The month ended ten days ago. The numbers you need to decide on a hire, a price change, or a slow season are still being keyed in from statements and receipts. By the time the books close, the month they describe is already history.

For a lot of small businesses along the lakeshore and the I-94 corridor, that is simply how the close works. One person, often part-time, often also doing payroll and answering the phone, spends the first two weeks of every month rebuilding the last one. The cost is not only their hours. It is every decision that waits on numbers that arrive late.

“Use AI for your books” is not a plan, though. Some of the close is ideal work for software. Some of it should never leave a person’s desk. The useful question is which parts are which.

What the research actually found

The best evidence so far comes from a study of accountants working with small and mid-sized firms, now published in the Journal of Accounting Research. Researchers from Stanford and MIT combined a survey of 277 accountants with field data from 79 firms and hundreds of thousands of transactions. Generative AI adoption was associated with a 7.5-day reduction in monthly close time, along with a 55% increase in weekly client support.

Two details in that study matter more than the headline. First, the time came back from data entry: roughly 8.5% of accountant time shifted away from routine entry toward higher-value work. Second, the accountants did not hand over the keys. Experienced users stepped in more when the software reported low confidence in its own answer. The books got faster because people stopped typing, not because people stopped checking.

The other side is worth stating too. In the Census Bureau’s survey of workers, among those who used AI on the job in the past week, 10% said it saved them no time and 4% said it made the work take longer. Pointing a tool at a messy process does not tidy the process.

Where the close splits

Every close has the same rough shape. Here is how it divides, in our experience, between work a system can carry and work that stays with a person.

When it isn’t the answer

If your close is a few dozen transactions and an afternoon of work, leave it alone. The setup, the testing, and the monthly review of exceptions can easily cost more than the afternoon. A spreadsheet and a calendar reminder are a perfectly good system.

It also isn’t the answer when the chart of accounts is a mess, or when nobody agrees what goes where. Automating a muddled process gives you muddled numbers faster. Fix the categories first, by hand, with whoever does the books. That alone often takes days off the close.

And it never replaces the review. Any setup that posts entries nobody looks at is not saving time; it is moving the risk to the day an auditor, a lender, or a tax notice finds it.

This is about capacity, not headcount

The person who does your books is not easy to replace. The Bureau of Labor Statistics projects employment of bookkeeping, accounting, and auditing clerks to decline 6% from 2025 to 2035, yet still expects about 144,100 openings a year, mostly to replace people retiring or changing careers. Fewer people are entering the work while experienced ones leave it. Taking the data entry off the person you already have is a way to keep them on the work that needs their judgment.

Start with one piece

You don’t need to rebuild your accounting. Pick the step that eats the most hours, usually receipt capture or bank matching, and fix that one inside the software you already use. Write down how it works so your bookkeeper can run it without us. Measure the next close against the last one. If the days come back, do the next piece. If they don’t, you’ve learned that cheaply.

A note on the numbers: studies of AI at work disagree with each other because they measure different people, different tasks, and different definitions of “use.” Every figure above links to its primary source so you can check the definition yourself. We keep a running Sources page grading each citation by its distance from original research.