More people are using AI at work. That much is clear.
Recent ONS data shows that reported use among UK businesses with 10 or more employees has risen sharply since late 2023. But rising business adoption only tells us that more organisations report using the technology. It does not tell us whether the work itself is getting better.
More businesses have access to new tools. More employees are experimenting with them. More tasks are being completed with some form of assistance. That is useful to know.
But it is still a long way from understanding whether the change is actually worthwhile. And I think that is where the conversation needs to move next.
Knowing that people use something tells you very little
We have become quite good at measuring adoption. Businesses can count licences, look at active users, ask employees what they are using, track training completion and measure usage frequency. All of that can be useful, but none of it answers the question I would want to ask if I were running the business:
What has actually changed in the work?
Something can be used every day without making the underlying process any better. A person might save ten minutes drafting something and spend another fifteen checking it. A team might produce more material but create a new review bottleneck for somebody else. One employee might find a genuinely better way of doing something while five others experiment with different approaches to the same problem.
Activity can increase while quality, consistency, effort and capacity remain largely unchanged. In some cases, they may get worse. That is the difference between adoption and improvement.
There is probably more happening than management can see
One of the more interesting patterns in current workplace data is the gap between what employees say they are doing and what organisations say they are doing. The measures are not directly comparable: the ONS comparison counts self-reported use for work or education among employed and self-employed individuals in Great Britain, while its business measure counts use of at least one AI technology among UK firms with 10 or more employees. But the wider point still matters.
Change can happen from the bottom up long before it becomes part of the operating model. Someone starts using a tool to prepare for a customer meeting. Another person uses it to analyse a spreadsheet. Someone else drafts an internal document with it. A manager finds a quicker way to summarise weekly updates.
None of those things necessarily require a formal transformation programme. They are often just people trying to get their work done more effectively. I do not think that kind of experimentation is a problem. In many cases, it is exactly how useful practices will be discovered.
The problem comes when experimentation becomes the permanent way the organisation operates. At that point, meaningful change can be happening across the business without much shared understanding of what changed, whether it helped, whether it created problems elsewhere, or whether anyone else should be working in the same way.
This is as much a people problem as a technology problem
There is also a risk in talking about adoption as though every employee experiences this change in the same way. They do not.
For one person, a new tool might remove an hour of repetitive administration every week. For somebody else, it might change a task they have spent years becoming good at. A manager might suddenly be responsible for reviewing far more work because their team can produce it faster.
A junior employee might gain access to knowledge or support they did not previously have. At the same time, they may get less exposure to some of the repetition through which people have traditionally learned how work is done.
Some people will experiment confidently. Others will worry about getting something wrong, breaching a policy they do not fully understand, or looking less capable because they need help. Some will inevitably wonder what all of this means for their role.
That matters because changing how work gets done changes more than task execution. It changes expectations, confidence, judgement, management behaviour and, over time, culture. If those things are ignored, it becomes difficult to claim that adoption is really understood.
The workflow matters more than the individual shortcut
I am also increasingly sceptical of measuring productivity one task at a time. Imagine somebody reduces a 60-minute task to 20 minutes. On the face of it, that sounds positive. But it is not enough information.
What happens next?
Does somebody now spend longer reviewing the output? Does the change create more corrections downstream? Can the person complete more useful work as a result? Does the organisation actually need more of that output? Has the faster method become a repeatable way of working, or does it still depend on one person knowing the right sequence of steps?
And what happens to the 40 minutes? That last question is often skipped completely.
Time saved creates potential capacity. It does not automatically create business value. Management still has to decide what that capacity is for. It might mean serving more customers, reducing a backlog, improving quality, giving managers more time to manage, avoiding future recruitment or simply making an overloaded job more sustainable.
All of those can be valuable outcomes. But the value comes from what happens next, not from the stopwatch.
Better evidence does not mean more surveillance
The obvious response is that organisations need more visibility. I agree, but that can quickly head in the wrong direction.
I do not think the answer is to monitor every prompt employees write or turn usage into another productivity score. That can tell you a great deal about activity and very little about whether work is improving.
The evidence I would want is much more practical. Where are people repeatedly finding better ways to do something? Where are outputs creating more checking or rework? Which changes are improving quality as well as speed? Where are several teams solving the same problem in different ways? Which practices are becoming repeatable? Where is genuine capacity being released? Where is human judgement still doing most of the important work?
And just as importantly, where has an experiment simply not worked? That is the kind of evidence management can actually use. The objective should not be to prove that every experiment was successful. It should be to learn enough to make a better decision about what happens next.
Adoption should eventually become an operating decision
There will always be some experimentation, and there should be. The technology is moving too quickly for businesses to design the perfect operating model in advance. People need room to try things, learn and adjust. But organisations cannot stay in experimentation mode indefinitely.
At some point, useful individual practices need to become organisational knowledge. That means understanding what worked, what did not, what should become repeatable and where further evidence is needed.
A practical progression might look like this:
Explore → Learn → Standardise → Measure → Improve
People explore where new tools might help. The organisation learns from that experience. Useful approaches become more repeatable where that makes sense. Evidence shows whether the change is actually improving the work. Then the process improves again.
That is very different from simply trying to increase adoption.
The next question is not whether people are using AI
Rising adoption is interesting, but it is increasingly the starting point rather than the conclusion. The more important questions are:
What changed for the people doing the work? What changed in the process around them? What evidence tells us whether the change was worthwhile?
Those are People, Process and Data questions. For smaller businesses in particular, answering them does not require a huge transformation office, endless dashboards or another layer of bureaucracy. It requires enough visibility to understand what is happening, enough structure to turn useful experiments into repeatable practice, and enough evidence to make sensible decisions about what comes next.
The businesses that get this right will not necessarily be the ones using the most AI. They will be the ones that get better at recognising when new technology genuinely improves work, understanding what that change means for the people and processes around it, and then doing something useful with what they learn.