How AI Is Changing Month-End Close for Small Businesses

AI is changing which parts of the month-end close a human touches. It has not yet changed when the close finishes, and for most small businesses it has not changed anything at all. Those three claims sound contradictory. They are all supported by the available evidence, and holding them together is the only honest way to talk about this.

Two decades of automation did not move the calendar

Start with the number that should embarrass the whole category. ISG Software Research, formerly Ventana Research, reported in December 2023 that 59 percent of organizations complete the monthly close within six business days, and described that against its 2019 finding of 60 percent as “a statistically insignificant difference.” Its own conclusion was blunt: organizations “have made little progress in the decades-long quest to close sooner.” ISG does not publish sample sizes for either wave, which is a real limitation on how hard you can lean on those figures.

APQC, an independent nonprofit benchmarking body, puts the median cycle time to complete a monthly financial close at 8.0 days across a sample of 3,389 organizations. Whatever finance departments bought between 2015 and 2023, it did not buy them a materially faster close.

That is the baseline any AI claim has to beat. It is also the reason to distrust the vendor statistics currently circulating. A widely repeated figure claiming 97 percent of finance departments adopted AI in 2026 does not trace to any identifiable source and is contradicted by every credible measurement below by a factor of three or more.

What the credible adoption data says

The best measurement available is a government one. Census Bureau researchers, in working paper CES-26-25 published in April 2026, surveyed more than 117,000 distinct firms between November 2025 and February 2026. They found 18 percent of firms using AI in a business function, rising to 32 percent when weighted by employment. Finance and insurance sat at 30 percent current use.

The employment weighting is the whole story. A separate Census analysis published May 26, 2026 found 37 percent of firms with 250 or more employees using AI, against fewer than 20 percent of firms with four or fewer employees, and reported that “AI use increased among firms with at least 20 employees but didn’t change significantly among firms with fewer than 20 employees.” The same Census working paper notes that about 75 percent of US firms have fewer than 10 employees.

So when a survey of finance leaders reports high adoption, check who was asked. Gartner’s November 2025 release found 59 percent of finance functions using AI, from a sample of 183 CFOs and senior finance leaders. That is a real finding about large finance organizations. It is not a finding about the company with a bookkeeper and a QuickBooks file.

Large companies are running AI in finance. Small ones, in aggregate, are not. Any article telling a ten-person business that it is behind is arguing with the Census.

Where the close did get faster

The strongest evidence that AI moves the close comes from a peer-reviewed field study, not a vendor deck. Jung Ho Choi of Stanford and Chloe Xie of MIT, in work accepted at the Journal of Accounting Research, surveyed 277 accountants and analyzed transaction-level data from 79 small and medium-sized enterprises. They measured a 7.5-day reduction in monthly close time and a 12 percent increase in ledger granularity, with roughly 9 percent of accountant time reallocated away from routine data entry.

Two cautions before anyone quotes that against the 8.0-day median above. The two figures measure different populations and different things, so subtracting one from the other implies a near-instant close, which is not what the paper found. And the 79 firms were all customers of a single AI accounting software vendor that supplied the data, so this is not a random sample of small businesses.

The mechanism is narrow and worth naming precisely. The automated task is transaction classification, with the model attaching a confidence score, and experienced accountants intervening selectively where confidence is low. That is not an AI closing the books. That is an AI proposing a coding for every line and a human adjudicating the uncertain ones.

The finding the marketing leaves out

The same study ran a framed field experiment and reported that accountants sometimes over-rely on inaccurate AI-suggested classifications. The tool that speeds you up is the same tool that makes a wrong answer look pre-approved.

This is the part of the story that deserves more weight than it gets. A misclassified transaction in a manual close gets caught because someone had to think about it to enter it. A misclassified transaction with a high confidence score gets waved through, and the error surfaces in a variance review three months later, or in a due diligence process two years later. Speed and accuracy are being traded against each other, and the trade is currently invisible in most vendor claims.

AICPA and CIMA research with NC State, released February 2026 from a sample of 1,735 executives, found fewer than one in five smaller organizations had the AI-skilled talent or systems they considered necessary. The firms adopting fastest are frequently the ones least equipped to audit what the model did.

The labor math underneath all of this

There is a reason this is happening now, and it has less to do with model capability than with hiring. The AICPA’s 2025 Trends report, released October 27, 2025, counted 55,152 combined bachelor’s and master’s accounting graduates in academic year 2023 to 2024, down 6.6 percent, with master’s degrees in accounting or taxation down roughly 15 percent to 14,335. Declines of 9.6 percent and 7.4 percent preceded it. New CPA exam candidates fell from 42,626 in 2023 to 28,082 in 2024.

Meanwhile the Bureau of Labor Statistics projects employment of accountants and auditors growing 5 percent from 2025 to 2035, faster than the 3 percent average across all occupations, with about 115,300 openings a year.

Fewer people entering, more roles to fill. Automation of classification work is not primarily a cost-cutting story for small businesses. It is a coverage story, and it explains why adoption is concentrated in the firms that cannot hire fast enough rather than the ones with the biggest software budgets.

What this means for a small business closing its own books

The honest position: AI will not compress your close from ten days to two, and anyone promising that is selling. What it plausibly does today is take the first pass at categorizing transactions, flag anomalies, and hand you a shorter list of things to think about.

For ecommerce businesses the effect shows up earliest in the areas with the highest transaction counts and the least judgment per transaction, which is marketplace settlement data. Tools built for that work, ConnectBooks and Bookkeep among them, pull Amazon, Shopify and Walmart settlements apart into sales, fees, refunds and COGS so the classification problem shrinks before anyone opens the ledger. The gain there is real and it is unglamorous.

Three things to hold onto. Measure your close before and after, because the baseline is the only defense against a vendor’s arithmetic. Keep a human on anything the model was unsure about, and keep the confidence scores where an auditor can see them. And judge the tool on whether the numbers tie back to the source report, not on how quickly it produced them.

The close will get shorter eventually. It has not yet, and pretending otherwise makes it harder to notice when it finally does.