Many established businesses are already using AI, but use is not the same as operating capability. Individuals can become productive with an assistant while the company remains unable to repeat the result, protect its information or explain where the value went.

The Office for National Statistics reported in July 2026 that self-reported AI use among UK businesses with ten or more employees had risen from around 12% in late 2023 to around 35%. Yet only 10% of adopting businesses described their use as extensive, and the average number of AI technologies used had moved only modestly. Adoption was growing; depth remained limited.

That is the maturity challenge. The aim is not to look advanced. It is to make a small number of valuable uses dependable enough to become part of how the business works.

Activity is not a maturity model

A company can buy an enterprise assistant and still have poor data, unclear ownership and no useful measures. It can run an impressive pilot that never survives contact with everyday workloads. It can issue a policy that reduces obvious risk while giving managers no practical route to approve good ideas.

Maturity should therefore be judged through evidence:

• leaders can explain where AI could strengthen or weaken the business;
• important information is reliable enough for the intended use;
• people have the skills and confidence to work differently;
• useful experiments become controlled, repeatable workflows;
• ownership, guardrails and monitoring match the consequence of the task.

The OECD’s 2025 paper on SME AI adoption identifies connectivity, data, algorithms and compute, skills and finance as important enablers. Its wider point is that SMEs start from different levels of digital maturity, so the route forward needs to fit the firm rather than a generic technology timetable.

Use START to find the constraint

START is a practical leadership lens: Sensemaking, Truth, Adoption, Repeatable workflows and Trust. It is not a certification or a reason to score the business to two decimal places. Its purpose is to expose the weakest condition holding useful adoption back.

Sensemaking

The leadership team needs a shared view of what AI changes for customers, competitors, roles and economics. Without it, each function follows its own interpretation and the portfolio becomes a list of disconnected tools.

Ask: which business outcomes matter, where might AI alter the basis of competition, and what must remain distinctly ours? The output should be a short set of choices, not a catalogue of possibilities.

Truth

AI cannot make inconsistent definitions dependable. If sales, finance and operations disagree about the customer, margin or order status, faster analysis may simply produce disagreement sooner.

Truth means knowing which information a use case needs, where it comes from, who owns its quality and what limitations users must understand. It does not require a perfect enterprise data platform before anything can begin. It requires data that is fit for the chosen decision or workflow.

Adoption

People need more than a demonstration. They need to understand how their role changes, which judgement remains theirs, when to challenge an output and how good practice will be reinforced.

The Department for Science, Innovation and Technology’s AI Adoption Research found that limited AI skills and a lack of identified need were common barriers. Among businesses already using AI, just over half said they felt ready to scale. Those findings suggest that access to a tool is only one part of readiness.

Repeatable workflows

Individual prompting can be useful for learning, but a business result depends on a process. The input, instruction, source information, human checks, output and hand-off all need to work together.

A repeatable workflow has a named owner, a defined starting point, an acceptable output and a fallback when the system is uncertain or unavailable. It can be measured against the previous way of working. It does not rely on one enthusiastic employee remembering an elaborate prompt.

Trust

Trust is earned through ownership, boundaries and evidence. Users need to know what the system may do, what it may not do, which information it can access and who is accountable when something goes wrong.

The UK Government’s AI Playbook is written for the public sector, but its practical principles travel well: start with the problem, be explicit about data and limitations, establish senior ownership of risk, integrate oversight and evaluate outputs. An SME can apply those principles proportionately without creating a bureaucracy.

Diagnose before adding more projects

Rate each START condition as absent, emerging, usable or repeatable. Require evidence for the rating. A policy document may support Trust, but it does not prove that exceptions are handled or incidents escalated. A training session may support Adoption, but it does not prove that a team changed its workflow.

Then look for the pattern.

If Sensemaking is weak, pause the shopping list and clarify business choices. If Truth is weak, narrow the use case to information the business can rely on. If Adoption is weak, involve the people doing the work in redesign and testing. If Repeatable workflows are weak, turn one successful experiment into an owned process before starting five more. If Trust is weak, set decision rights, guardrails and monitoring before increasing consequence or scale.

The lowest-scoring condition often determines the pace of the whole portfolio. Funding a stronger model will not compensate for a workflow nobody owns.

A focused 90-day maturity agenda

The first three months should build one complete line of sight from strategic intent to operating evidence.

Days 1–30: establish Sensemaking and Truth. Agree two or three outcomes that matter to the business. Inventory current AI use, including features inside existing software. Select one workflow where improvement would be commercially useful and the data is sufficiently dependable. Name the executive sponsor and operational owner. Record the baseline for time, quality, cost or customer effect.

Days 31–60: build Adoption and a Repeatable workflow. Redesign the whole task with the people who perform and receive the work. Define inputs, instructions, checks, hand-offs and fallback. Test normal cases as well as incomplete and awkward ones. Train the team in the new process, not simply in the tool. Capture where human judgement adds value.

Days 61–90: establish Trust and decide. Set proportionate access, review and escalation rules. Monitor the agreed measures and collect user feedback. Decide whether to scale, change, stop or defer the workflow. Record what the business learned, including evidence that challenges the original case.

At the end of 90 days, the most valuable output may not be an automated process. It may be a confident decision to stop. Maturity includes rejecting work that cannot yet meet the business standard.

Avoid the three common shortcuts

First, do not treat tool deployment as transformation. A licence gives access, not ownership, adoption or value. Second, do not wait for perfect data everywhere. Improve the information needed for a bounded use case and learn from operating it. Third, do not scale activity because a demonstration was fluent. Scale only when the workflow produces an acceptable outcome under real conditions.

This keeps the agenda proportionate. Established SMEs do not need to imitate a global technology company. They need enough direction, discipline and evidence to make better decisions than they made last quarter.

A practical next step

Run a 60-minute START review with the leadership team. Give each condition one evidence-based rating, identify the weakest one and choose a single workflow through which to improve it over the next 90 days. Write down what would cause you to scale, change or stop.

If you need an independent view of the gaps and a sequenced executive agenda, explore The AI Centre’s AI maturity support. The aim is a practical route to stronger decisions, not another assessment that sits on a shelf.

Sources

• Artificial intelligence in UK businesses: 2023 to 2026, Office for National Statistics, 20 July 2026.
• AI adoption by small and medium-sized enterprises, OECD, 9 December 2025.
• AI Adoption Research, Department for Science, Innovation and Technology, updated 13 February 2026.
• Artificial Intelligence Playbook for the UK Government, Government Digital Service, 10 February 2025.