Research
An NBER survey of nearly 6,000 executives across the US, UK, Germany and Australia finds 69% of firms actively use AI, yet nine-in-ten report no measurable impact on productivity or employment over the past three years
Executives personally using AI average just 1.5 hours a week with it. Looking ahead, they expect AI to lift labour productivity by only 1.4% and cut employment by 0.7% over the next three years, a modest number now being cited on both sides of the argument over whether this year's AI capex is justified.
The gap this paper documents is between adoption and impact, not between hype and adoption. 69% of firms actively using AI is a high number, higher than most sceptics would have guessed a year ago, and it means the productivity paradox here isn't about firms dragging their feet on AI, it's about firms using it without the usage showing up in the metrics a CFO actually reports. That distinction matters for how this data gets used in the market, because 'firms aren't adopting AI' and 'firms are adopting AI without visible payoff' argue for completely different trades, and this paper is squarely the second one.
1.5 hours a week of personal executive AI usage is the number that undercuts the more dramatic version of the AI-transformation narrative. Task-level studies the paper references show individual productivity gains of 14% to 55% when someone actually uses these tools intensively on a specific task, drafting, coding, searching documents, so the technology clearly works at the task level. What isn't happening is that task-level time saving compounding into firm-level output, revenue per employee or margin, which is the exact translation problem every prior computer-productivity paradox has hit before eventually resolving, and the open question is whether AI resolves faster or slower than personal computing did.
The three-year forward numbers, 1.4% productivity, 0.8% output, a 0.7% employment decline, are why this paper is already being cited in the AI-capex-selloff argument rather than staying an HR-and-productivity story. Hundreds of billions of dollars in AI infrastructure spending are being underwritten by expectations well above a 1.4% productivity lift, and a survey of nearly 6,000 of the executives actually deploying this technology inside their own firms putting a number that low on the outcome is exactly the kind of data point a skeptic cites against the capex thesis, and exactly the kind of data point a bull can wave away as too early and too backward-looking to capture what's coming.
The employee-employer expectations gap the paper flags, workers expecting AI to raise employment 0.5% at their own firm against employers expecting a 0.7% decline, is a smaller finding but a telling one for anyone trying to price second-order effects. If a firm's own workforce is more optimistic about AI's employment impact than its executives are, that's either workers underestimating the technology's substitution effect or executives overestimating how cleanly they'll be able to cut headcount without disrupting the operations AI is supposed to be augmenting, and which of those two turns out true will show up in the earnings calls that cite AI-driven headcount reductions over the next several quarters.
Read the original: National Bureau of Economic Research - An NBER survey of nearly 6,000 executives across the US, UK, Germany and Australia finds 69% of firms actively use AI, yet nine-in-ten report no measurable impact on productivity or employment over the past three years. Commentary is the independent editorial view of Share Trading; the original article is credited to its publisher.