0.5% in three years: So far, the AI boom mostly pays off for the companies selling it

What this is about

On November 5, the Bank of England (BoE) published its Monetary Policy Report, the regular report the UK central bank uses to explain its interest rate decisions. It contains a number that fits this week's mood: according to companies themselves, AI has boosted their productivity by about 0.5 percent since early 2024. Annualized, that's 0.1 to 0.2 percentage points of extra productivity growth. The BoE itself writes that there is "only evidence of a modest AI-driven increase."

Our thesis: AI demonstrably saves a lot of time on individual tasks. But what has reached economies so far is mainly the money being invested in AI. Two things follow from that. First, growth depends on investment, which is why markets react so nervously to any sign that fewer data centers might get built. Second, job cuts made in AI's name don't prove higher productivity.

What exactly the Bank of England measured

The 0.5 percent comes from the Decision Maker Panel (DMP), a regular BoE survey of finance chiefs at British companies. So the figure is the companies' own assessment, not a measurement from official statistics. According to reporting by The Register and The Telegraph, the report also says:

• AI-related activity explains just under a fifth of UK GDP growth since early 2026. The BoE attributes this mainly to the "upstream development and adoption of AI technologies," not to companies becoming more productive by using AI. That means programming, IT consulting and information services, which include data center services.
• Productivity growth has picked up in AI-adjacent industries such as information and communication, but it's still below the pre-pandemic average.
• So far, AI has had "virtually no aggregate effect" on employment. Some employers, though, are hiring less or leaving vacated positions unfilled because of AI.
• Over the next three years, companies expect about 0.9 percentage points of extra productivity growth a year. Roughly half of that is supposed to come from more output, the other half from lower employment.

We were only able to read some of the AI passages in the report itself. It does confirm, among other things, that demand for AI products such as memory chips is pushing up export prices in Asia and therefore Britain's import prices. According to secondary reports, the detailed AI figures are in Boxes C and D.

The BoE also cautions against attributing all of the recent rise in productivity to AI. Part of it simply comes from companies hiring less, reversing pandemic-era staffing decisions and responding to higher payroll costs. Consultancy RSM UK put numbers on this in September: since 2024, output grew by about 2.8 percent while employment grew by only about 0.5 percent. RSM considers the evidence for an AI boom "thin." The upswing is spread across almost every industry, it says, and finance, consulting and administration, where AI should show up first, have grown more slowly than hospitality and transport.

The industry data: IT services are the winners

A more detailed picture comes from an analysis by BoE economist Sandra Batten, published on August 6 on the Bank Underground blog. She compares individual industries' contributions to growth in output per hour between 2010–2019 and 2023–2025:

Among industries that only use AI, office and business support services perform best: classic automation cases like scheduling and invoice processing. Manufacturing also contributes positively. Batten places this in a familiar pattern: with electrification and the computer, the makers of the new technology became more productive first, and the users only years later. Britain, she writes, "appears to be in phase one."

The evidence is explicitly weak. The link between AI use and productivity explains only about a tenth of the differences between industries (R² = 0.10), and it isn't statistically significant by the usual standard (p = 0.065). Batten speaks of correlation, not causation. For the negative figure in finance, she cites measurement problems or offsetting effects as possible causes. In a second BoE post the same month, job postings in the most AI-exposed occupations fell 15 percent over three years, and by more than 20 percent in customer service and administration.

Other hard numbers: the same pattern everywhere

We lined up the BoE figure against other measurements:

What stands out: in both the BoE survey and the four-country study, companies expect roughly five times as much over the next three years as they've achieved so far. In the UK, that's 0.9 instead of about 0.17 percentage points a year; in the CEPR study, 1.4 instead of 0.29 percent. Our take: that's not necessarily wrong. It's the same J-curve earlier technologies followed, and BoE Governor Andrew Bailey expects one for AI. But these are expectations, not results. The 2024 ifo figures are even more optimistic, but the question was asked differently (five years, own business), so they're only roughly comparable. There's no sign of a boost in official German statistics: according to Statista, based on official data, labor productivity per hour worked rose by about 0.4 percent in 2025.

Why big time savings on tasks shrink in the big picture

At the task level, the effects are large and well documented. In the best-known field study, by Brynjolfsson, Li and Raymond, covering 5,179 customer service agents, workers with an AI assistant resolved 14 percent more requests per hour on average, and newcomers as much as 34 percent more. The International AI Safety Report 2026 sums up the research like this: gains of 20 to 60 percent in controlled studies and 15 to 30 percent in most experiments under real working conditions, with outliers in both directions.

A quick calculation shows why so little of that survives. According to the four-country study, US workers use AI for about 1.8 hours a week on average. Out of a 40-hour week, that's about 4.5 percent of working time. Make that time 15 to 30 percent more productive and, on paper, you get only a one-off gain of about 0.7 to 1.4 percent. On top of that, time saved doesn't automatically turn into more output. An HBR study cited by Fortune found that employees often filled the freed-up time with extra work. A result like that would be hard to tell apart from noise in the statistics. So the BoE's order of magnitude is plausible and doesn't contradict the lab studies.

The money is arriving, and it's making markets fragile

What's easy to measure so far is the investment side. In the US, GDP grew at an annualized 3.4 percent in the third quarter, according to the BEA. Spending on information technology alone, mainly computers and servers, contributed 0.86 percentage points. Because much of that hardware is imported from Taiwan and South Korea, imports subtract a big chunk of it again. That's why Goldman Sachs arrives at only about 0.1 percentage points net. "We are currently passing the peak of the boost from the AI boom," the Goldman economists wrote, according to Fortune.

If growth comes more from building than from using, it's vulnerable. That explains this week's sell-off: when investors feared after the US midterms that a Democrat-controlled Congress could slow data center construction, the Nasdaq fell 5.5 percent in a single day and the SOX semiconductor index dropped 12.2 percent. Our take: if the investment wave dries up before users' productivity gains arrive, a gap opens up. That's exactly the risk markets are pricing in right now.

Job cuts aren't proof of productivity

Productivity means output per hour worked. Cutting staff only raises it if output stays the same or grows, and that only shows up in the numbers later. Two cases from this week:

• Berenberg is cutting about 30 jobs in its research and support teams in London, according to Private Banking Magazin and Bloomberg. According to its 2025 annual report, the research department had 107 analysts. The bank has been using Google's Gemini Enterprise since February and says certain workflows are sped up "by up to 90 percent." That's a vendor-style claim about individual workflows, not a measurement for the whole bank. Berenberg has also been restructuring repeatedly since December 2025, and Germany's financial regulator BaFin removed its management board in June. So the cuts have several possible causes.
• Microsoft is offering voluntary buyouts to about 60,000 employees, roughly 27 percent of its workforce, according to GeekWire and IBTimes. Microsoft itself expects fewer than 5 percent to accept, meaning fewer than 3,000 people. The company says the offer doesn't mean it needs fewer staff and doesn't mention productivity gains. At the same time, it invested $45.2 billion last quarter, mostly in data centers. This looks more like costs adjusting to AI investment than AI measurably replacing work.

Our assessment

The Bank of England's 0.5 percent isn't a verdict against AI. It shows where the technology is visible in the economy today: among those who build, run and sell it. What's established is that AI substantially speeds up individual tasks and that AI investment is propping up GDP, accounting for just under a fifth of UK growth this year. The broad productivity boost among users, so far, is only claimed or expected. Companies in every survey promise it for the next three years, at roughly five times what they've achieved so far.

Our take: history suggests users will catch up later. Batten's pattern and the J-curve Bailey is counting on aren't excuses. But anyone citing job cuts or stock valuations today as proof of higher productivity is confusing expectation with results. What matters over the next one to two years is whether productivity picks up in industries like finance, administration and consulting before the investment wave subsides. Until then, statistically speaking, the AI boom rests on construction sites and server imports, and therefore on political decisions like this week's.