Short answer: AI readiness is not really a technology question, and most readiness checklists ask the wrong things. The official UK numbers show why. The ONS found about 35% of businesses with ten or more employees using at least one AI technology by June 2026, but the average adopter used only 1.6 technologies against 1.4 in late 2023 — adoption that is wide and very shallow. DSIT, counting deliberate deployment rather than self-reported use, puts it at 16%, with 80% of businesses neither using AI nor planning to. The figure that should decide your next move is this: among businesses that have adopted AI, 75% report improved workforce productivity and 77% report no change at all in revenue. Readiness, in practice, means being able to answer five questions before you spend anything — what work changes, whose data it touches, who checks the output, what happens when it is wrong, and how you will know it worked. If you cannot answer them you are not unready for AI; you are unready for a build, which is a different and far cheaper problem to fix.

Three UK surveys, three different answers.

If you look up how many UK businesses use AI, you will find three numbers from the same year, and they do not agree.

The Office for National Statistics reported in July 2026 that about 35% of businesses with ten or more employees used at least one AI technology, up from about 12% in late 2023. It comes from Wave 159 of the Business Insights and Conditions Survey with 38,637 businesses responding, so it is not a small poll. By size: 28% of businesses with nine staff or fewer, 49% of those with 250 or more.

DSIT AI Adoption Research, carried out by IFF Research and Technopolis Group across 3,500 telephone interviews with private-sector businesses of five or more staff, puts it at 16%, with a further 5% planning to adopt and 80% doing neither.

A British Chambers of Commerce survey widely reported during 2026 puts it above half.

None of them is wrong. They measure three different things: whether anybody in the building has used an AI tool, whether the business has deliberately deployed one, and whether a self-selecting group of Chamber members describe themselves as actively adopting. The spread between 16% and 54% is the distance between an employee with a browser tab open and an organisation that has changed how it works.

The surveys disagree by a factor of three because “using AI” is not one thing. Working out which of the three you are is most of a readiness assessment.

One ONS figure settles it more honestly than any headline rate. The average number of distinct AI technologies used by an adopting business rose from about 1.4 to about 1.6 in nearly three years. The ONS calls this adoption “relatively shallow”, which is restrained. Most adopters have added one thing, and the breakdown says which: large language models 18%, visual content creation 16%, machine learning on data 12%, image processing 6%, robotics 2%.

So the typical AI-adopting UK business is a business where somebody writes with a chatbot. That is a real and useful change. It is also not a system, it does not survive that person leaving, and it will not show up in the accounts.

Same year, same country, three questions "Has anyone here used an AI tool?" BCC members, 2026 ............... 54% "Does this business use at least one AI technology?" ONS, 10+ staff, June 2026 ....... 35% ONS, 0-9 staff .................. 28% ONS, 250+ staff ................. 49% "Has this business deliberately deployed one?" DSIT, 5+ staff, fieldwork 2025 .. 16% ...planning to ................ 5% ...neither .................... 80% Depth, not breadth, is the story. Average AI technologies per adopter late 2023 ~1.4 -> June 2026 ~1.6

The number that should decide your next move.

Inside the DSIT data sits the most useful pair of figures a small business owner could read before committing budget.

Among businesses that have adopted AI, 75% reported improved workforce productivity. Asked about employees’ overall productivity specifically, 56% reported an increase, 35% no change and 1% a decrease.

And 77% reported no change in revenue. 12% reported an increase, 1% a decrease, and 11% did not know.

Three quarters of adopters say their people got more productive. More than three quarters say it made no difference to revenue. Both are true, and the gap between them is where AI budgets go to die.

That is not a contradiction, and it is not evidence that AI does not work. It is the oldest result in operations: saved time is not saved money until the freed capacity is pointed at something. Twenty minutes a day given back to nine people is three hours a day that will quietly refill with other work unless somebody decides what it is for.

These are self-reported figures and DSIT says to treat them as estimates, so take them as direction rather than precision. The direction is not ambiguous.

The practical consequence is simple. An AI project that cannot name the thing that changes on the other side — a job you stop doing, a hire you do not make, a quote you turn round in a day instead of a week, a service you can now sell — is a project heading for the 77%. Naming it costs nothing. Not naming it is the expensive option.

The biggest barrier is not cost, and not skills.

Asked why they had not adopted AI, businesses in the DSIT survey gave two answers far more often than any other: no identified need for AI (71%) and limited AI skills or expertise (60%).

Cost is raised less often than either, though among businesses that do raise a barrier the ones rated most significant are ethical concerns (80% of those citing them), high costs (76%) and unclear or uncertain regulation (72%). So cost is not what stops most people starting. It is what worries the people who have already started thinking.

“No identified need” is the finding worth sitting with, because it is rarely literally true. A business confident that AI has nothing to offer it is usually a business where nobody has spent half a day looking. The need is not absent. The search is.

Which is what a readiness assessment is actually for — and why a score on its own is the wrong output. A number tells you where you sit. It does not tell you what to do on Monday.

The six areas an honest assessment scores.

We built a free AI readiness assessment because the lead-capture version of this — eight questions, an email gate, a score, a sales call — is not useful to anyone. Ours is twenty questions, no sign-up, and it scores six areas. You can sign in to save a result if you want to. You do not have to.

Whether you use ours, work through our fuller AI readiness assessment and checklist for SMEs, or write your own on paper, these are the six areas, and the question each one is really asking.

Strategy and leadership

Not “do you have an AI strategy”. The real question is whether one named person can say which business outcome the project is meant to move, and has the authority to stop it when it does not. Projects without that person rarely fail loudly. They just never finish.

Data and systems

Is the information the tool would need digital, organised, and reachable by something other than a human opening a file? This is the area that most often turns a six-week project into a six-month one, and it is almost always discovered after the contract is signed rather than before it.

Tools and technology

Do your existing systems talk to each other, and can anything get data in and out of them without a person retyping it? An AI layer on top of systems that do not integrate mostly relocates the copy-and-paste.

Skills and people

DSIT found limited AI skills were the top hindrance to wider adoption even among businesses already using AI, cited by 54% of users. Readiness here is not whether anybody can write Python. It is whether the people who would use the thing are confident enough to tell you when it is wrong.

Processes and operations

Is the work you want to automate genuinely repeatable, and is it written down anywhere? You cannot automate a process that lives in one person’s head, because the first thing the project has to do is get it out of there — and that is a separate piece of work with its own cost and its own timeline.

Governance and risk

Who decides what may be put into a third-party AI tool, and what happens when an output is wrong in a way that reaches a customer? This is the area almost everybody fails, so it gets its own section.

The governance gap almost nobody is closing.

DSIT found that very few businesses had internal AI policies or guidelines, and that most of those without one had no plans to create one. Larger firms were more likely to have something, or to be building it.

At the same time, 84% of AI users apply at least some human checking to AI outputs, 67% apply significant checking, and 2% apply none.

Read those together and the picture is clear. Businesses are managing AI risk through individual caution rather than policy. That works until it does not. It does not survive growth, it does not survive the careful person leaving, and it gives you nothing to show a client, an insurer or a regulator who asks how you control it.

A written AI policy is not bureaucracy. It is the document that lets you say yes to things, because somebody has already decided where the line is.

It does not need to be long. The useful version answers four questions: which tools are approved, what categories of information may never be pasted into them, which decisions always need a human signature, and who to tell when something goes wrong. That is one page.

If it helps, the same free tools set includes a starter AI acceptable-use policy generator, and a red, amber and green check against UK GDPR for whether a particular piece of information can safely go into a public AI tool at all. Both are free and neither asks for an account.

What a low score actually means.

The failure mode of a readiness assessment is treating a low score as a verdict. It is not a verdict. It is a price signal.

A business scoring well on data, process and governance can sensibly commission a build, because the expensive unknowns are already known. A business scoring badly should not commission a build — not because it is not ready for AI, but because it is not ready to be quoted accurately, and a fixed quote against unknown data is a quote somebody is going to lose money on. Often both parties.

The right response to a low score is a smaller first step, and the steps are cheap next to the thing they de-risk.

Readiness score -> sensible first move Low across the board Pick ONE process. Write it down. Costs: a morning of someone's time. Need is clear, data is not Idea validation sprint, GBP 1,500 fixed. Buys: feasibility, a data check, and a real estimate. Data is fine, value unproven Prototype, from GBP 5,000. Buys: a working thing real users can be put in front of. Proven, and people keep using it Integration or MVP build. Buys: the system the pilot proved. Each row costs less than the row below. Taking them out of order is how the 77% happens.

Our published ladder starts with a fixed £1,500 idea validation sprint for exactly this reason: it is the cheapest way to turn “we think AI could help with X” into either a real estimate or an honest no. Four in five businesses in the DSIT survey were neither using AI nor planning to. Some of them are right to be. Finding out which costs very little, and the assessment itself costs nothing.

What this means for a buyer.

Start with the business decision, audience, and evidence the project must produce. Simam Digital can turn that into a focused discovery, prototype, MVP, or production roadmap across AI applications, SaaS platforms, digital twins, real-time 3D, XR, and interactive systems.

Sources and further reading