When Money's Tight, the AI Bet Gets Bigger

It is planning season for a lot of small businesses right now, and somewhere in a lot of those plans is a line with no number next to it. Next year we do more with the same headcount, because AI picks up the slack.

That line is carrying a lot of weight. Research the New York Fed published this morning suggests it is carrying that weight in a lot of plans at once, and that it is heaviest in the plans with the least room to be wrong.

Owners using AI see a different year than owners who aren't

The data comes from the 2025 Small Business Credit Survey, the Fed system's annual survey of small firms. Economists Will Aarons and Asani Sarkar looked at the 5,248 employer firms that answered its AI questions. Forty-six percent said the business or its employees use AI tools, with another 15 percent planning to start within twelve months.

Then they compared what the two groups expect. The measure is a net figure, the share of owners expecting an increase minus the share expecting a decrease, looking out twelve months.

On employment, AI users came in at plus 33 points. Non-users, plus 15. On revenue the spread was wider: plus 48 against plus 21. Controls for firm and owner characteristics narrow the gaps but do not close them.

Two sets of owners, same economy, same year ahead. One set has AI in the building and sees a noticeably better one.

The gap is widest where the cushion is thinnest

This is the part worth slowing down for. The authors split the sample by how owners rated their own firm's financial condition.

Among owners who called their condition fair or poor, the employment gap between AI users and non-users was 21 points and the revenue gap was 24. Among owners in good shape or better, those same gaps were 10 and 15.

So the businesses under actual pressure are the ones leaning hardest on AI as the thing that turns the year around. Roughly double the optimism gap of their healthier neighbors.

One more finding sits underneath it. Among AI users, the owners who had genuinely woven AI into their processes were no more likely to expect higher employment than the owners still poking at it, once you account for everything else. The optimism is not tracking how far along anyone actually is. It tracks having started.

What has shown up so far is smaller

The same survey asked those owners what AI has done for them to date.

Seventy-seven percent reported no change in their firm's labor costs from using AI. Thirty-one percent reported increased sales. Among users, 63 percent called AI somewhat or very important to production, and 51 percent had integrated it partially or fully into a business process.

None of that is a failure. Important to the work and not yet visible in labor costs is a completely normal place to be in year one of anything. But it is a long way from plus 33 and plus 48.

Two caveats worth stating plainly, both of which the authors raise themselves. The survey was fielded between September and November of 2025, so it is a photograph of last fall and the tools have moved since. And these are associations, not proof of cause. Owners who adopt new tools early may simply be the kind of owners who expect to grow.

Even with both of those, the shape is hard to miss. Expectation is running ahead of measured result, and it is running furthest ahead in the businesses that can least afford a miss.

Point it at one number you already watch

The answer here is not to be less optimistic. Most of those owners will get something real out of AI. The answer is to make the optimism checkable, so that in twelve months you know rather than feel.

Pick the number the tool is supposed to move. Not efficiency, not productivity. A number already on your desk: hours a week spent writing quotes, days from job complete to invoice sent, share of inbound calls answered by a person, admin cost per job.

Write down where that number sits today, before anything changes. Almost nobody does this, and without it you cannot tell a working tool from a good month.

Give the tool one job that touches that number. One workflow, start to finish, with a person reviewing the output until reviewing it gets boring.

Put a date on the check. Sixty or ninety days out, on the calendar, with the baseline written next to it. On that date the number has moved or it has not, and both answers are worth having.

If your plan only works when AI delivers

Say that out loud now rather than in March.

A tool that takes four hours a week off your plate is real and it is worth paying for. A tool that closes a cash gap on its own is a different claim, and no survey data I have seen supports it yet. The first one you can buy this month. The second one is a bet, and bets belong in a plan labeled as bets.

That distinction is most of the difference between the owners who will be able to point at a result next fall and the owners who will still be describing a feeling.

The Fed's numbers say small business owners are betting on AI harder than anyone, and hardest of all when the money is tight. Fair enough. Just write down what you are betting on, and when you will know.

If you want help picking that one number and the workflow behind it, that is most of what I do. Start at nexeraintelligence.com.