I am going to say something that might annoy you. Most of what passes for AI adoption in UK businesses is theatre. Not malicious. Not dishonest. Just performative. People are buying tools, running pilots, ticking the “we do AI” box, and then wondering why the numbers do not move.
The data backs this up. And I think it is time we talked about it honestly.
The Numbers Do Not Match the Narrative
The Office for National Statistics has been tracking AI adoption in UK businesses since late 2023. Their most recent data, covering June 2026, shows that 35% of businesses with 10 or more employees report using at least one AI technology. That is up from 12% in late 2023. A near-tripling. Sounds impressive.
Here is the number nobody quotes. The average adopting business uses 1.6 AI technologies. Up from 1.4 three years ago. That is not transformation. That is a chatbot subscription and a meeting transcript tool.
The ONS data also shows that just 7% of UK organisations are pursuing an enterprise-wide AI strategy. The overwhelming majority are experimenting at the margins. One tool here, one there. Nothing connected. Nothing that changes how the business actually operates.
Studio Graphene commissioned a Censuswide survey of 500 UK senior decision-makers in early 2026. Seventy-eight percent of UK businesses said they were using AI in some capacity. Rising to 85% for mid-sized organisations. But only 31% said they had seen any positive ROI. Eighteen percent said their AI projects had never delivered the benefits they expected. And fewer than half, 41%, could even define what success would look like.
Let me put that another way. Nearly four in five UK businesses are using AI. Two-thirds of them cannot point to a return on that investment. And most of them did not bother to decide what a return would even look like before they started spending.
That is not a strategy. That is a vibe.
The Efficiency Trap
Reading Room surveyed 150 senior leaders at medium-to-large UK organisations and found something revealing. Ninety-seven percent said AI had delivered some level of measurable value. But 22% said that impact had fallen short of expectations. And while 71% reported efficiency gains, just 18% reported direct financial gains through new or additional revenue.
This is what I call the efficiency trap. You deploy AI inside a process that was designed before AI existed. The AI makes that process faster. You save time. Maybe you save headcount. But you have not changed the process. You have not created new value. You have just done the same old thing a bit quicker.
Amanda Falshaw, AI Enablement Lead at Reading Room, said it better than I can. Efficiency gains should be treated as just the starting point. Organisations should look at decision quality, customer experience, innovation capacity. Not just minutes saved.
She is right. But almost nobody is doing that. Because measuring minutes saved is easy. Measuring whether your customers are happier because your AI-powered support bot gave them a better answer is hard. And hard measurement does not look good in a board update.
Why This Happens
I have spent 27 years building software. I have watched this pattern repeat with every technology wave. Cloud. Microservices. Agile. Big data. Each one promised transformation. Each one delivered, mostly, incremental improvement. The gap between the promise and the delivery was always the same thing. People bought the tool and skipped the work.
The CBI published a report on 18 August 2026 called “The Adoption Decade: Closing the Execution Divide and Making AI Work for Britain,” produced with Oliver Wyman. It surveyed 415 CEOs and found that among firms leading on AI deployment, 49% report ROI meeting or beating expectations. Among laggards, the figure is 15%. A three-to-one gap.
The report identified what separates the leaders from the laggards. It was not budget. It was not access to better tools. It was not even talent, though that matters. The discriminating variable was workflow redesign. Deployment leaders are redesigning workflows at 49%, against 32% for firms stuck in pilots.
The CBI’s own words: “It is the single clearest finding on AI in this year’s survey: Deployment drives ROI.”
And then the report’s advice to businesses, which is stronger than anything it asked of government: “Treat AI as business transformation, not a technology roll-out.” And “prioritise a few of the highest-value use cases and move them from pilot to scale, with senior ownership, redesigned workflows, proportionate governance and security built in.”
There it is. The answer is not more tools. The answer is fewer tools, better chosen, deployed into processes that have been rebuilt to take advantage of them.
The Governance Surprise
Here is something that genuinely surprised me. AIBL Media surveyed 755 UK mid-market business leaders for their State of UK AI Adoption 2026 report. They found that half of the companies surveyed could show a measurable return on AI. But only 14% had AI running across three or more functions with a number behind it.
They expected the separating factor to be tooling, talent, or data. It was governance. Companies with no governance reported 22% measurable ROI. Companies with the most mature governance reported 85%. On much the same tools.
And here is the counterintuitive bit that should make every HR director wince. Companies with a written AI policy that nobody follows report 16% measurable ROI. Companies with no policy at all report 33%. A policy on paper that nobody follows is worse than no policy. It creates the illusion of control while removing the urgency to actually build one.
Government is not the answer to this. The CBI asked for four things: a delivery group, a voluntary code, a literacy standard, and a regulatory map. All reasonable. None of them redesign a single workflow. A literacy standard is cheap, popular, measurable, and almost certainly worth doing. But the execution divide is a 34-point gap. A literacy standard is worth maybe two to three points. These are not the same order of magnitude.
The Lloyds Business Barometer, published the same day as the CBI report and drawn from 1,200 UK firms, confirms this. Six in ten firms use AI. Above £10 million turnover that rises to 79%. Cost is the leading barrier at 18%. Data quality and access to skills are tied at 17% each. These are real constraints. But they are constraints on execution, not adoption. The tools are available. The willingness is there. What is missing is the discipline to use them properly.
What Actually Works
So what does work? Based on the data, based on what I have seen in 27 years of building software, and based on what the companies reporting real ROI are doing differently, here is the pattern.
Pick one thing. Not five. Not three. One. The single highest-value use case in your business. The one where AI could save the most time or generate the most value if it actually worked. Not the one that sounds most impressive in a press release.
Redesign the workflow before you deploy the tool. Do not drop an AI assistant into a process that was designed for humans doing manual work. Rethink the process from scratch. What does the AI do? What does the human do? Where do they hand off? What happens when the AI gets it wrong? If you cannot answer those questions, you are not ready to deploy.
Own it at the top. The CBI data is clear. Leaders combining senior ownership with workflow redesign are the ones reporting returns. Delegated pilots are the profile associated with the laggard figure. If the CEO is not involved, it will not work. Not because the CEO needs to pick the tool. Because the CEO needs to remove the political obstacles to changing how work gets done.
Measure something real. Not time saved. Not tasks automated. Measure something that matters to the business. Revenue. Customer retention. Error rates. Decision quality. If you cannot define what success looks like before you start, you are doing theatre.
Govern it properly. Not a policy document that sits in a shared drive. Actual governance. Who is accountable? What are the guardrails? What is the review cycle? The AIBL data is unambiguous. Governance is the single biggest lever between 22% ROI and 85% ROI. On the same tools.
The Uncomfortable Truth
Here is the contrarian bit. The problem is not that AI is overhyped. AI is genuinely useful. The problem is that most businesses are using it to do the wrong things faster.
The ONS data shows that around 12% of AI-using businesses report increased income from AI. Three-quarters report productivity gains. That gap between efficiency and revenue is the whole story. We have collectively built a landscape where AI makes existing things cheaper but very rarely makes new things possible.
That is not an AI problem. That is a strategy problem. And no amount of tool adoption will fix it.
The businesses that will win the next decade are not the ones with the most AI tools. They are the ones who had the discipline to pick one thing, rebuild the process around it, measure whether it worked, and then do it again. Slowly. Deliberately. Boringly.
That is not exciting. It is not a headline. It will not get you on a conference stage. But it is the difference between 22% ROI and 85% ROI, and the data is right there in the surveys if you bother to read past the executive summary.
Stop buying tools. Start changing how work gets done. That is the whole post.