What the AI adoption numbers actually count
If you run a small business, someone has told you AI will save you hours. The real question is whether AI automation for small business pays off in yours. Many of the numbers going around don't get at that. They count something else.
The New York Fed's Liberty Street Economics reported in October 2026 that 46% of firms with at least one employee said their business or employees were using AI tools. Another 15% planned to start within the next twelve months. The survey ran September to November 2025, across 5,248 employer firms.
Start with the definition. It changes what the number means. "Business or employees using AI tools" covers one employee asking a chatbot to draft an email. It doesn't mean a system that does a job on its own, and the gap between the two is wide. The Fed's 2026 report on employer firms points the same way. Among firms using AI, about half were still experimenting. Another 44% had partly built it into how the business runs, and only 7% had fully done so.
Other surveys give very different answers on how many businesses use AI, because each one defines "use" differently:
- The Census Bureau found 19.8% of businesses using AI as of May 3, 2026. It's a nationally representative survey of firms of all sizes, and it asks whether a business used AI "in any business function" in the past two weeks.
- Intuit, a software company that sells to small businesses, put it at 77% of U.S. businesses that "use AI regularly", from a survey of more than 34,000 owners across four countries.
The same topic, and the answers run from about 20% to 77%, depending on who asked and what counted. None of it tells you whether the AI paid for itself.
What owners say they got
On the Fed's report page, firms using AI were asked what changed. 71% reported higher productivity, 31% reported higher sales, and most reported no change in labor costs. All of it is self-reported, though. The Liberty Street authors describe the link between AI use and owners' hiring and revenue expectations as an association. They say whether adopters get what they expect is a question for future research.
The way you ask changes the answer. A Goldman Sachs survey of owners in its 10,000 Small Businesses program found that 93% of AI users said AI had a positive impact. The respondents were a self-selected program group, not a random sample. Ask whether AI helped, and most owners say yes. Ask the narrower questions the Fed did, of a broader sample, about sales and labor costs, and the yes gets a lot smaller.
Independent data on what AI automation returns or costs a small business is thin, and that's the part you'd most like to have. The price ranges that turn up in search come mostly from vendor and agency blogs, and that's sales content.
Why some AI projects don't pay off
You may also run into a number from an MIT report: 95% of organizations get zero return from generative AI. Before you apply that to your business, look at how it was built. The headline hides a lot. The report dates from July 2025 and calls itself preliminary. Its sources are 52 organization interviews, 153 survey responses from senior leaders, and a review of over 300 public AI initiatives. It counts a project as a success only if it got past the pilot stage with measurable KPIs. That's a different question from "did it help." Its authors say the figures are "directionally accurate" and not official company reporting. The sample wasn't built around small businesses, so don't read it as your odds. You can read the report itself through a mirror copy.
Closer to a small business, the Fed's AI users report problems too. Among firms using AI, the Fed found that 46% cite accuracy and 43% cite adapting the tools to meet their business needs as challenges. Among Goldman's AI users, 50% named data privacy and security concerns, 49% lack of technical expertise, and 48% difficulty choosing the right tools.
Read together, these point to a couple of places things go wrong: no measure of success, or a tool that doesn't fit how the work gets done. That's a reading of the evidence, not a measured result.
Sometimes the answer is no
The Fed found that 33% of employer firms have no plans to use AI. A San Francisco Fed research brief on the 2024 survey found that firms not using AI cite accuracy and intellectual property concerns, the importance of human interaction, and a view that AI doesn't apply to them. Those can be sound reasons, and some may be yours. Take them seriously rather than buying something just to keep up.
These are judgment calls, not survey findings. Hold off if any of them sound like you:
- The task only comes up a few times a year, so there are few hours to save.
- Nobody on your team will have time to check the output. Since 46% of users name accuracy as a challenge, you'd be carrying that risk with nothing to catch it.
- The task is the conversation itself, and the customer is paying for a person.
- You can't yet describe the task step by step. If you can't write down how it gets done today, automating it won't sort that out for you.
Questions to ask before you buy
Here's a short test. If you can't answer one, that's your answer for now.
- Which one task is this for? Pick a single repeated job and write down the steps. A tool bought for "AI" in general rarely gets the chance to prove itself.
- How many hours does that task take today? Count a normal week. Any savings get measured against that number.
- Who checks the output, and how long does that take? Count the checking time as part of the cost. In the Fed's survey, 46% of AI users cite accuracy as a challenge.
- What result would tell you it worked? Pick the measure before you start, like hours saved or replies sent on time. The MIT report counted a project as a success only if it had measurable KPIs.
- Does it fit how you already work? Adapting tools to the business is a challenge for 43% of the Fed's AI users.
- What data does it touch, and who can see it? In the Goldman survey, 50% of AI users named privacy and security concerns.
- What does it cost in full? Ask about setup, the monthly fee, who maintains it, and what happens if you leave. Get the answers in writing.
So is it worth it?
AI automation for small business can pay off when it's pointed at one named, repeated task, the checking is planned for, and the result gets measured against what the task costs now. From the evidence here, it is more likely to disappoint when it's bought across the board with no task in mind. Nobody can promise you a payback, and none of these surveys can tell you what your business will get.
If you have a task in mind, the AI Workflows page shows how TGC builds repeated tasks into the tools a business already uses, then keeps them running. Or book a call with Jason and talk yours through.
