6,000 executives: 9 in 10 report no AI impact yet

The most useful AI adoption number we have seen this year is not a benchmark. It is a survey of nearly 6,000 executives across four countries, and it says that 69% of firms use AI while nine in ten report no measurable effect on employment or productivity. Both things are true at once, and the gap between them is where most small businesses currently live.

What actually happened

The paper is Firm Data on AI (NBER working paper w34836), issued in February 2026 and revised in March. The underlying survey ran November 2025 through January 2026 and covered roughly 6,000 CEOs, CFOs, and senior finance managers in the US, UK, Germany, and Australia, plus about 3,000 US employees. It has been circulating again this week.

From the NBER digest, the numbers that matter:

• 69% of firms actively use AI — 78% in the US, 59% in Australia
• 41% use text generation specifically
• Executives personally use AI about 1.5 hours a week; US workers about 1.8
• More than 90% report no effect on employment over the past three years; 89% report none on labor productivity
• Among the minority who did report a productivity effect, the average was 0.29%

Forward-looking answers flip hard. Over the next three years the same executives expect AI to raise productivity 1.4%, cut employment 0.7%, and lift output about 0.8%, with roughly two-thirds of the headcount reduction coming from slower hiring rather than layoffs. US workers and their employers disagree about direction: workers expect employment to rise 0.5%, executives expect it to fall 1.2%.

Why "adoption" is the wrong metric for your business

Look at the two headline figures together. Sixty-nine percent adoption. One and a half hours a week. That is not a workforce transformed; that is a subscription and a browser tab. The survey is measuring use, and use is nearly free — anyone can turn on a seat license. What it is not measuring, because almost nobody has it, is a workflow where the AI step is load-bearing and the old step is gone.

That distinction is the whole game. If a tool is optional, it produces no measurable effect by construction — the fallback path absorbs every failure, and the fallback path is what you are still paying for. Nine-in-ten reporting nothing is exactly what a world of optional tools looks like.

So stop asking whether you have adopted AI and ask a narrower question: which single process runs differently now, and what did it used to cost? Pick one — invoice intake, quote generation, first-touch support triage, inventory reconciliation. Write down the current cycle time and the current labor hours before you change anything. That baseline is the only thing that will ever let you answer the productivity question for your own business, and the 89% figure above exists largely because nobody wrote theirs down.

The 0.29% average among firms that did report gains is the other honest signal. Real effects, when they show up, are small and specific. Anyone quoting you a 40% efficiency gain on a discovery call is selling. The operators who will report something in next year's survey are the ones running one measured workflow, not twelve pilots.

Sources: NBER working paper w34836, NBER Digest.