Summary: A February 2026 NBER working paper surveyed nearly 6,000 phone-verified executives across four countries and checked responses against national statistics. It found broad AI adoption but roughly nine in ten executives reported no measurable employment or productivity effect over three years, even as they forecast a 1.4 percent productivity gain ahead. The evidence shifts enterprise AI debates from adoption claims to depth of use and measurable output.
In February 2026, NBER put out working paper 34836, “Firm Data on AI,” the work of a dozen-plus economists — Ivan Yotzov, Nicholas Bloom, Steven J. Davis, Brent H. Meyer, Paul Mizen, Gregory Thwaites, and Ben Zhe Wang among them — drawn from the Federal Reserve Bank of Atlanta, the Bank of England, the Deutsche Bundesbank, and Macquarie University. A February 20 story from AI News pulled the paper into the tech conversation, and its central finding is blunt enough to sting: firms have adopted AI at scale, yet the productivity impact anyone can actually measure hovers near zero.
What earns the survey a serious reading is how it was built. Close to 6,000 executives — phone-verified, unpaid, and overwhelmingly CEOs and CFOs — were polled across the United States, the United Kingdom, Germany, and Australia, and the responses were then cross-checked against ten years of macro output and employment data from each country’s national statistics agency. Set against the vendor questionnaires and Twitter polls that usually fill this space, it stands as some of the most solid evidence available on what AI is genuinely doing inside firms.
Adoption Is Broad, Measured Impact Is Thin
The figures cut a stark contrast. 69 percent of the firms surveyed say they actively use AI in the business, adoption runs markedly higher among younger, more productive companies, and more than two-thirds of the executives themselves use AI tools on a regular basis. Use cases cluster around LLM text generation first (41 percent), then visual content creation (29 percent) and machine-learning data processing (28 percent); within the UK alone, firm-level adoption rose from 61 to 71 percent over the course of 2025.
Wind the clock back three years, though, and roughly nine in ten executives say AI has produced no measurable effect on their firm’s employment or productivity. How hard the tools actually get used is another problem: the average executive logs about 1.5 hours of AI use per week. Broad uptake paired with shallow usage is, for now, the whole picture.
Bosses and Workers Are Expecting Different Futures
The expectations half of the paper is where it gets genuinely interesting. Asked to look three years ahead, these same executives on average forecast a 1.4 percent productivity gain, 0.8 percent output growth, and a 0.7 percent employment decline. Optimism peaks among US executives, who project 2.25 percent productivity growth against 1.86 percent for the UK. Planned downsizing stays modest as well: two-thirds of the adjustment UK firms anticipate is meant to arrive through slower hiring rather than redundancies.
Line that up against workers and the gap jumps out. Through the Survey of Working Arrangements and Attitudes, US employees expect employment to grow 0.5 percent where their executives expect a 1.2 percent contraction, and employees peg productivity gains at 0.92 percent — under half the 2.25 percent their bosses project. One technology, two opposing stories: the employer pictures a smaller headcount, the worker pictures a bigger paycheck, and one of those camps is guaranteed to be disappointed.
What Keeps the Numbers This Flat
A few structural explanations are worth spelling out:
- Complementary investment hasn’t landed yet: process redesign, data cleanup, and worker training tend to arrive after the tools do — productivity research’s familiar J-curve playing out once again.
- Usage stays too shallow to matter: 1.5 hours a week will not bend a company’s cost structure.
- Surveys measure different things: McKinsey counts 88 percent of firms using AI in at least one function, while the US Census Business Trends and Outlook Survey climbed from roughly 9 percent in early 2024 to 18 percent by December 2025 — a fivefold spread driven by how the questionnaire asks, and a caution against treating any single adoption rate as gospel.
- Selection effects cut the other way too: adoption piles into firms that were already younger and already more productive, which makes crediting AI itself for the gains nearly impossible.
Takeaways for Product Teams
For one, the adoption pitch has run its course; the next procurement argument will be measurable change in output, which reads as a directional endorsement of every ROI-obsessed enterprise AI product on the roadmap. For another, depth of use is the new north-star metric: a tool used 1.5 hours a week and a tool built into the daily workflow are commercially different animals. And expectation management is now table stakes: with executives banking on an average 1.4 percent productivity gain, any pitch promising tenfold returns falls apart in front of a CFO. The evidence, for now, sits with the incrementalists — and should the next three years prove them wrong, this paper is the baseline everyone will be measured against.