The riskiest hire is the one you keep reposting. NFIB's August jobs survey had 476 member firms answering through the end of the month. 56 percent of owners were hiring or trying to hire, and the report puts 82 percent of those hiring as finding few or no qualified applicants. 35 percent had openings they could not fill. A seasonally adjusted net 31 percent had already raised compensation. 23 percent named labor quality or availability their single most important problem, against a historical average of 12 percent. I think that quietly changes the question owners are asking. It isn't whether the role is affordable at the number you wrote down. It's whether it's still affordable at the number that actually fills it.

A loan application gets read, not argued. The Federal Reserve Banks surveyed 3,654 small businesses that had applied for financing. 42 percent got everything they asked for. 22 percent got nothing. Among the ones who didn't get all of it, 37 percent named already carrying too much debt. That isn't something you talk a lender out of on the call. It's in the file before anyone picks up. I think that's the part that surprises owners. By the time you're asking, most of the answer is already written down.

A lot of owners think they've handed off the back office. Then someone asks who actually does the work. A small business advocacy group put that question to more than 325 owners: who in your company spends the most time on federal regulatory compliance? 70 percent answered themselves. Accounting, in-house or outside, came to 6 percent each. 51 percent outsource none of it. Hiring the accountant and handing off the work are two different purchases, and I think most owners have only made the first one.

Worry about whether AI is right shows up after you start using it, not before. In the Fed's 2025 Small Business Credit Survey, among the 2,203 employer firms already using AI at work, accuracy was the top challenge, named by 46 percent. Among the 810 firms that only plan to start, accuracy was named by 25 percent. Their leading worry was finding a tool that fits, at 54 percent. People who haven't started are shopping. People who have started are checking. And an IDC survey of finance leaders, sponsored by an accounting software vendor, found 71 percent would reject a tool that was 99 percent accurate if it couldn't explain how it got there. I think that gap is the whole product problem in this category. The ask isn't a better score, it's being able to see the work.

On business returns the IRS barely fines anyone for getting the numbers wrong. It fines them for being late. Fiscal 2025, partnership returns: 50.2 billion dollars of civil penalties assessed. 49.3 billion of that was delinquency, which means the return went in late, not that anything in it was wrong. S corps look the same. 540,452 of 545,882 penalties were delinquency. Accuracy penalties on corporate returns came to 46.5 million, roughly 1 percent of the 4.2 billion assessed that year. I think most owners have this exactly backwards. They stall on filing because the numbers aren't perfect yet, and the stalling is the part that actually costs money. Being roughly right in March beats being precisely right in November.

"Most deals fall apart because the financials are a mess. The seller cannot tell their story clearly." That is Erik Salazar in Forbes. He runs an M&A advisory firm for businesses under 20 million dollars in revenue, so he watches this happen for a living. His fix is boring. Keep the bookkeeping clean from day one, whether that is an AI tool or a CPA. He says it alone puts a seller ahead of 90 percent of the market. McKinsey expects 6 million US small businesses to change hands over the next decade and reckons about 1 million of them are genuinely sellable. I think the gap between those two numbers is mostly a records problem wearing a valuation costume. Nobody wakes up on the day of the offer and reconstructs years of books. The diligence file is whatever you have been keeping all along.

888 receipts and invoices are sitting across 11 company files on one bookkeeping platform, matched to nothing. They aren't spread evenly. 7 of those companies hold all of them and 4 hold none. The largest single pile is 213 documents at one company. That split is the part worth looking at. Unmatched receipts aren't a volume problem, they're a habit problem, and the habit breaks at the company level rather than a little bit everywhere. The cost lands at tax time, when the deduction is real and the paper behind it is in a pile nobody indexed. Count yours. If nobody owns that number it only goes one direction.

A credit card can't owe you money. On one balance sheet, four cards carry ordinary credit balances and the fifth sits at a debit balance of 39,407.88. That's not a mystery, it's a signature. It's the shape a ledger takes when the payment side of a card imports and the charge side doesn't. Every payment goes in. None of the spending does. When that happens the P&L is missing the expenses and the balance sheet is carrying something that reads like an asset. Open your balance sheet and read the credit card section. Any card sitting on the wrong side of zero is telling you half its data never arrived.

21 card balance payments from August are sitting in a clearing account because the other half of the pair never showed up in the feed. The bank side is there. The money left. The card provider's leg does not exist to match against. No amount of categorization work clears that. It is not a judgment call about which account something belongs in. It is a missing record. A backlog is usually a mix of those two things and they get treated as one. One needs somebody to decide. The other needs somebody to go and get the data.

Across 11 company files on one bookkeeping platform, 10 have zero classes defined in their books. The one that does has 6. Classes are the field that answers what a program cost, or a product line, or a location. Without them the P&L can tell you the total went to contractors. It cannot tell you which thing you were building. Nothing breaks when you skip the setup. That is exactly why it gets skipped, and why the gap only surfaces the first time somebody asks a question the books are structurally unable to answer.

An automated categorizer wanted to book a 199.50 card processor payout as Tax Revenue. It was 70 percent confident. A payout is not income. It is money that was already recorded when the charges settled, moving from the processor to the bank. Book it as revenue and you have counted the same dollar twice. The confidence number is the part worth looking at. It is not measuring whether the answer is right. It is measuring how much this description resembles descriptions it has seen before. Widen out from that one row. Across 11 company files on the same platform, 253 transactions were waiting and 137 carried a suggested category that is itself a holding account. On more than half the backlog the categorizer was saying it did not know, in a format that looks exactly like knowing.

I hate when pricing is gated behind a meeting. You want to know what a firm charges, and the only way to find out is to put a call on your calendar. A discovery call where you explain your business to someone who is quietly qualifying you, then a questionnaire, then a proposal a week later if you chase it. Do that with three firms and you've spent a month of your own time on a question that should have taken minutes. Then the number lands and you can't really compare it to anything, because it was built around you. It's shaped by your raise or your revenue, and by how badly you seemed to need the help. So we made ours public. A rate card with every line item on it, and a calculator that gives you your price before you ever talk to us. It costs us something. A funded company and a bootstrapped one might pay the same rate for the same work. I'd still rather that than make you sit through a call to learn something you could have read in 60 seconds.

A founder told me yesterday that he took us seriously because he never had to talk to me first. Our demo is on the site. No signup or meeting needed. We took that idea from @mercury. Gating a demo only filters out the people who wanted to look before they booked. demo.medianfi.com

Something worth checking if you use Stripe. There's often a real balance sitting inside it on the last day of the month that hasn't hit your bank yet. If there's no account for it in your ledger, your revenue is short by every processing fee you've paid, and the fees never show up as a cost.