On October 8, the Federal Reserve Bank of New York published a piece on its Liberty Street Economics blog looking at what small business owners who use AI expect to happen next. The data comes from the 2025 Small Business Credit Survey. The AI questions went to 5,248 firms with at least one employee.
Start with the adoption number. 46 percent said their business or employees used AI tools. Another 15 percent planned to start within twelve months. Among the firms already using it, 63 percent called it somewhat or very important to what they produce, and 51 percent had worked it into their processes partly or fully.
That is the setup. The finding is the gap.
What the gap actually shows
Ask owners whether they expect to employ more people over the next year and net out the ones who expect fewer. Among AI users, that net figure was plus 33 points. Among non-users it was plus 15. On revenue, AI users came in at plus 48 against plus 21.
A gap that wide usually means you are comparing two different kinds of companies. Bigger firms adopt earlier. Younger firms adopt faster. Owners who were already growing are more likely to try new tools and more optimistic overall.
So the authors controlled for firm and owner characteristics, and the employment gap fell to 13 points. Then they controlled for what the firms had actually done in the prior year, and it fell to 10. The revenue gap settled at 14 points. Smaller than the raw number, and still there after you take out the obvious explanations.
The part worth sitting with
The gap was widest among the firms in the worst shape.
Owners who described their financial condition as fair or poor showed a 24-point revenue gap between AI users and non-users. Owners in good condition showed 15. On employment, the gap was 21 points, versus 10. Among AI users, those who reported trouble with technology expected 16 points more revenue growth than those who did not.
Read that the way an operator would. The tools appear to be doing the most for the businesses with the least slack — the ones where nobody has time to chase invoices, nobody is answering email after six, and the owner is doing three jobs. That is a constraint problem, not a technology problem, and it is the argument I have been making for two years. The question is never which model is best. It is which hour of your week gets eaten by something that doesn’t need you.
Two more numbers. Thirty-one percent of respondents said AI use had increased their sales. Seventy-seven percent said it had not changed their labor costs at all. Nobody is being replaced. The work is just getting done.
The caveat the Fed put in writing, and you should too
These are expectations. Not results. The authors say it plainly: whether these firms actually get the revenue and employment they anticipate is a question for later research. Optimism is not revenue. I have watched plenty of owners feel good about a new system in month two and quietly abandon it by month six.
What separates the two outcomes is usually not the tool. It depends on whether it got wired into how the business runs or sat beside it. Fifty-one percent said they had integrated AI, at least partly. The other half have a subscription. If you are picking AI solutions for small business use, the setup work — the accounting, the task management, the CRM, the places the handoffs happen — is where the return comes from, not the model you chose.
Start by finding the hours. I wrote AI Today for exactly this: where to begin, which hours in your week you can automate, and how to measure whether any of it paid off. The Fed data says the owners doing this expect a better year than those who aren’t. In twelve months, we will know whether they did.