The National AI Centre published its quarterly read on small and medium business AI adoption, covering December 2025 to February 2026. The headline is flat. What sits underneath it is not.
The numbers
- 44% of SMEs reported some level of AI adoption in February 2026, the strongest month in several. Across the full quarter it was 43%, down from 45% in the September to November quarter.
- Broad adoption, meaning AI embedded across multiple parts of a business rather than used once, hit its highest level in seven months.
- Limited, one-off use is declining. In the Centre’s words, those who have committed are doubling down.
So adoption is not growing, but it is deepening. The people who got in are going further in. The gate is the problem, not the room.
The three reasons the other half are not in
This is the part worth reading twice, because none of the three barriers are what a software vendor would tell you they are.
Trust, at 65%. Around 65% of non-adopting businesses cited either distrust of AI decision making or a strong preference to keep human control of their processes. Not cost. Not capability. Control.
Relevance, at 54%. More than half of non-adopters said AI is not relevant to their business. The Centre calls this the most addressable barrier, and frames it as an absence of visible, relatable examples rather than a rejection.
Not knowing where to start, at 19%, up two points on the previous quarter. The report describes this group precisely: “they aren’t cynical or resistant, they are disoriented.”
The relevance gap tracks by industry. Fewer than 30% of Construction and Agriculture businesses are adopting. More than half of Health, Education and Services businesses are. The Centre’s read on that difference is blunt: it is not capability, it is context.
What adopters are actually doing
Among those using or planning to use AI, the leading applications are content generation and data analytics at 54% each, then cybersecurity and threat detection at 48%. The report notes that agentic AI, supply chain optimisation and AI-assisted HR remain largely untapped.
In other words: almost everybody is using it to make things and look at things. Almost nobody is using it to run anything.
One finding nobody should skip
On responsible AI, the report says practice is ahead of policy. About half of current users check AI outputs before they reach a customer. But transparency with customers about AI use, and formal processes for customers to raise a concern, both lag significantly.
That gap is going to become somebody’s problem. If your inbox is answering customers and you have not told them it is automated, and there is no route for a customer to object, you have built the exact risk this paragraph is describing.
Reading it honestly
We have an obvious interest in this data, so treat the next three paragraphs as argument rather than reporting.
The 54% relevance figure is the one that matters, and it is not a marketing problem. When a builder says AI is not relevant to their business, they are usually right about the thing they are picturing. Nothing about a chat window is relevant to a builder. What is relevant is that their quotes go out three days late and two enquiries a week die unanswered, and nobody has ever described that to them as an AI problem, because it is not one. It is a systems problem that AI happens to be good at.
The 65% trust figure is the reason we publish the bounds rather than the capability: a ceiling in dollars, gates on anything irreversible, and a readable trail. The people in that 65% are not asking whether it works. They are asking who is in control.
And the 19% who do not know where to start is why the most useful page we have is a day by day breakdown of an install rather than a feature list. Disoriented people need an order of operations, not a demo.
