Yes, but only the businesses that decide what AI should change, and how they will measure it, before they buy it. Public money and business adoption are both rising quickly. Productivity evidence within individual firms is not keeping pace.
In 2026, 32% of UK businesses used at least one AI technology, up from 16% a year earlier, according to the government’s AI adoption survey. Yet Air IT Group’s research with 504 UK decision-makers found that 83% had spent on AI without a measurable return or were not measuring returns at all. Closing that distance is now a management problem more than a technology one.
Public Money Is Flowing Faster Than Business Returns
The Chancellor’s October 2026 AI package shows where public investment is heading. It commits £10 million a year to work on how AI will reshape jobs and industry, an extra £23 million over three years for the AI Economics Institute and £30 million to start local AI partnerships with Mayoral Strategic Authorities.
For a leadership team deciding whether to build in-house or brief an AI development company in the UK, this spending changes the context but not the task. Public funds build skills, research and infrastructure. They do not redesign your claims process, your scheduling or your month-end close.
The Chancellor of the Exchequer, John Healey MP, said that “every firm, community, hospital, school, and every aspect of our economy can benefit.” AI Minister Kanishka Narayan MP added that “Britain must have the capabilities to protect its people and shape its own future.” Both quotes describe national intent. Neither tells a board whether its own AI spend is paying back, and that is where the gap sits.
The returns question is sharper at the top. In the EY-Parthenon CEO Outlook Survey of UK chief executives, half ranked AI tools among their top three productivity drivers of the past year. Only 14% said they have comprehensive, real-time visibility of AI costs, and 32% reported significant gaps in assessing return on investment. CEOs believe AI is helping. Far fewer can show it in numbers a CFO would sign off on.
Why Spending Is Not Becoming Output
Air IT Group calls this the AI decision gap, and its data points to process failures rather than weak technology. Some 41% of respondents bought or introduced an AI tool before agreeing the problem it should solve. Among organisations with at least one active AI initiative, 74% said not every initiative had both an agreed business objective and a success measure.
Pilots are where the gap becomes visible. Of the organisations that formally track AI pilots, 62% abandoned at least one in the past 12 months, with integration difficulties (34%) and employee non-adoption (29%) among the reasons given. Neither is a model problem. Both are delivery problems.
Two further findings explain why even working deployments often fail to move productivity figures.
The Verification Tax
Almost half (47%) of respondents said AI had made a task or process take longer once checking, correction, setup and management were counted. Call it the verification tax. A tool that drafts a supplier contract in 2 minutes but requires 40 minutes of legal review has saved nothing if the old process took 35 minutes. Business cases that count only the drafting time will overstate the gain.
Unbanked Hours
Only 13% said their organisations consistently convert time saved by AI into measurable gains in output or productivity. An hour saved is not productivity until it is redeployed to work that produces revenue, throughput or quality. When free time is not reflected in targets, staffing plans or service levels, it drifts into email and meetings and never reaches the productivity figures.
What Separates Firms That See Returns
Research from The Productivity Institute at Alliance Manchester Business School, commissioned by Bytes Technology Group and published in October 2026, puts AI’s most plausible contribution to UK productivity growth at 0.5 to 0.7 percentage points a year. The condition is investment in managers, training and new ways of working, not tools alone. The report also found that UK firms rated as well-managed in 2020 were more likely to have adopted AI three years later, regardless of their sector, size or age.
The table below maps each failure point in the data to a change in how AI work is scoped and the measure that shows whether it worked.
| Failure Point | What to Change | What to Measure |
| Tool bought before the problem | Start from one P&L line or service metric | Baseline value recorded before the build |
| Verification tax | Build review into the workflow, with confidence thresholds that send only uncertain outputs to people | End-to-end cycle time, not task time |
| Integration failure | Treat data access and system integration as phase one | Share of transactions running in production |
| Unbanked hours | Reset targets, rotas or capacity plans when time is freed | Output per FTE or cost per case |
| Non-adoption | Train line managers first and redesign roles around the new process | Weekly active use within the target team |
A simple test for any AI proposal is whether the board pack would show the last column moving within two quarters.
Where an AI Development Company in the UK Fits In
Because integration and adoption sink more pilots than model quality does, choosing a build partner is mostly a question of delivery discipline. A capable AI development company in the UK markets should be able to show how it connects models to your existing ERP, CRM or case management systems, how it tests accuracy against your own data before launch and how it plans adoption for the team that will use the output.
Shortlist questions should be practical. Ask how the partner will set the baseline, what the evaluation dataset looks like, who owns monitoring after go-live and how cost per transaction will be reported. That last answer addresses the cost-visibility gap CEOs flagged in the EY survey. Providers such as Appinventiv, which offers AI development services for UK businesses, can be assessed against exactly these questions.
Data protection belongs in scope from day one. Under the UK GDPR, an AI system that processes personal data requires a lawful basis, and processing likely to pose a high risk to individuals requires a data protection impact assessment before processing begins. This is general information, not legal advice.
Start With One Workflow and One Number
The UK’s AI investment will build capacity across the economy, but it will not, by itself, lift productivity within a single business. The firms best placed to see returns pick one workflow, agree the number it should move and fund the process and people changes alongside the build.
The practical next step is a short scoping exercise with your build partner that maps the workflow, records the baseline and defines an MVP small enough to reach production within a quarter. That gives the board a cost estimate and a success measure before the larger budget is committed.
Frequently Asked Questions
What Did the Chancellor’s October 2026 AI Package Include?
The package covered AI security, homegrown capability and regional growth. Commitments include £165 million for a National AI Biosecurity Initiative, an additional £120 million for the AI Security Institute, an extra £60 million for the Sovereign AI Fund with a new physical and industrial AI focus area, and £18 million for a UK Advanced Chip Design Academy.
How Is the UK Public Sector Already Using AI?
The government cites two examples. In the NHS, AI helps clinicians analyse more than 210,000 chest X-rays a month, around a third of all NHS chest X-rays. HMRC’s AI-powered assistant, Ask HMRC, supported around 6.3 million customer interactions in 2025-26.
What Support Does the Package Offer Smaller Businesses?
The most direct route is the £30 million for local AI partnerships with Mayoral Strategic Authorities. These partnerships bring together councils, businesses, universities, colleges and SMEs, with a focus on AI adoption and building workers’ AI skills. Firms should check with their mayoral authority about how to participate.
How Should a Business Measure AI Productivity?
Measure the whole process, not the task. Record a baseline before deployment, then track end-to-end cycle time, output per employee or cost per case. Include the hours staff spend checking and correcting AI output, and agree with finance how freed time will show up in targets.





