There is a question a buyer has always been able to ask about a set of books, and until about a year ago the answer was always a person. Who made the entries?
In the twenty years I spent on the institutional side, that question had a long answer. A controller, a staff accountant, an outside firm at quarter end, an auditor once a year, each with a name, a signature and something to lose. On Main Street the answer was shorter. The owner, a bookkeeper two days a week, a CPA who saw the file in March. Shorter, but still a person, and a person can be asked what they did and why.
That is changing, and it is changing faster than the people who lend against these books have noticed.
What happened last week
On October 27 Intuit told its accountant conference in Las Vegas that its Accountant Suite now runs “more of the monthly close process automatically,” following each firm’s own procedures and keeping an audit log, with accountants reviewing, approving and managing exceptions. The suite, it said, has been adopted by more than 250,000 accounting firms in its first year. The transaction-level work was already there: in July 2025 Intuit put an Accounting Agent into QuickBooks Online that, in its words, “automates bookkeeping and transaction categorization” and “assists in reconciliation.” (CPA Practice Advisor, October 27, 2026; Intuit press release, July 1, 2025.)
Read those two announcements together and the picture is this. For the 36.2 million small businesses in this country (SBA Office of Advocacy, 2025 Small Business Profiles), the entries are increasingly made by software, the reconciliation is increasingly proposed by software, and the month-end close is increasingly run by software, with a human somewhere in the loop whose job is to handle exceptions. I do not think that is bad. The categorization that a bookkeeper did by hand on a Tuesday afternoon was never the valuable part of the job, and it was wrong often enough. But it changes what verification means, and a buyer who does not notice the change will pay for it.
What the agent is good at, and what it cannot know
An agent that categorizes transactions learns from the categorizations it has seen. If the owner coded his truck payments to cost of goods sold for three years, the agent will code the fourth year the same way, faster and with more confidence. If a personal credit card has run through the business since 2019, the agent will keep it running. If a customer deposit was booked as revenue the day it arrived, the agent has no reason to think otherwise. The books will be internally consistent, the reconciliations will tie, and the audit log will show every step. None of that tells you whether the business earns what the books say it earns.
This is the point I keep coming back to. Clean books and true books are different things, and the gap between them is where small-business acquisition prices get set. A buyer paying 3.0x SDE and 4.3x EBITDA at the median (the trade-contractor sales I wrote about last issue; averages 3.3x and 6.9x) is paying a price that assumes the books are a fair account of the business. When software makes the entries, that assumption gets harder to test by talking to someone, and easier to test by doing the work.
The test, on a real set of books
Here is what I mean by doing the work, taught on the composite sample on our site. (Hartwell Mechanical Services is a composite built for illustration; every name in it is fictional; the arithmetic is real.)
Hartwell is a Charlotte heating and air contractor with $4,806,000 of FY2025 revenue and an asking price of $3,600,000. Its books are tidy. Its seller proposed nine add-backs. We did not ask who made the entries; we asked the bank. Every deposit in the bank records was matched to the revenue on the books, every disbursement to the expenses, month by month. The seller’s normalized earnings came out at $730,549 of EBITDA for FY2025 and $783,770 for the trailing twelve months to June 2026. Of the nine add-backs, five were accepted, two were modified and two were rejected, and the reasons are in the report, not in a judgment call.
What the cash proof found that the books could not: the business carries no working capital in the proposed structure, so the day-one cash requirement is $346,871, one month of trailing expenses before depreciation and interest. The largest customer, a property management account, is worth $158,885 of FY2025 earnings, and if it leaves, debt service coverage on the proposed loan falls from 1.25x to 0.98x. At the ask, the buyer is paying 4.9x FY2025 normalized EBITDA and 4.6x on the trailing twelve months, against a DealStats median of 4.6x on the same base. The lender’s coverage standard under the SBA’s SOP 50 10 8.1 is 1.25x. Hartwell clears it by $375 a year.
None of those facts live in the general ledger. A categorization agent, however good, would have produced the same tidy books and the same nine add-backs, because the ledger is the input it learns from. The facts came from putting the bank records, the tax returns and the books next to each other and refusing to let any one of them speak for the other two.
What I would ask for
If you are buying a business whose books are kept in QuickBooks, Xero or anything like them, three things are now worth asking for that were not on the list two years ago.
First, the audit log. Intuit says its close agent keeps one. Ask for it, for the full period, and look at how many entries were approved by a person and how many went through on the default. The ratio is a fact about the books.
Second, the exceptions. An agent that manages exceptions is producing a list of things it was not sure about. That list is the most useful document in the file, and it is the one no one thinks to send.
Third, the bank records, every month, every account. Not because the agent is dishonest but because the agent cannot be. It only knows what it was shown.
Where this goes
I believe the entries will be made by software for nearly every small business within a few years, and that the firms that keep the books will become reviewers of software rather than makers of entries. That is a good outcome for owners. For buyers and lenders it means the one question that used to have a human answer, who made the entries, now has to be answered a different way: by tying the books to the cash and the returns, mechanically, for every month in the period, and then applying judgment to what the tie-out shows.
That is what institutional investors have always done, because at scale no one ever trusted the person who made the entries either. The difference is that it used to cost a team and six weeks. Now it is the cheapest part of the work, and the judgment is the part worth paying for.
The reason a person still has to sit at the end of that chain is not sentiment. Every agent in the process, the one that made the entries and the one that tests them, draws its confidence from the data it was given, and the question in an acquisition is whether that data describes the business. Deciding that means asking the seller why the truck is in cost of goods sold, reading the lease, and knowing from experience which of the nine add-backs a credit desk will reject. A reviewer with a name on the report, and something to lose if it is wrong, is the same safeguard the institutional side has always relied on. The software changed what the reviewer spends time on. It did not change who answers for the number.
G.F.S.
Workup is the analyst’s read on small business acquisition
Gray F. Smith spent twenty years as a senior portfolio manager at some of the largest global hedge funds, including Millennium, Caxton, and J. Goldman, before founding his own fund, Franklin Street Capital Partners. A Harvard economics graduate and Institutional Investor Hedge Fund Rising Star, he is the founder of Workup, bringing institutional-grade diligence to small business acquisition. Learn more at getworkup.ai.