From November 2025 through summer 2026, central banks and universities in the US, UK, Germany, Canada, Australia, Mexico, and South Africa added the same AI questions to the regular executive surveys each conducts. Nearly 9,000 CEOs, CFOs, and senior financial executives across the seven countries answered two questions: whether AI had changed their company's headcount over the previous three years, and how they expected it to change headcount over the next three.
More than four in ten executives expected AI to increase or decrease their company's headcount over the next three years. But the projected cuts and increases offset each other, leaving the seven-country average nearly unchanged at a 0.4% decline. The figures come from the October 15, 2026 revision of “Firm Data on AI,” a paper by 21 authors including Ivan Yotzov of the Bank of England, Jose Maria Barrero of ITAM, and Nicholas Bloom and Steven J. Davis of Stanford.
Among roughly 9,000 executives in seven countries, 87% said AI had not changed their headcount over the past three years
The employment questions asked how AI adoption had affected company headcount over the past three years and how executives expected it to affect headcount over the next three. Respondents chose from five categories: increased by more than 5%, increased by less than 5%, no effect, decreased by less than 5%, or decreased by more than 5%. The researchers assigned values of +7.5%, +2.5%, 0%, -2.5%, and -7.5% to the categories to calculate an average.
Of the 8,744 executives who answered the question about the past three years, 87% chose no effect. Six percent reported a decrease in headcount (1% by more than 5% and 5% by less than 5%), while 7% reported an increase (5% by less than 5% and 2% by more than 5%). Applying the category values produces an average of +0.06%, effectively zero. Germany had the highest share reporting no effect, at 95%, followed by the US, Canada, and the UK at 88%, South Africa at 82%, Mexico at 80%, and Australia at 74%. The portion of productivity growth attributed to AI, with productivity defined as sales per employee, was just +0.46% cumulatively over the past three years, or 0.2% per year.
These figures are based on executives' own judgments about whether AI caused changes in headcount, not administrative data such as payroll records. In their conclusion, the authors describe what they measure as “(perceived) effects.” The researchers offer two reasons to trust the responses. The US and UK surveys initially verify respondents' positions and companies by phone, and respondents receive no compensation. At the Federal Reserve Bank of Atlanta, which runs the US survey, staff call to verify respondents before admitting them to the panel. In addition, US and UK executives' one-year forecasts for sales and employment closely matched actual outcomes. That validation, however, covered sales and employment overall; it did not separately verify whether changes were caused by AI.
Expected cuts and increases offset each other, leaving a seven-country average decline of 0.4%
Executives were divided when asked about the next three years. Of the 8,869 who answered, 28% expected headcount to fall (8% by more than 5% and 20% by less than 5%), while 16% expected it to rise (10% by less than 5% and 6% by more than 5%). Another 57% expected no effect, 30 percentage points below the 87% who reported no effect over the past three years. The three figures add up to 101% because of rounding.
| Executive response | Past three years | Next three years (expected) |
|---|---|---|
| Headcount decreased or will decrease | 6% | 28% |
| No effect | 87% | 57% |
| Headcount increased or will increase | 7% | 16% |
| Average effect | +0.06% | -0.4% |
| Responses | 8,744 | 8,869 |
Together, the shares expecting a decrease or an increase add up to 44%. Applying the values assigned to each category, the projected cuts and increases offset each other, leaving an average projected change of -0.4% across the seven countries. The authors put it this way:
“more than 40% of firms anticipate positive or negative employment effects that roughly offset only in the aggregate.”
Original: “more than 40% of firms anticipate positive or negative employment effects that roughly offset only in the aggregate.”
The authors read this as a sign of substantial job churn between firms. The two sides are not equal, however. Nearly twice as many expect a decrease as an increase, so the average points to a decline rather than zero. Forbes reported on the paper under the headline “It’s A Wash,” but the figures in the paper tilt more toward a decrease than that headline suggests.
Executives also expected productivity to rise over the same three years. In this survey, productivity means sales per employee. Across the seven countries, executives expected a cumulative 1.7% rise in productivity, a 0.4% fall in headcount, and a 1.3% increase in sales, or output in the paper's terminology. If companies improved productivity solely by cutting staff while keeping sales unchanged, a company expecting productivity to be 1 percentage point higher should also expect headcount to fall by about 1 percentage point more. Comparing companies, the researchers found that for each 1 percentage point of higher expected productivity, the expected headcount decline was only 0.07 percentage points greater (Figure A24 in the paper). The authors interpret this result as follows: “Firms therefore expect a substantial share of the productivity gains from AI to arrive through higher output rather than lower headcount.” In other words, companies expect a substantial share of AI productivity gains to come from selling more rather than cutting staff. There are also objections to defining productivity as sales per employee. In comments on a February 2026 article in The Register about the first edition, readers argued that this measure excludes sectors without sales, such as government or healthcare, and that companies do not accurately know office workers' productivity in the first place.
The authors offer their own caveat. They say -0.4% could be too pessimistic because it excludes the economy-wide effects of AI productivity gains, jobs at companies that do not yet exist, and young companies that are underrepresented in the sample. In a footnote, they also report that companies' AI employment expectations are unrelated to how inaccurate their usual employment forecasts have been. That is evidence that the figures are not simply the result of executives being pessimistic by nature.
Decreases and increases can also offset each other within a company. The survey asked only about total company headcount, so it cannot show which employees are lost or added within a firm. The authors acknowledge this limitation:
“a decline in employment among younger workers could be offset by an increase among older workers, leaving little change in a firm’s total employment.”
Original: “a decline in employment among younger workers could be offset by an increase among older workers, leaving little change in a firm’s total employment.”
When the same researchers separately asked US executives about hiring over the next 12 months by education level, the expected cuts differed. US executives expected AI to reduce hiring of college graduates by 0.8% and hiring of people without a college degree by 1.1%. They expected the largest decrease, -1.8%, in hiring people without a college degree in business services.
Nearly half of US and UK executives expected cuts, while more Mexican executives expected growth
The extent to which the two expectations offset each other varies by country. In the US and UK, nearly half of executives expected their company's headcount to fall, putting the average projected decline in each country at more than three times the seven-country average of -0.4%. In Mexico, the reverse was true: more executives expected headcount to rise than to fall.
| Country | Executives expecting a decrease | Executives expecting an increase | Average expectation (cumulative over three years) |
|---|---|---|---|
| US | 48% | 14% | -1.26% |
| UK | 47% | 12% | -1.31% |
| Mexico | 16% | 26% | +0.585% |
Among the other four countries, Australia was close to the seven-country average at -0.50%, while Canada at -0.08%, Germany at -0.06%, and South Africa at +0.06% were close to zero. South Africa's average expectation was positive, but so close to zero that it did not point clearly in either direction. The data also came with different conditions in each country. Germany's latest survey did not include the employment question, so the researchers used responses from January 2026. Mexico surveyed only establishments with more than 100 employees, and South Africa had responses from just 121 establishments.
UK executives expected sales per employee to rise 2.62% and headcount to fall 1.31% over the next three years. The largest expected headcount declines were in finance and insurance (-1.9%), followed by wholesale and retail and professional and scientific services (-1.7% each).
A Bank of Korea survey that was not included in the seven-country analysis also appears in the appendix. When 749 Korean companies were surveyed in June 2026, 54% expected headcount to fall over the next three years, 24% expected a decline of more than 5%, and the average expectation was -1.8%. Korean respondents used AI for about 4.1 hours a week, roughly twice the seven-country executive average of 2.1 hours. The researchers did not combine these results with the main analysis because the survey's scope differed.
Beginning in March 2026, the US and UK surveys also asked about spending on AI, including software, subscriptions, hardware, training, and IT support. Just as headcount expectations differed across companies, AI spending was concentrated among a small number of firms.
In the US, 1.8% of companies account for half of projected national AI spending in 2026; in the UK, the figure is 1.3%. The Gini coefficient for inequality in AI spending between companies is 0.90 in the US and 0.92 in the UK, far above the countries' income Gini coefficients of 0.42 and 0.32, respectively. According to figures summarized by the same researchers on the Atlanta Fed blog, the median US company spends $200 or less on AI per employee, while a company must spend at least $2,800 to reach the top 10%. The paper does not separately report whether companies that spend more on AI expected larger headcount reductions.
To see how AI may change jobs, look beyond the average to which companies and countries expect cuts or growth
In their conclusion, the authors write that AI's major effects may come through differences, or dispersion, across companies, industries, and workers rather than through economy-wide totals. This survey directly shows the divide between companies. That raises the next question: which companies expect to cut and which expect to grow? The closest figure in the paper concerns executives' own time spent using AI.
Among UK executives grouped by their own AI use, those who did not use AI expected their company's headcount to fall 0.7% and productivity to rise 1.1%. Those who used it for more than five hours a week expected headcount to fall 2.2% and productivity to rise 3.9%. This is a correlation; it does not mean that spending more time with AI causes executives to decide on layoffs. The authors interpret the result as showing that executives with direct experience of AI see both larger productivity gains and larger headcount reductions, rather than viewing the technology only positively. Executives who spend more time using AI themselves expect both greater staff reductions and greater productivity gains.
Forecasts and discussions about AI and jobs have focused on how many jobs the economy as a whole will gain or lose. Now that four in ten executives expect their company's headcount to change, it is time to focus on which companies expect to cut staff and which expect to add them.
For a closer look
Firm Data on AI (October 2026 revision) The 84-page paper contains the seven-country survey questions, country-level distributions, and the concentration of AI spending. stevenjdavis.com
How Much Are Firms Spending on AI (and What Will Happen to Headcounts)? In this Atlanta Fed article, the same researchers briefly summarize the distribution of AI spending among US companies and hiring expectations by education level. atlantafed.org
Firm AI Use Dashboard The researchers' dashboard shows survey results by country and the actual survey questions. firmdataai.com