AI adoption is accelerating across UK businesses.
Around 35% of businesses with 10 or more employees reported using at least one AI technology in June 2026, according to the Office for National Statistics. That is almost three times the proportion recorded in September 2023.
But beneath those headline figures, the picture is more complicated. Only 10% of businesses already using AI said they were using it extensively, while just 15% said more than half of their employees used AI in their daily work.
That points to a growing gap between having access to AI and successfully embedding it across an organisation.
For many businesses, experimentation is no longer the difficult part. The challenge is turning that activity into something useful, measurable and repeatable.
THE AI ADOPTION GAP
Different stages of AI adoption are often grouped together.
An employee experimenting with an AI assistant is not the same as a team incorporating AI into an established process. A tool being deployed does not mean employees are using it consistently, and widespread usage does not automatically mean an organisation is generating measurable value.
It is more useful to think about adoption as a progression:
Experimentation → Deployment → Adoption → Value → Scale
Recent research illustrates the gap. McKinsey’s 2025 global survey found widespread AI use, yet around 62% of AI-using respondents still described their organisations as experimenting or piloting rather than scaling or operating at full scale.
So where does progress start to break down?
WHERE DO ORGANSITIONS GET STUCK WITH AI?
1. They struggle to identify where AI will create value
For organisations at the beginning of their AI journey, one of the biggest obstacles is deciding what to use it for.
Research commissioned by the Department for Science, Innovation and Technology found that 71% of UK businesses identified a lack of a clear use for AI as a barrier to adoption.
Starting with the technology can contribute to this problem. Giving employees access to an AI platform and asking them to find ways of using it may generate plenty of ideas without establishing whether they solve meaningful business problems.
A better starting point is the work itself.
Where are processes slow? Where is repetitive administration consuming significant time? Where are errors, delays or customer frustrations occurring?
Once the problem is clear, organisations can assess whether AI is genuinely an appropriate solution.
2. Promising experiments struggle to become part of everyday work
A successful demonstration can make AI implementation look deceptively simple.
Moving that same idea into an operational environment exposes a much wider set of considerations. Organisations may need to resolve data access, security, privacy, integration, human oversight and process ownership before an experiment can become part of normal work.
The workflow itself may also need to change.
McKinsey’s 2025 research found that 55% of its small group of self-identified AI high performers reported fundamentally redesigning workflows, compared with 20% of other AI-using respondents. While that does not prove workflow redesign caused better performance, it highlights an important difference between simply providing AI tools and changing how work gets done.
Organisations making progress are looking beyond individual tasks and considering where AI fits within the wider process, where people remain responsible and how work moves between the two.
3. Skills and leadership capability don’t keep pace
Among businesses already adopting AI, 54% identified limited AI skills, expertise or knowledge as an obstacle to wider adoption in DSIT research.
This can become more apparent as organisations move beyond initial experimentation. Trying an AI tool may require relatively little specialist knowledge, but embedding it into everyday work introduces more complex decisions about where it should be used, how processes need to change, how outputs should be evaluated and what risks need to be managed.
Those responsibilities also sit across different parts of an organisation.
Employees need to understand AI’s strengths and limitations, handle information appropriately and evaluate outputs. Process owners need to identify opportunities and redesign workflows. Technical teams may need deeper expertise around integration, data and security.
Leaders also need enough understanding to identify opportunities, prioritise investment and decide what should progress. Separate UK employer research found that only 34% rated their senior leaders as good at identifying new AI opportunities.
As adoption develops, the skills organisations need develop with it. Building capability therefore means understanding what different roles need to know and do, rather than treating AI proficiency as a single skill that can be developed uniformly across the workforce.
4. They struggle to prove whether AI is actually working
AI can feel useful without necessarily creating measurable business value.
A 2025 study by METR provides an interesting example. Experienced open-source developers believed AI had made them faster, yet measured completion times in that particular experiment were actually 19% slower when AI tools were available.
That does not mean AI makes developers less productive. Instead, it demonstrates why perceived productivity and measured productivity need to be separated.
Other research has found clear gains. A study of 5,172 customer-support agents found that introducing a generative AI assistant increased issues resolved per hour by an average of 15%, although the impact varied according to employee experience.
AI’s impact depends on the task, employee, tool and surrounding workflow. Organisations therefore need to decide what success looks like before a pilot begins and measure outcomes such as time, quality, errors, customer experience or cost.
5. What works in one team doesn’t automatically scale
Even a successful use case can encounter another barrier when an organisation tries to reproduce it elsewhere.
Scaling requires clear ownership, repeatable processes, appropriate governance, training and mechanisms for sharing what has been learned.
Without them, organisations can accumulate isolated pockets of successful AI activity without developing wider organisational capability.
Scaling also does not mean making every experiment permanent. Knowing when to stop an unsuccessful pilot, adapt a promising one or invest further in something that works is an important part of AI maturity.
WHAT ARE ORGANISATIONS MAKING PROGRESS DOING DIFFERENTLY?
There is no universal AI adoption playbook, but several behaviours appear consistently across the research.
Organisations making greater progress tend to start with business problems rather than AI tools, establish clear ownership and consider how workflows need to change around the technology.
They recognise that capability needs to exist at different levels, from confident users and process owners through to technical specialists and informed leaders.
They also measure outcomes rather than simply monitoring usage. The objective is not to maximise the number of people using AI. It is to identify where AI improves something that matters and create a reliable way of reproducing that improvement.
WHERE SHOULD LEADERS START?
Leadership teams do not necessarily need to begin with another technology decision.
Start by understanding what is already happening. Employees may already be experimenting with AI, while teams may have very different levels of confidence and experience.
Then identify a business problem worth solving and define what better performance would look like. Consider what needs to change around the technology, including workflow, skills, data, governance and ownership.
Finally, test against a clear baseline and decide whether to stop, adapt or scale.
Not every process needs AI. Deciding that a particular application does not create enough value can be just as useful as identifying one worth expanding.
MOVING FROM AI USE TO AI CAPABILITY
The challenge facing organisations is increasingly less about getting people to try AI.
The harder task is translating experimentation into high-value use cases, better ways of working, appropriate skills, measurable outcomes and repeatable practices.
That is where businesses can begin to turn experimentation into meaningful value.
Not sure where your organisation is getting stuck with AI?
If your organisation is already exploring AI but you’re unsure how to take the next step, our AI Skills Consultations can help you assess your current position, identify opportunities and understand where skills, leadership or organisational capability may need to develop.