate returns about 600 dollars. I think the rebate is marketing and the 1,100 or so coded transaction records a year are the actual product, though somebody on thin margins would tell me I have that backwards.

Read the agreement for the personal guarantee before anything else in it, because that single clause changes what the product is and it is never the clause anybody leads with in a demo. Deposit backed cards mostly skip it. Cards extending real credit frequently want one, and the difference between those 2 products is worth more attention than the rebate rate that gets advertised on both of them.

A short warning to anyone about to cite this

Sweep spreads appear nowhere I could find. Operators keep their network bank lists private until you sign, which means the one thing you would most want to check beforehand is the one thing you cannot. And nobody publishes what happens to their ledger if they stop trading. That answer has to come out of your own custodial agreement.

Everything else in this piece you can check yourself in an afternoon. The limit and its structure sit in the FDIC's own rules. The institution count and the failure list came from BankFind on 29 July 2026. The complaint figures came from the Bureau's public database for the 12 months to July 2026, aggregation buckets rather than rows, 1 request per product and no key required.

I am not your accountant. None of this is advice on your treasury.

I cannot tell you how common the mistake is, because nobody publishes a survey of what founders believe about the 250,000 limit, and I looked across 5 sources for one. What I can say is that I have made it myself and been corrected in public, and I keep thinking about how confidently I got it wrong in front of 40 people. I still cannot decide whether one hundred banks is a real constraint or just a number that sounds like one.