Small business · AI adoption

The Fed Asked 5,248 Small Businesses About AI. The Firms Using It Are the Most Likely to Hire

The Federal Reserve's first dedicated AI survey of small firms found that businesses using AI are the most likely to hire, that most firms see no change in their payrolls or sales yet, and that the strongest optimism comes from the companies in the weakest shape.

Coffee Cup AI · October 30, 2026 · 8 min read

In fall 2025, the Federal Reserve Banks added a first-of-its-kind artificial intelligence module to the Small Business Credit Survey, creating evidence that can inform AI consulting and implementation planning and ground the debate over AI job loss and workforce change. The annual questionnaire reaches firms with fewer than 500 employees, the companies that make up 99.7 percent of U.S. employer establishments. 5,248 firms answered detailed questions about whether they use AI, what they use it for, and what it changed. Three Fed research papers built on those answers landed between July and October 2026, from the San Francisco, Philadelphia, and New York Fed.

Taken together, the results embarrass both sides of the AI debate. The hype camp, which sells instant transformation, meets a population where 57 percent of adopters describe themselves as merely "experimenting with the technology" and only 7 percent have fully integrated it. The doom camp, which predicts AI-driven job destruction, meets something stranger: small businesses that use AI are the most likely to say they plan to hire, and the enthusiasm is strongest among firms in poor financial condition.

The results describe AI as a capacity tool: a way for thin-staffed companies to buy the skills they cannot hire. That reading comes straight from the New York Fed researchers themselves, and it carries a practical lesson for any owner deciding whether to adopt, and how.

What the survey is, and why it deserves more weight than another vendor poll

The Small Business Credit Survey has run annually since 2016 under the Federal Reserve Banks, most recently coordinated out of the Cleveland Fed. It is not a marketing study. The 2025 wave collected responses between September 3 and November 14, 2025, weighted the results to represent all U.S. small businesses, and the Fed has now published three analyses of the AI module: a qualitative brief from the San Francisco Fed on owner attitudes, a performance and labor study from the Philadelphia Fed, and an expectations analysis from the New York Fed's research group.

The Fed is plain about the survey's limits: it draws from a convenience sample, and much of the most interesting material is about expectations, what owners think the next twelve months will bring, with realized results still thin.

The reality check on the hype: AI at small firms is mostly clerical, and mostly unfinished

On adoption, forty-six percent of small employer firms reported that their business or employees were using AI tools, another 15 percent planned to adopt within a year, and roughly one third said they have no plans to adopt at all (2026 Report on Employer Firms). The San Francisco Fed's qualitative work found owner sentiment spanning "enthusiasm to opposition," including firms that deliberately opted out over accuracy concerns, intellectual property worries, or because human interaction is the product.

The headline 46 percent is modest, and so is the depth beneath it. Among adopters, 57 percent describe their adoption as experimenting. Only 7 percent say AI is fully integrated.

Writing, notetaking, research, admin. Here, AI is clerical support. The Philadelphia Fed found that among adopters in its region, 44 percent said AI was not even important to their core goods and services. The pattern mirrors what the San Francisco Fed captured in owner quotes: genuine enthusiasm ("We use it for everything that we can pretty much"), applied to ordinary work.

The realized results so far match that modesty. The survey asked firms what changed because of AI. Most firms said nothing changed: 76 percent reported no change in labor costs, 70 percent no change in sales, 72 percent no change in customer engagement. Among the firms that did see movement, productivity gains were the dominant outcome (70 percent of changers), with 30 percent reporting increased revenue.

The doom check: AI users are the small businesses planning to hire

The New York Fed researchers, Will Aarons and Asani Sarkar, compared the twelve-month-forward expectations of firms that use AI against those that do not. The gap is large on both jobs and revenue:

Firms using AI show a net expectation of higher employment 18 points above non-users, and a net revenue expectation 27 points above. And this sits against a bleak backdrop: the same survey found that small-business expectations for both revenue and employment growth fell to their lowest levels since the 2020 survey, with tariff-driven cost increases hitting more than four in ten firms. Against the gloomiest Main Street mood in five years, the AI-using subset is the one leaning toward hiring.

Aarons and Sarkar anticipated the obvious objection: maybe optimistic people are the ones who adopt AI, and the technology is an innocent bystander. They tested that four ways, and the gap survives every test. Controlling for owner and firm characteristics (age, profitability, size, location) leaves a 13-point employment gap. Adding the firm's actual hiring over the prior year leaves 10 points. The revenue gap holds at 14 points after the same treatment.

The data punctures the optimism story from the other direction too: if AI adoption were just a symptom of healthy, confident firms, the user/non-user gap should be widest among firms doing well. Instead, among owners who describe their current financial condition as fair or poor, AI users exceed non-users by 21 points on employment expectations. Among firms in good condition, the gap is only 10 points. For revenue, financially strained firms show a 24-point gap, and healthy ones 15.

The firms getting the biggest expectation boost from AI are the ones that could not hire their way out of their problems: a small contractor without a marketing person, a shop whose owner does the books and the social media at midnight. For them, AI stands in for staff they could never hire. The technology is filling capability gaps, and the revenue optimism concentrates exactly where those gaps are.

And the workers? Skills shift, headcount holds

The Philadelphia Fed asked the labor question head-on, asking AI-using firms how adoption changed their demand for labor. The answers describe recomposition.

Work is changing shape at these firms. Certain skills and certain job types are shrinking, which is real displacement pressure on specific tasks. But at the level where it hurts, the headcount, the dominant movement is neutral or positive: 3 percent cut jobs, 17 percent plan to add them. Combine that with the national hiring expectations and the picture is consistent: small firms deploy AI to raise capacity per person.

Strong expectations, thin results so far

The strongest results above are expectations, and the survey itself says realized gains have been thinner. Only 30 percent of firms report that AI increased revenue so far, against the 48-point net share that expects revenue growth ahead. Owners can be wrong, and self-reports tend to run generous.

The selection problem deserves respect too. Even with controls, no survey of this kind separates what AI caused from what adopters were already inclined to expect. The Fed's tests make pure optimism an unlikely explanation, but they do not establish causation, and the convenience sample means the respondents are self-selected. Treat the direction of the evidence as strong and the size of any effect as unsettled.

What this means for an owner deciding on adoption

Nobody is as far ahead as the hype implies. More than half of adopters are still experimenting, and a third of firms have no plans at all. If you have not moved yet, you are squarely in the mainstream.

Adopt against a constraint. The optimism premium sits with firms that use AI to close a specific gap: marketing they cannot staff, analysis they outsource or skip, writing that eats the owner's evenings. The Philadelphia data shows what works in practice starts with writing, individual productivity, and planning, at 80, 60, and 50 percent of adopters. Pick the one bottleneck whose removal would unlock revenue, and put the tool there.

Expect the time savings before the money. Seventy percent of firms with any change report productivity gains, only 30 percent report revenue gains. Budget the project against a baseline of hours saved and output produced, and treat revenue lift as the second milestone. That expectation discipline is also why you should measure before you expand a pilot.

If you manage people, the labor lesson is recomposition. Certain skills will shrink in value; plan the retraining and the redeployment now, because the firms in the data are holding headcount while the work inside it changes. The small businesses that pair AI with hiring are the ones betting that capacity compounds.

The Fed's survey does not settle the AI question for small business. In the meantime, it shows what the median firm is doing, and the most hopeful results are landing at the firms that needed a hand the most. The next wave of survey data will show whether those expectations turn into results. Until then, the evidence points to narrow adoption, honest measurement, and hiring for the work the tool frees up.