A National Bureau of Economic Research study released in February 2026 — and back in the headlines this week after Fortune broke it down for a wider audience — surveyed 6,000 CEOs, CFOs, and senior executives across the US, UK, Germany, and Australia. The headline finding is stark enough that we have read it three times to make sure we did not misread it.

Nearly 90% of firms reported that AI had no measurable impact on either employment or productivity over the past three years. This is despite over $250 billion in AI investment in 2024 alone.

The number that should worry the Fortune 500

Two-thirds of executives in the study said they personally use AI tools — but only about 1.5 hours per week. Another 25% reported no AI use at all. And yet the same executives forecast 1.4% productivity gains and 0.8% output increases over the next three years. The gap between expectation and reality is now a chasm.

Apollo's Chief Economist Torsten Slok put it most bluntly: 'AI is everywhere except in the incoming macroeconomic data.' Nobel laureate Daron Acemoglu, who has spent two decades studying technology and labour markets, called the measured 0.5% productivity gains 'just disappointing relative to the promises.'

The 1987 echo no one in tech wants to remember

Economists have started openly comparing 2026 to 1987, the year economist Robert Solow famously wrote: 'You can see the computer age everywhere but in the productivity statistics.' Despite transistors, microprocessors, integrated circuits, and the entire IT revolution, US productivity growth had slowed from 2.9% (1948-1973) to 1.1% in the years after. It took roughly fifteen years before computers actually showed up in the productivity numbers.

If history rhymes — and it usually does on these things — the AI productivity dividend may take another decade to land at the macro level. For shareholders waiting on $250 billion of capex to pay back, that is a long time.

Why big companies are stuck

Buried inside the NBER study and a complementary Boston Consulting Group analysis are three explanations that line up with what we hear from the small business owners we talk to every week:

  • Tool sprawl. BCG found that workers using three or more AI tools at once report 'brain fog' and increased error rates. Enterprise procurement loves to buy AI tools; humans can only use a few.
  • No workflow integration. A chatbot bolted onto an unchanged process saves nothing. The change has to happen at the process level, which most large companies are too slow to do.
  • Executive disconnect. Two-thirds of the executives in the study use AI 1.5 hours a week. That is not enough time to understand what it can and cannot do, let alone redesign workflows around it.

Why small businesses are quietly the exception

Here is the part the macro studies miss. They sample CEOs and CFOs of firms large enough to have CFOs. They do not sample the four-person bakery in Portland we wrote about last month, or the two-person consultancy that runs entirely on ChatGPT and Notion, or the eight-person agency that just cut its admin time by 40% with Canva and Buffer.

Small businesses have three structural advantages that the Fortune 500 cannot replicate at speed:

  • The owner uses the tool. There is no executive disconnect when the CEO is also the person opening ChatGPT at 7am. The feedback loop between 'this works' and 'change the workflow' is instant.
  • There is only one workflow to change. Redesigning how you handle customer emails takes an afternoon for a four-person shop. It takes a year and a 40-page change-management document for a 4,000-person enterprise.
  • Tool stacks stay small. A two-person business that adopts three AI tools is operating at the BCG-recommended limit by accident. A Fortune 500 division running fourteen AI vendors at once is operating well past it.

What this means for you

If you run a small business in 2026, the headline of this study is not a warning. It is a quiet permission slip. You do not need to wait for the macro data to validate that AI is worth your time — you can already see it in your own week. You also do not need to feel guilty for not adopting more tools. Three used well will beat ten used badly, every time.

What the macro data does suggest: if you have a competitor that is much bigger than you, they are probably not getting nearly as much out of AI as their LinkedIn posts suggest. The playing field, on this one specific dimension, is more level than it has ever been. And it may stay that way longer than anyone in San Francisco wants to admit.

The bottom line

The biggest economic story in tech this week is not a new model release or a $40 billion investment round. It is that 90% of executives just admitted, in print, that AI has not done what their boards and their consultants told them it would.

And small businesses — the ones too small to have boards or consultants — keep on quietly using AI to save four hours a week, send invoices faster, and write better customer emails. That gap between the headline story and the ground truth is, frankly, where the most interesting opportunity in business sits today.