Does AI actually boost productivity? What 2026 research shows

Every AI sales pitch comes with a productivity number. Save ten hours a week. Double your output. Then you look at the research, and it says something closer to "maybe, eventually, for some firms." That gap is one of the most important things a business owner can understand about AI right now, because it decides whether your AI spending turns into results or just turns into spending.

What happened

Two large studies of executives, both published as National Bureau of Economic Research working papers this year, landed on very different sounding answers.

The first, a survey of nearly 750 corporate executives by economists including researchers at Federal Reserve banks, found that AI's effect on labor productivity is positive, varies by sector, and is expected to strengthen in 2026. The biggest gains were in high-skill services and finance. Two details stood out. Executives' perceived gains were larger than the gains the researchers could actually measure, possibly because revenue benefits arrive late. And while larger firms expected AI to shrink their workforce, smaller firms expected modest job gains.

The second paper, covered by People Matters in August, surveyed nearly 6,000 CEOs, CFOs and senior executives in the US, UK, Germany and Australia. About 69% of firms said they actively use AI. Yet 89% reported no measurable effect on labor productivity over the previous three years, and over 90% saw no measurable effect on employment. The executives themselves averaged about 1.5 hours of AI use a week. Looking ahead, they expected productivity to grow 1.4% over the next three years.

Then there's the hidden cost side. In June, The Next Web covered a Harvard Business Review piece by Matthias Holweg and Thomas Davenport arguing that low-quality AI output slowly degrades the information companies rely on. It builds on earlier research from BetterUp Labs and the Stanford Social Media Lab that coined the term "workslop": AI-made work that looks finished but isn't. In that survey of 1,150 US workers, 41% had received some in the past month, and each case took close to two hours to sort out.

Why it matters for your business

Perceived savings are not real savings. The first study found that executives feel more productive than the numbers show. You probably do too. Faster first drafts feel like progress, but if someone spends an hour checking and fixing them, the net gain shrinks.

Light use gives light results. An hour and a half a week is a dabble, not a change in how work gets done. Firms that bolt a chatbot onto the side of an unchanged process shouldn't expect it to show up in the numbers.

Workslop moves cost, it doesn't remove it. When one person uses AI to produce a polished but shallow report, the cost lands on the person who has to read, fix or redo it. On a small team, that's often you.

Small firms may have the edge. It's worth noticing that smaller firms in the first study were the ones expecting AI to add jobs. Fewer layers means a change in one process can show up quickly in the whole business.

The optimists have a fair point: big technologies often take years to appear in productivity statistics, and executives in both studies expect gains to grow. Early evidence of "nothing yet" is not proof of "nothing ever." But it is a good reason to measure instead of assume.

What we'd do

• Pick one process, not ten. Choose a single, repetitive workflow such as quoting, intake, scheduling or invoice matching, and apply AI there properly.
• Measure before and after. Time the task, count errors, track turnaround for a month before the change and a month after. Use the clock, not impressions.
• Count the checking time. Include review and correction in your numbers. That's where workslop hides.
• Set a quality bar. Make it a rule that anything AI-drafted is reviewed by the person who sends it. The sender owns the output, not the tool.
• Redesign the work, don't just add a tool. The firms seeing gains change steps, hand-offs and roles. A new subscription alone rarely moves the needle.
• Cut what isn't working. If a tool hasn't shown a measured gain in a quarter, cancel it.

Our position is simple: AI can make a business more productive, but the research says most companies haven't gotten there yet, mostly because they adopted tools without changing work. Owners who treat AI as a process project with a stopwatch will find out what it's worth. Owners who treat it as a perk will keep feeling productive while the numbers stay flat.