The Fed Counted How Many Small Businesses Use AI. The Real Story Is What Sits Underneath It.

Estimated read time: 7 minutes

New Federal Reserve research put a number on something most small business owners already suspected: close to 40 percent of the small businesses surveyed are either using artificial intelligence already or planning to. That is the line every outlet ran with. The more useful finding sits three paragraphs down and has almost nothing to do with AI. When owners were asked what was actually getting in the way, they named software costs, staff training, system upgrades, and plain uncertainty about how to implement any of it.

Not one of those is an AI problem. They are plumbing problems. The AI purchase did not create them. It just turned the lights on.

What the Fed actually counted

The research came out of the Small Business Credit Survey, which is the closest thing the United States has to a census of how small firms actually operate rather than how they say they operate on LinkedIn. The qualitative findings put AI use and planned use at roughly four in ten respondents, spread across marketing, customer service, analytics and general operational work.

That is a real number and it is worth sitting with, because it is lower than almost every vendor-published figure you have seen this year. The Fed is not selling anything. It also asked the follow-up question that vendor surveys tend to skip: what is stopping you. The answers were costs, training, system upgrades and uncertainty about implementation.

Read that list again. Three of the four are about the business, not the technology. Only “software costs” is about the AI itself, and even that is usually a symptom of buying tools one at a time without a plan.

Why one survey says 40 percent and another says 77

Intuit’s 2026 AI Impact Report puts regular AI use among US small and midsize businesses at more than three in four, up from roughly half in mid 2024. The Fed says four in ten. Both are probably right, because they are counting different things.

“Regular use” in a vendor survey often includes an owner who pasted a customer email into ChatGPT last Tuesday. The Fed is closer to measuring AI as an operational commitment, something wired into how the business runs. The distance between those two numbers is the entire story of small business AI in 2026: an enormous amount of casual usage sitting on top of a very small amount of actual integration.

Clutch found the same shape from a different angle in August, reporting that a large share of small businesses already using AI still have visible gaps in data readiness, governance and overall technology strategy. In other words, they bought the tool and skipped the part that makes the tool work.

AI does not organize your data. It reads it faster.

Nick Kurkov, who runs the agency and software firm iTishniki, put it bluntly to CPA Practice Advisor this week: “AI can expose weaknesses in a business that were already there.” His version of the failure is one most owners will recognize instantly. You add a tool expecting to save time, and then somebody is still copying information between systems, the website still does not talk to the CRM properly, and customer records still live in three places with no agreement on which one is correct.

The specific thing that breaks is the source of truth. An AI assistant answering a customer question has to read something. If your product pricing lives in a spreadsheet, your inventory lives in the point of sale system, and your shipping promises live in an email signature somebody wrote in 2023, the assistant will confidently answer from whichever one it can reach. That is not a hallucination. That is your filing system, out loud, to a customer.

Before you connect anything to anything, pick the system that wins ties. One system per category of information. Customers here, inventory there, financials in the third place, and everything else reads from those. It is an unglamorous afternoon of work and it is the single highest-return thing on this list.

The copy and paste tell

There is one diagnostic that costs nothing and works on every business. Watch what your team retypes.

Every time somebody moves the same information from one screen to another by hand, that is an integration gap wearing a staffing costume. The lead comes in through the website form, gets typed into the CRM, gets typed again into the invoicing tool, and gets typed a third time into the spreadsheet you use to track which jobs are done. Four systems, one fact, three opportunities to get it wrong.

Owners usually read that as “we need more help.” It is much more often “these two tools have never been introduced.” The fix is a connection, not a hire, and the going rate for a connection is roughly twenty dollars a month, which is a useful thing to know before you post a job listing. If the retyping is happening because two tools both think they own your records, our breakdown of Notion vs Airtable is a useful way to think about which one should.

Keep a list for one week. Every retyped field goes on it. At the end of the week you will have a ranked, evidence-based automation roadmap that cost you a notepad.

Almost nobody is measuring the return

Gartner research reported by the Wall Street Journal found that 85 percent of the functional leaders surveyed planned to increase AI spending in 2026, while 23 percent could not say what return their existing AI investments were producing. That is a large company statistic, but small businesses do not get a pass here. They get a harsher version of it, because a two hundred dollar a month mistake is proportionally much worse at eight employees than at eight thousand.

The reason nobody measures is that “AI” is not a measurable thing. “Time from customer inquiry to first reply” is measurable. “Invoices sent per week without manual data entry” is measurable. “Percentage of quotes that go out the same day” is measurable. Pick the number before you buy the tool, write it down somewhere you will see it again, and check it in sixty days.

If you cannot name the number, that is not a reason to buy carefully. It is a reason not to buy yet.

The 45 minute audit before your next AI purchase

This is not a consulting engagement. It is four questions and a sheet of paper.

1. Where does each kind of information actually live? Write one line per category: customers, orders, money, files, conversations. If a category has more than one home, mark it. Those are your breakage points.
2. What does the team retype? The list from the previous section. Rank by frequency, not by how annoying it feels.
3. Is the underlying software still the right size? A setup that worked at five people frequently becomes the bottleneck at twenty. The expensive part is rarely the new tool. It is keeping five old ones alive because replacing the process underneath them feels disruptive.
4. What number should move, and by how much? One metric. A target. A date to check it.

If you get through all four and the answer is still “yes, buy the AI tool,” buy it with confidence. You will be in a small minority of businesses that actually knew what they were buying it for, and your odds of it working are dramatically better than average. Our SaaS stack audit guide is the companion piece if step three turned up more than you expected.

If you are already three tools deep

Most readers are not standing at the start line. The typical small business is already running several tools with AI features switched on by default, often without anyone deciding to switch them on. Your accounting software started categorizing transactions. Your email started drafting replies. Your scheduling tool started suggesting slots.

That is fine, and mostly harmless, right up until two of those tools disagree about a customer. Then you have an AI-flavored version of a problem that used to be solved by one person keeping the master spreadsheet. The move is not to rip anything out. It is to run the same four questions retroactively, decide which system wins ties, and turn off the AI features that are writing into systems you have not designated as authoritative.

The businesses that are getting real returns from AI this year are not the ones with the most tools. They are the ones with the fewest arguments between their tools.

Frequently asked questions

Does this mean I should wait to adopt AI?

No. It means you should spend the first afternoon on your data instead of on a demo. Waiting is its own cost, and the Fed numbers suggest a meaningful share of your competitors are already past the starting line. The point is sequence, not delay.

What is data readiness in plain language?

It means your information is in one findable place per category, is reasonably current, and has an agreed owner. That is it. It does not require a data warehouse, a consultant, or the word “governance.”

How much should a small business spend on AI tools per month?

Most land somewhere in the low hundreds once you count the AI features bundled into software they already pay for. The more useful question is what share of that spend has a named metric attached to it. If the answer is none of it, the amount is beside the point.

Is an integration problem really cheaper to fix than hiring?

Usually, yes, when the work being duplicated is moving structured data between two systems that both have modern connections. It stops being true when the work requires judgment. The retyping list from above separates the two cleanly.

Related Coverage

• Is Your Favorite Business Software About to Shut Down? How to Audit Your SaaS Stack for AI Risk. The long-form version of question three, with a scoring method for tools you are not sure about.
• Best Business Intelligence and Reporting Dashboards for Small Business. What to use once your data actually has a single home and you want to see it.
• Should You Hire or Use AI? The Real 2026 Small Business Guide. For the retyping list that turns out to need a person after all.

Faceted Media Magazine covers business, AI, and entrepreneurship for the people building what’s next.