Most writing about AI and small business assumes the only thing standing between an owner and adoption is money or know-how. The Australian data says otherwise, and it says it clearly enough to be worth changing your approach over.
National AI Centre figures put SME adoption at 43% in the quarter to February 2026 — down from 45% the previous quarter. Not a slowdown in growth. An actual decline.
What is actually stopping people
Around 65% of businesses not adopting AI give a reason that comes down to trust: they distrust AI decision-making, or they want a human to stay in control of how the business runs. By comparison, cost accounts for roughly 21% of reported barriers.
That is close to a three-to-one split, and it inverts the usual sales pitch. Making a tool cheaper does not address the objection most owners actually hold.
Why a decline makes sense
A falling number is what you would expect if a wave of businesses tried something, did not trust the output, and stopped. That is not irrational. An owner who cannot verify whether an answer is right is being sensible to distrust it — particularly on anything touching money, staff or compliance, where being confidently wrong is expensive.
The problem is rarely the technology being incapable. It is that nobody set up a way to check the work.
The practical response: pick tasks where you can see the answer is right
The trust problem largely dissolves if you start where verification is cheap and obvious.
- Good starting tasks: drafting something you were going to write anyway, summarising a long document you also have, tidying data you can spot-check, preparing a first version of a quote or proposal you will review line by line. In every case you already know what right looks like.
- Poor starting tasks: anything where you would have to take the output on faith — a tax position, an employment decision, a legal interpretation, a number you cannot independently confirm. Not because a tool cannot help, but because you cannot yet tell when it is wrong.
The pattern is the same one you would use with a new staff member. You give them work you can check, you check it, and you extend the scope as their reliability becomes evident. Nobody hands a new hire the BAS in week one.
Trust is built by checking, not by being reassured
If distrust is the barrier, the fix is not a better sales pitch. It is a routine that lets you find out how reliable the thing actually is on your work, which nobody else can tell you.
A version that takes about twenty minutes:
- Pick one recurring task you already do well. It has to be something you can judge instantly — a quote, a customer reply, a summary of a supplier contract.
- Run it three times, on three real examples. Once is an anecdote. Three starts to show you the pattern, including how it fails.
- Mark each one honestly: would you have sent it as-is, sent it after an edit, or thrown it away? Write the answer down rather than carrying an impression.
- Look at the failures specifically. Did it invent a detail, miss context it was never given, or get the tone wrong? Those are three different problems and only the first is a reason to stop.
Most owners discover the output is neither as good as the marketing nor as bad as they feared — it is a decent first draft that needs a knowledgeable person over it. That is a useful thing to know either way, and it is knowledge you can only get from your own work.
The other two barriers are skills and time, not money
Trust is the largest barrier but not the only one. Skills and time come next, and they matter because they compound: the owner with least time to learn is usually the one whose business would benefit most from removing a repetitive task. Twenty minutes on one task beats a half-day course nobody schedules.
Keeping the human in control is a legitimate design choice
Wanting to stay in control is not resistance to change, and it does not have to be argued out of anyone. It is a requirement you can design for: the tool drafts, a person decides. That arrangement is available today, it addresses the objection two-thirds of non-adopters actually raise, and it happens to produce better results than either extreme.
If you have tried AI in your business and quietly stopped, the useful question is not whether to try again. It is which specific task you could hand over where you would immediately know whether the answer was any good.
Sources
- National AI Centre — AI adoption insights, December 2025 to February 2026 — the source for the 43% adoption figure and the trust-related barriers.
- iStart — SMEs believe in AI, but don't trust themselves to use it — reporting the same adoption data alongside the MYOB Business Monitor cost figure.