Invent a partner at an accountancy in Leeds, a composite who exists only to make the numbers below easier to picture. A senior trainee has just handed in her notice. A year ago the partner would have advertised the post the same afternoon. This time she opens a spreadsheet and counts how many audit files the firm’s document-review tools now clear before a human looks at them. Then she writes “hold” next to the vacancy and goes back to her email, where the monthly panel questionnaire from the Bank of England is waiting. One question asks what AI has done to her firm’s productivity. She types a small number.

Thousands of small answers of that kind now sit behind a sentence in the Bank of England’s Monetary Policy Report, published on 6 November. Firms in the Bank’s Decision Maker Panel estimate that AI has raised their productivity by around 0.5% over the past three years. The Bank translates that into a contribution of 0.1 to 0.2 percentage points to annual productivity growth since early 2024. It is the first time the central bank has put a figure on the effect, and the figure is small.

Where AI shows up in the national accounts

The report finds AI in the growth figures, but in the wrong place for a story about adoption. Bank staff built a proxy for AI activity from three industries: computer programming and consulting, information services, and publishing, which includes the building of AI agents and application software. After adjusting for the longer-run expansion of technology work, the Bank says novel AI-related activity accounts for just under a fifth of UK GDP growth since the start of 2026. The same passage says what that growth is made of. It reflects “the upstream development and implementation of AI technologies rather than broader gains from their adoption elsewhere in the economy”. The proxy is the Bank’s own and only partly captures AI; the report notes that not all the output of those sectors is AI.

So the economy is paying people to build and install the technology, and the benefit to the firms that bought it is, so far, modest. The report adds that productivity growth in information and communication, a sector heavily exposed to AI, has picked up but has not risen above pre-Covid averages.

What firms say about jobs

On employment the panel firms report “close to zero” impact so far. The report says that matches several empirical studies, among them the Stanford Digital Economy Lab paper by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen. That paper, revised on 12 August, uses payroll data from ADP covering millions of American workers through June 2026. It finds no evidence of widespread, economy-wide job displacement. It also finds that employment of workers aged 22 to 25 in the most AI-exposed occupations is 19% below where it would have been had it kept pace with less exposed peers, and that the gap works through reduced hiring rather than extra separations.

The Bank’s Agents, who talk to businesses around the country, describe the same shape. In the Monetary Policy Report they say AI adoption is reducing the likelihood of firms hiring or replacing workers, “but not materially increasing the likelihood of firing workers so far”. The October summary of their visits is more specific. Labour-market effects, it says, are seen “primarily through lower replacement hiring, slower recruitment and reduced trainee intakes rather than widespread job losses”. Entry-level and routine professional roles are the most exposed, while contacts report pay premia for AI-literate and judgement-intensive roles. Governor Andrew Bailey saw an early sign of this in February, when he told a conference in AlUla that new online vacancies in the most AI-exposed roles had fallen by more than twice as much as in the least exposed group over three years.

Nothing in the Bank’s account contradicts the Leeds partner’s spreadsheet. A firm can employ as many people as before in total while the door at the bottom of the profession narrows, and an aggregate survey answer of “close to zero” will not register it.

The case for caution, and for the opposite

The Bank is careful about what the productivity figures prove. Productivity growth has also risen in sectors with little exposure to AI, such as wholesale and retail trade and accommodation and food services, where employment growth has fallen. The report links that to higher labour costs from the national living wage and employer National Insurance contributions, which may have thinned out low-paid, low-productivity jobs and so lifted the average. It says most of the recent strength in productivity growth appears to be cyclical, and in its central projection productivity growth falls back towards a trend of 1%.

A second caution comes from outside the Bank. On 24 October Torsten Slok, chief economist at Apollo, wrote that the technology “has delivered zero financial payoff to corporate America outside of tech”, citing flat revenue per worker and margins across the S&P 493 and a utilisation-adjusted productivity series from the San Francisco Fed that shows no acceleration. His evidence is American market data and his phrasing is far sharper than the Bank’s, but the two point the same way: the spending is real and the gain elsewhere is not yet visible.

The report also leaves room for the opposite outcome. It says there are “increasing upside risks” to productivity growth from AI, and that its Agents’ contacts increasingly report that firms using AI proficiently and at sufficient scale can automate routine administrative work and accelerate knowledge tasks. Panel firms expect AI to lift productivity growth by about 0.9 percentage points a year on average over the next three years, split roughly equally between slower employment growth and faster output growth. That is several times the 0.1 to 0.2 points they have seen. The Agents’ October summary says costs are changing shape too: labour savings come with higher spending on software, cloud services, cyber security and governance, and professional-services clients are pressing for a share of the efficiency gains, and fee inflation is slowing.

Bailey called himself in February “a realistic optimist”, observing that progress so far had come from well-defined, task-based work and that productivity gains from new technologies take time. The November report carries the same attitude. It does not say that AI has failed to pay. It says that a three-year-old technology has so far added about half a percent, and that the firms producing the numbers expect the next three years to differ.

Back to the panel

The partner in Leeds sends her answer and then reads the question that follows it, the one about the next three years. It asks what she expects. The vacancy still says “hold”. Whether the answer she gives is 0.5% again or something larger depends on what the review tools do with next year’s files and on whether a trainee’s first two years can be bought back later. Both are things the panel cannot ask her, and the Bank will not be able to measure them until the next trainee intake comes and goes.

References

  • Bailey, A. (2026). Speech at the IMF-Saudi Ministry of Finance conference, AlUla (February). Bank of England. bankofengland.co.uk
  • Bank of England (2026). Agents’ summary of business conditions, October 2026. bankofengland.co.uk
  • Bank of England (2026). Monetary Policy Report, November 2026. bankofengland.co.uk
  • Brynjolfsson, E., Chandar, B., & Chen, R. (2026). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab. digitaleconomy.stanford.edu
  • Slok, T. (2026). No Signs of AI Payoff Outside the Tech Sector. Apollo, The Daily Spark. apollo.com