89% of Executives See No AI Productivity Impact. Here Is What the Data Actually Says.

A new working paper from the National Bureau of Economic Research surveyed nearly 6,000 senior business executives across the US, UK, Germany, and Australia. The finding that is getting attention: 69% of those firms actively use AI, and 89% of executives report no measurable impact on labor productivity over the past three years.

That number sounds like a scandal. It’s not. It is a system problem with a specific cause.

The NBER working paper is not arguing that AI does not work. It is documenting exactly where the benefit disappears. Adoption is running at the organizational level. Impact is running at the task level. The gap between those two planes is where most enterprise AI investment is currently getting lost.

The research is consistent with a pattern showing up across multiple data sources this year.

Stanford’s HAI AI Index 2026 reports that 88% of organizations now use AI in at least one business function. Gallup’s February 2026 survey of nearly 24,000 US employees found that half of American workers now use AI in their role at least a few times a year, up from 21% in mid-2023. Employee adoption has tripled in three years.

And yet the P&L is not moving.

Brynjolfsson, Li, and Raymond published a study in the Quarterly Journal of Economics documenting a 14% average productivity gain for customer support agents using a generative AI assistant. For novice workers, that gain climbs to 34%. For experienced agents, the effect nearly disappears. A separate METR study found that experienced open-source developers given AI coding tools completed tasks 19% slower, not faster.

These findings point to the same underlying structure. AI works well at the task level for specific skill gaps and specific workflow types. It does not automatically aggregate up to firm-level productivity gains. The reason is that firm-level productivity depends on organizational structure, workflow design, and decision architecture, none of which change when you issue a software license.

Microsoft’s 2026 Work Trend Index found that only 16% of AI users qualify as what Microsoft calls “Frontier Professionals,” meaning workers who have actually redesigned their workflows around AI rather than bolting the tool onto existing processes. The remaining 84% are using AI as a faster version of the same process they had before. That does not move a P&L.

This is the real productivity gap. Not capability. Process.

The NBER data reinforces this from the executive side. Those same executives who reported no current impact predicted that AI will boost productivity at their firms by an average of 1.4% over the next three years. They believe the gains are coming. They are just not seeing them yet, in the same period when adoption has gone mainstream.

That expectation gap is where enterprise leadership needs to focus.

The firms that will close this gap are not the ones buying more AI tools. They are the ones doing the harder work: restructuring the processes that AI is supposed to accelerate. That means questioning whether the workflow AI is being applied to is the right workflow, not just whether AI is being applied at all.

Three questions worth running against any major AI initiative right now:

One: Are the workers most affected by this tool the workers with the highest skill gaps in this domain? The data shows AI generates the largest gains for novice and low-skilled workers in specific task categories. If you are deploying AI primarily to your most experienced people without redesigning what those people do, expect modest gains.

Two: Has the underlying workflow been redesigned, or just accelerated? Speed improvements on a broken process produce a faster broken process. The BCG study in Organization Science this year found that consultants using AI completed 12.2% more tasks at 40% higher quality inside their skill zone, but were 19 percentage points more likely to produce wrong answers on tasks outside it. The workflow design matters as much as the tool.

Three: What is the governance mechanism that connects AI usage to business outcomes? The NBER data shows a specific finding: executives who regularly use AI personally average only 1.5 hours per week of actual AI usage. These are the people setting AI strategy. If senior leadership is not in the workflow, the strategy is being set from outside the problem.

The 89% number is not evidence that AI does not work. It is evidence that workflow transformation is harder than software deployment. Those are two different problems with two different owners.

The firms that close the productivity gap in the next 18 months will not be the ones that bought the most tools. They will be the ones that treated AI deployment as an organizational redesign project from the start.

---

Robin Green is an enterprise technology executive and the author of The Intelligence Loop. He is Chief Revenue Officer at Occams Advisory.