The Federal Reserve went and asked small business owners about AI in their own words, and the answers are less about robots than about filing cabinets.
The research brief, drawn from the Small Business Credit Survey, found nearly half of small businesses already using AI, with another 15% planning to start within twelve months. The most common uses were exactly what you'd guess: writing and marketing, personal productivity, and planning or analysis. Most owners said productivity went up. Most also said their labor costs stayed flat, which is its own quiet finding.
A third of businesses have no plans to adopt at all — over half of those because they don't see how it applies to their work, and about 30% because they simply prefer not to. Fair enough. But among the owners who did adopt, the barriers they named weren't about the models being bad. They named cost, staff training, time, and not knowing how to implement the thing effectively.
That last one is where the money goes missing.
A fast tool on top of a messy process
Nick Kurkov, who runs the agency iTishniki, put it well in a piece for CPA Practice Advisor this month: "AI can expose weaknesses in a business that were already there." You add a tool expecting hours back. Instead someone is copying a phone number out of the inbox and into the booking system, the website form doesn't reach the customer list, and nobody's sure which spreadsheet is current. You haven't automated anything. You've just made a faster machine that runs on bad inputs.
This is not a technology problem, and no subscription fixes it. It's a decision nobody has made.
For each kind of information in your business, there should be exactly one place that's right. Can you name it?
The twenty-minute version
Open a blank page. Write five categories down the left side: customers and leads, work in progress, money, brand and content, calendar. Next to each one, write the single system that holds the correct version. Not the systems that have a copy — the one you'd trust if two disagreed.
Most owners get two lines in and stall. Customers live partly in the inbox, partly in the booking tool, partly in a spreadsheet, and partly in one employee's text messages. That stall is the finding. You just located the reason your AI tool feels underwhelming: it can only work with what it can reliably reach, and right now nothing can reliably reach the truth.
Then run a second pass and look for retyping. Anywhere a person moves the same information from one screen to another by hand — leads, invoices, customer requests, form submissions — you've found a connection problem wearing the costume of a staffing problem. Those are the spots worth automating, and they're only visible once you know where things are supposed to live.
Pick the ugly one on purpose
The source of truth doesn't have to be sophisticated. A clean, boring spreadsheet that everyone actually updates beats an expensive CRM that three people ignore. What matters is that it's declared out loud, that the whole team knows, and that every other tool — including whatever AI you point at it — reads from that one place instead of guessing.
Once that's true, the AI work gets dramatically easier. You can hand an assistant your customer list and trust the output. You can automate follow-up without wondering which number it'll dial. You can ask for a monthly summary and believe the numbers. None of that was ever gated on a better model.
Nearly half of small businesses are in this now, and more are coming in the next year. The ones who get real hours back won't be the ones who bought earliest. They'll be the ones who spent an afternoon deciding where things live before they automated anything on top of it. The cheapest upgrade available to you this month isn't a subscription. It's a decision.