Plenty of small firms have tried AI. Far fewer run on it, and the 2026 surveys show how wide that gap is. Each figure below stays with its own survey, since each one defines “use” its own way.
Most firms are still testing
The Federal Reserve’s Small Business Credit Survey drew 6,525 responses in the fall of 2025. It found that 46% of US small employer firms used AI for work. Among those users, 49% were experimenting, 44% had partly integrated it into their processes, and just 7% had fully integrated it.
Two other surveys tell a similar story. In the Goldman Sachs 10,000 Small Businesses survey, 76% of owners used AI, but only 14% said it was fully embedded in their core operations. In the UK, the British Chambers of Commerce found 54% of firms using AI. Most firms rely on generic tools, and only around one in ten has a system built for its own business.
Depth is where the payoff shows
The OECD’s 2026 survey of SMEs found that 56.6% of AI users apply it only to isolated tasks, and 19% across several functions. In its Japanese sample, 2% of firms using AI for isolated tasks reported transformational benefits. Among firms using it enterprise-wide, 23% did. The OECD warns that its sample isn’t representative and that the benefits are self-reported, so treat the gap as a signal, not a law.
Lighter use isn’t wasted, by owners’ own account. Among the Fed’s AI users, 71% reported higher productivity and 31% higher sales.
What gets in the way
The Fed’s users named accuracy (46%) and adapting tools to the business (43%) as their top challenges. Firms planning to start cited finding the right tools (54%) and the time to set up and train staff (37%). In the Goldman Sachs survey, 73% of owners said they’d benefit from more training and implementation help.
A 90 day plan
This plan is our advice, built on the obstacles above. It isn’t a measured recipe.
- Days 1 to 30: pick one process and one owner. Choose a recurring job, like answering quote requests or writing product pages. Writing and marketing is the most common use among the Fed’s AI users, at 83%. Name one person to own it, write down what data may go into the tool, and record how long the job takes today.
- Days 31 to 60: put AI inside the process. Make the tool one step of the job itself, used every time the job comes up. Add a human check where accuracy matters, keep a count of what it catches, and put training hours on the calendar.
- Days 61 to 90: measure and decide. Compare time, output and errors with your starting point, then decide whether to keep the process, fix it or drop it. If it stays, write the method down so someone else can run it, and pick the next process in another part of the business.
For the adoption figures behind this, see our five numbers every SME owner should know about AI and why Europe’s small firms lag on AI.
All sources were re-read on November 18, 2026. The Goldman Sachs figures come from the wire copy of its March 17, 2026 release. Corrections to hello@bearingbridge.com.







