What did the Bank of England find about AI and productivity?
The Bank of England published its November 2026 Monetary Policy Report on 5 November. Box D of the report examines UK productivity and asks how much of the recent improvement comes from artificial intelligence. The answer is smaller than the public debate suggests.
Firms in the Bank's Decision Maker Panel survey report a modest productivity gain. They estimate that AI has raised their productivity by around 0.5% over the past three years. The Bank says that points to AI adding about 0.1 to 0.2 percentage points to annual productivity growth since early 2024.
The Bank says the result is in line with similar impacts reported in other countries. That makes the UK estimate less likely to be a local anomaly.
The same firms expect a much larger effect ahead. They forecast that AI will lift productivity growth by about 0.9 percentage points a year over the next three years. The Bank says those gains would be split roughly equally between faster output and slower hiring.
The report's language is careful. It says there is only evidence of "a modest AI-driven rise in UK productivity growth" so far. It also says AI investment is supporting demand more clearly than it is supporting supply.
The finding arrived in a significant week for monetary policy. The Monetary Policy Committee voted 8-1 on 4 November to raise Bank Rate by a quarter point to 4%. The BBC reported that this was the first increase in more than three years.
Inflation explains the move. Consumer prices rose 3.5% in the year to September. The Bank expects inflation to reach around 5% by the middle of 2027, driven mainly by energy costs.
Productivity matters in that setting because it determines how fast the economy can grow without adding to inflation. A weak AI effect would make the Bank's job harder.
The report does not treat AI as a failure. It treats it as an early-stage technology whose macroeconomic effects are still small and uneven. That distinction runs through the rest of the analysis.
Why is AI adding only 0.1 to 0.2 percentage points a year?
The Bank does not claim that AI is doing nothing. It argues that other forces explain most of the productivity pickup since early 2024.
The first is an unwind of labour hoarding. Many firms held extra staff in 2022 and 2023 because hiring was difficult. As demand recovered and hiring slowed, output per worker rose.
The second is a shift in the job mix. Higher minimum wages and employer National Insurance contributions have reduced lower-paid employment. Those jobs tend to have low measured productivity, so their loss lifts the average.
The Bank says these effects on productivity growth are likely to be limited. That is unless minimum wages or employer NICs rise sharply again.
The Bank also examined sectors most exposed to AI. Productivity growth in information and communication has improved since early 2024. It has not risen above pre-pandemic averages.
Chart D.3 in the report makes a similar point. Productivity growth picked up in less AI-exposed sectors, including wholesale, retail, accommodation and food services. It also rose in information and communication. A broad cyclical pattern is doing much of the work.
That pattern matters because cyclical gains tend to fade. The Bank's central projection has productivity growth falling back towards its assumed trend rate of 1%.
The Bank's Agents, its regional business contacts, report that firms using AI well can automate routine administration and speed up knowledge work. Those reports are qualitative rather than representative.
The Agents also say some contacts have reached their lower bound for headcount.
Spare capacity within firms now reflects slack in physical capacity more than slack in labour. That limits how much more productivity can come from thinner staffing.
The result is a productivity story with two layers. The visible improvement is real but mostly cyclical. The AI layer is positive, small and uncertain.
The gap between the realised 0.1 to 0.2 points and the expected 0.9 points is the central puzzle. It is also the number that matters for inflation and interest rates.
If the cyclical sources fade before the AI gains arrive, productivity growth could slow again. The Bank says the hoarding unwind cannot continue indefinitely.
How can AI drive a fifth of GDP growth but so little productivity?
The report contains a second finding that looks inconsistent with the first. AI-related activity accounts for just under a fifth of GDP growth since the start of 2026.
The Bank built a proxy for AI activity from three industry groups. They are computer programming and consultancy, information services, and publishing, which includes firms developing AI agents and applications.
The proxy is imperfect. Not all output in those industries is AI-related, and some growth reflects a longer expansion in technology services. The Bank adjusts for that trend and still finds a large contribution.
The important detail is where the activity sits. The Bank says the contribution comes from upstream development and implementation of AI. It does not come from broad productivity gains among firms that adopt the technology.
Business investment grew by nearly 5% in the first half of 2026. Investment in information and communication has been strong, and the Bank's Agents link part of it to AI spending.
This is demand, not supply. Building data centres, buying chips and writing software adds to GDP now. It does not automatically make existing workers more productive.
The report says the rollout appears to primarily reflect a boost to demand within the economy. If AI investment starts to feed through to productivity, it would support potential supply.
There is an inflation consequence. Demand for AI products such as memory chips has pushed up export prices in China and Asia. That raises UK import costs.
So AI can raise costs before it raises output. That helps explain why the Bank is cautious about declaring a productivity boom.
The two findings can coexist. An economy can sell a lot of AI equipment and still be waiting for the productivity payoff.
The distinction also matters for investors. Revenue from AI construction and hardware is visible in the national accounts. The return on that spending depends on gains that have not yet appeared in the productivity data.
How much are UK firms actually using AI?
Adoption is broad. A survey of nearly 9,000 executives across seven countries finds that 88% of UK firms use at least one AI technology. The findings appear in NBER Working Paper 34836.
That figure depends on weighting. The same UK responses give 88% when weighted by employment, 86% unweighted and 83% when weighted by the number of businesses.
The spread comes from firm size. Large firms are more likely to use AI, and surveys that over-sample them report higher adoption.
Across the seven countries, 78% of firms use AI. The United States is highest at 95%, followed by the UK at 88%. Adoption is highest in UK finance and insurance and in information and communication.
Spending is rising but remains uneven. UK firms spent about $422 per employee on AI in 2025. They expect to spend about $725 in 2026, against roughly $2,000 per employee in the United States.
The share of UK firms making no AI spending is expected to fall from 33% in 2025 to 17% in 2026. The share spending more than $1,000 per employee is expected to rise from 8% to 13%.
The spending is highly concentrated. Just 1.3% of UK firms account for half of all AI spending. The top 10% account for 69%, and the UK spending Gini coefficient is 0.92.
Use is also shallow for many firms. About 82% of UK respondents report no productivity impact from AI over the past three years.
Where gains appear, they cluster. Information and communication firms report a 1.3% productivity gain, and professional and scientific firms report 1%.
Executive use is rising. About 80% of executives say they use AI regularly, for an average of 2.1 hours a week. That is up from 1.5 hours at the start of 2026.
Broad adoption, concentrated spending and mostly flat reported productivity can all be true at once. That combination is what the Bank's modest estimate describes.
What does this mean for jobs and interest rates?
The employment picture is quieter than the public debate suggests. UK firms report an average employment effect of about -0.1% over the past three years.
About 88% of UK firms report no employment impact at all. The Bank says the aggregate effect is close to zero.
The channel is hiring, not firing. Firms adopting AI are less likely to hire or replace leavers, according to the Bank's Agents. They are not materially more likely to cut staff.
Expectations are more negative. UK firms expect AI to reduce employment by about 1.3% over the next three years. That is the largest projected fall alongside the United States.
Even that is modest in aggregate. The survey authors caution that it misses new firms, general equilibrium effects and job creation elsewhere.
For interest rates, productivity is the key variable. Faster productivity growth lets the economy expand without stoking inflation. That gives the Bank more room to keep rates lower.
That is why the Bank watches productivity so closely when setting rates. A durable gain would ease the trade-off between growth and inflation.
Slower productivity does the opposite. It makes the current inflation problem harder to solve. It also limits how quickly the Bank can cut rates once the energy shock passes.
The Bank's central case assumes 1% trend productivity growth. If AI delivers the 0.9 point gain firms expect, that assumption would look too low.
If it delivers only the realised 0.1 to 0.2 points, the recent pickup will fade without a lasting supply boost.
The MPC has said it will review its judgments for trend productivity and trend labour supply. Those reviews will shape the path of rates beyond the energy shock.
For workers, the near-term effect is a slower rate of hiring rather than a wave of redundancies. For borrowers, the gap between the realised and expected gain is the main uncertainty for the path of rates.
Why do estimates of AI's productivity effect vary so much?
The 0.5% figure is based on what firms say, not on a controlled experiment. Executives choose from broad categories such as small or large positive effects.
Most firms report no effect, which makes the average sensitive to a minority. About 82% of UK firms report no productivity impact.
Survey design matters. The Decision Maker Panel covers businesses with ten or more employees, so the smallest firms are underrepresented.
Questions also differ. Some surveys ask whether firms use AI at all, while others ask about regular use or spending. That produces a wide range of adoption estimates.
The seniority of the person answering also matters. The NBER paper finds that CEO and CFO respondents report 89% adoption, against 84% for other roles. It says three design factors each shift measured adoption by five to twelve percentage points.
The NBER paper illustrates the problem. UK adoption estimates range from about 9% in an ONS survey to 25% in an LSE-CBI survey. An Institute of Directors survey put the figure at 49%, while the NBER team's own estimate is 88%.
Productivity measurement is being revised at the same time. The Office for National Statistics has adopted a new component method. It will publish quarterly estimates up to the third quarter of 2026 on 17 November.
The Bank says the revisions suggest productivity growth since the financial crisis was stronger than previously thought. That changes the baseline against which AI is judged.
Timing is another caveat. General-purpose technologies often take years to appear in aggregate data. The Bank says uncertainty about the size and timing of AI's effect remains large.
Firms' forecasts are not evidence. The 0.9 point expectation is a useful signal of intent, but it may reflect optimism as much as measurable capability.
Other evidence is also mixed. A Gartner survey reported by The Register found about three-quarters of board members expected AI productivity and cost savings. Meanwhile, 38% of companies were over budget on AI.
The Bank's own conclusion is conditional. It sees material upside risk if AI adoption accelerates, not a guaranteed gain.
What should you watch next?
The first test arrives on 17 November. The ONS will publish quarterly productivity data under its new method, giving a cleaner read on whether the recent pickup is durable.
The next Monetary Policy Report, due in February 2027, will update the Bank's view. The MPC has said it will review its assumptions for trend productivity and labour supply.
The Bank will also take account of industry-level data under the new ONS method in February. That should show which sectors are converting AI spending into output.
Watch business investment and AI spending. If firms keep buying but productivity stays flat, the gap between promises and results will widen.
Watch the labour market. A continued hiring slowdown could lift measured productivity without any AI-driven output gain.
Watch prices. AI demand is already pushing up memory chip and import costs. That complicates the inflation outlook.
The Bank for International Settlements has compared the AI spending surge to the dotcom boom, according to The Register. The concern is that infrastructure spending runs ahead of returns.
There is upside too. The Bank says AI adoption could accelerate and lift potential productivity growth, and it flags that risk explicitly.
The practical conclusion is not that AI is useless. It is that the UK has so far banked a small gain and a large expectation.
The next year of data will show whether the expectation becomes investment returns or another productivity disappointment.