There is a version of the AI conversation that plays out in boardrooms and conference keynotes, involving large budgets, dedicated teams, and multi-year transformation roadmaps. It gets most of the coverage. It shapes most of the assumptions. And for a lot of smaller businesses, it quietly reinforces a belief that has become its own kind of obstacle: that AI is something you graduate into, once you are big enough, funded enough, or structured enough to take it seriously.
A Federal Reserve survey published earlier this year suggests that belief is already out of date. Nearly half of small businesses in the survey reported using AI in some form, and a significant share of those reported meaningful productivity gains. Not pilot programmes. Not experiments on the roadmap. Actual, working use, in businesses that look a lot like yours.
That is worth sitting with for a moment, because it shifts the question considerably.
The conversation that has already moved on
The ‘are we ready for AI’ discussion tends to assume a threshold, some point at which a business becomes large enough, or organised enough, to begin. But the data does not support that framing. The businesses in that survey were not waiting to reach a threshold. They were using tools that were available to them, applying them to problems they already had, and getting time back as a result.
The tools themselves have made this possible. A lot of what is genuinely useful in AI right now does not require integration work, a technical team, or a significant budget. It requires someone willing to try something specific on a real problem, and the patience to iterate until it works. That is not a size requirement. It is a disposition.
Which means the more honest question is not whether a business is ready. It is whether the people running it have found one or two places where AI actually fits the work they are already doing.
What ‘using AI’ actually looks like at this scale
It rarely looks like the enterprise version. There is no transformation programme, no steering committee, no phased rollout. It looks more like a founder who stopped spending two hours a week drafting client update emails because they found a faster way to get a solid first draft. Or an operations manager who built a simple process for summarising supplier communications so nothing gets missed. Or a small creative team that found a way to move from brief to first concepts without the slow, blank-page start that used to eat half a morning.
None of those examples are dramatic. That is precisely the point. The productivity gains the Federal Reserve survey captured are not coming from grand deployments. They are coming from small, specific applications of available tools to real, recurring friction. The businesses seeing results are not the ones who committed to AI as a strategy. They are the ones who got specific about a problem and tried something.
Why the headline version of this story misleads
Enterprise AI coverage is not wrong, exactly. Large organisations do face genuinely complex questions about AI adoption, governance, and integration. But when that version of the story dominates, it creates a distorted picture for everyone else. It makes AI feel like infrastructure, something you plan for years and implement at scale, rather than something you can apply quietly and practically to the work in front of you this week.
The small businesses in that survey did not wait for the infrastructure conversation to resolve itself. They found something useful and used it. The gap between where they started and where they got to was smaller than the headline version of AI would suggest.
That is probably the most useful thing the data offers: not a competitive warning, not a call to accelerate, but a recalibration of what is actually within reach. The businesses already seeing gains are not outliers with unusual resources. They are ordinary businesses that asked a more practical question than most.
Not: are we ready for AI? But: what is one thing we do every week that takes longer than it should?
That second question tends to have an answer. And the answer tends to be closer to a solution than most people expect.
