Augmentic BV
Haaswijkweg east 12B
3319 GC Dordrecht
The Netherlands
A position paper on why productivity research on AI systematically misses the mark, and what it can tell us.
“Productivity gains from AI? Most companies don't see it yet.” That's what the Financieele Dagblad wrote on March 14, 2026, based on a large-scale survey of six thousand CEOs and CFOs in four countries. The conclusion: 89% of the surveyed companies see no effect of AI on productivity or employment. Only 10% report some improvement.
These are impressive figures. They sound convincing. And they are also largely correct.
But the article draws the wrong conclusion. Not because the data is incorrect, but because it doesn't ask the deeper question: Why Do those 89% not see anything? And what do the 10% who do see results do differently?
The answer to that question changes the whole story.
What the research actually says
The study referred to by the FD is thorough. It concerns NBER Working Paper 34836, authored by economists from Stanford and four central banks—Atlanta Fed, Bank of England, Deutsche Bundesbank, and Macquarie University—with identical research questions in four countries. The sample is representative, the respondents are senior, and the methodology is transparent.
The figures mentioned by the FD are exactly correct: 69% of the surveyed companies use AI, but only 11% report a productivity increase, and more than 89% see no effect. These are figures that seem to explain themselves: AI is there, but it doesn't work.
Solely: the study has three fundamental limitations that the article does not mention.
Firstly, it's a self-reported survey. CEOs and CFOs estimate if they see productivity gains without objective measurement. That's a fundamental difference. A 2025 study with software developers showed they believed AI made them twenty percent faster, while they were actually nineteen percent slower on complex tasks. Therefore, the perception gap can be almost forty percentage points.
Secondly, researchers measure productivity as “revenue per employee”—a crude measure that completely misses quality improvements, error reduction, shorter lead times, and the increased complexity of the remaining human work. A lawyer who accomplishes in six hours with AI what would have taken twelve hours otherwise, but uses the freed-up time for additional client contact instead of more billable hours, is not visible in revenue per employee. The productivity gain is real; the metric does not see it.
Third - and this is the most striking detail that the article omits - respondents use AI for an average of an hour and a half per week. At that level of use, it is hardly surprising that the effect is immeasurable. An employee who spends an hour and a half a week using a tool while his work processes are otherwise unchanged is not going to move his organization's productivity numbers.
And there's something else: the researchers themselves are not pessimistic. The CEPR article based on the same research is titled “Firms predict an AI productivity boom is coming.” The same CEOs and CFOs who see no effect today expect an average of 1.4 percent productivity growth per year in three years — and a quarter expect a 0.7 percent decline in employment per year. That's not the picture of people who think AI doesn't work. That's the picture of people who realize they aren't working with it in a way that yields results yet.
The other research that the FD does not mention
While the article concludes that the effect “remains shrouded in mystery,” there is a growing body of research showing where AI demonstrably works – with hard numbers.
The most relevant study for the European context is an analysis by the Bank for International Settlements, the European Investment Bank, and CEPR, published in 2026. Researchers analyzed over twelve thousand European companies using a causal methodology—not a survey, but an instrumental-variables method that isolates real effects. Outcome: AI adoption increases labor productivity by an average of four percent, with no short-term employment loss.
But the most interesting thing in that research isn't the average. It's the multipliers. Every additional percent that companies spend on employee training strengthens the productivity effect by 5.9 percentage points. Every additional percent on software and data infrastructure by 2.4 percentage points. In other words: technology itself is only one part of the equation. Organizations that only purchase an AI tool might see a one or two percent improvement. Organizations that simultaneously invest in training, infrastructure, and process adjustment see ten to twenty percent.
A pre-registered experiment with 758 BCG consultants — the so-called “Jagged Frontier” study by Harvard Business School — showed something similar. Consultants who used AI for tasks for which the tool is suited completed 12 percent more tasks, worked 25 percent faster, and produced 40 percent higher quality. But consultants who deployed AI for tasks beyond its scope of competence performed nineteen percentage points worse than colleagues without AI. The tool works — but only if the employee understands when it works and when it doesn't, and adapts their workflow accordingly.
And then there are the organizations that have implemented AI not just as a tool, but as a catalyst for fundamentally different work. Klarna reduced customer service resolution time from fifteen minutes to under two minutes—a drop of over eighty percent—and lowered cost per transaction by forty percent. AstraZeneca achieved fifty percent shorter development timelines and seventy percent time savings on regulatory documentation. Rolls-Royce increased machine utilization by thirty percent through AI-driven planning. In the legal sector, 62 percent of European professionals in a recent Wolters Kluwer study report saving six to twenty percent of their work week.
None of these results are visible in a survey asking if the CEO sees an impact on revenue per employee.
We have seen this story before
There's a pattern in how transformative technologies are received throughout economic history—and it's the same every time.
In 1987, economist Robert Solow wrote his famous paradox: “You can see the computer age everywhere but in the productivity statistics.” Computer use had grown strongly for ten years, but productivity growth had fallen from nearly three percent to just over one percent. It sounded like a verdict: computers don't work.
Until they started to work. Between 1995 and 2005, American productivity growth rose by one and a half percentage points per year — and that increase was concentrated in precisely the sectors that had fundamentally redesigned their work processes around the new technology.
The parallel with electricity goes even deeper. Electricity became commercially available starting in the 1880s. But its impact on productivity wasn't visible until around 1920—a forty-year delay. Factories that simply replaced their steam engines with electric motors saw minimal improvement. It wasn't until a new generation of factory designers completely rethought the production logic—single-story buildings, individual motors per machine, different logistics, different staffing—that productivity exploded. Electrified factories that took that step achieved thirty percent more output. Electricity was responsible for half of all productivity growth in the 1920s.
The technology had been around for years. The difference wasn't in the technology. It was in the willingness and ability to redesign work itself.
89% does the wrong thing and 10% does the right thing — and that's exactly right
This is where the research gets interesting.
Erik Brynjolfsson, the Stanford economist cited in the FD article, publishes research himself on what he calls the “J-curve”: the pattern where organizations first see a drop in productivity after introducing new technology—due to investments, learning costs, and process disruptions—before the profits become visible. Those profits do materialize, but only once the complementary investments have been reaped: processes redesigned, people trained, infrastructure adjusted, and the organization reorganized.
In an interview with Fortune, Brynjolfsson states that only ten to fifteen percent of companies are truly investing in those complementary adjustments. The remaining 85 to 90 percent are falling behind. This figure is almost identical to the 89/11 split from the NBER study cited by the FD.
McKinsey's State of AI Survey 2025 confirms this. Of 25 investigated factors, work process redesign has the strongest effect on AI's profit impact. Only six percent of companies belong to the group achieving significant results—and they are three times more likely to be engaged in process redesign than the rest. MIT reports that 95 percent of generative AI pilots have no measurable bottom-line impact, primarily because organizations deploy AI without changing their ways of working.
These are not conflicting findings. They tell the same story. The 10% that do see results in the NBER study are almost certainly the organizations that do what historical logic dictates: not pasting AI onto existing work, but redesigning work around the new possibilities.
What it actually means
There is a fundamental difference in how organizations approach AI.
Most organizations approach it as a tool: an additional aid on top of existing processes. An employee uses ChatGPT to write an email faster. A team uses an AI assistant to summarize meeting minutes. Handy, certainly. But the existing workflow remains intact. The structure of meetings doesn't change. The way decisions are made doesn't change. The division of tasks doesn't change. The work is the same; there's just a new layer on top.
A small group of organizations approaches it differently. They ask a different question: what could this work be if we were allowed to redesign it from scratch, with the possibilities that AI offers? That yields different answers. Not “how do I use AI in writing this report,” but “should this report actually exist, and if so, what would the process look like if AI were a structural part of that process?” Not “how does AI help me with this task,” but “which part of this work truly adds value, and how do I organize the rest so that people can focus on that part?”
That's a different question. And it's a harder question, because it touches on how work is organized, what people do and why, which processes are truly necessary, and which exist because they always have. It asks for a willingness to see work itself as the object of change — not the tool, but the work. That's precisely the distinction between organizations that adopt AI and organizations that redesign work. It sounds like a nuance. It's a fundamental difference in outcome.
That's exactly why most companies see no effect. Not because AI doesn't work. But because they treat it as an IT project instead of an opportunity to redesign work.
What this means for decision-makers
If you are a director, manager, or decision-maker reading this, there are a few things worth considering.
The question “are we already seeing productivity gains from AI?” is the wrong question. It measures whether the layer you've added on top of your existing workflow is delivering anything. This is rarely the case, and that makes sense.
The better question is: which part of our work takes the most time and delivers the least value? What would our employees do if they were freed from that burden? And what does the work process look like if we design it with AI as an integral part rather than an add-on tool?
These are questions that no AI can answer. They call for leadership, a willingness to redesign, and the courage to question work processes that have existed for years. The technology is available. The question is what you do with it — and how fundamentally you are willing to look.
It's also worth knowing that the timeline is long. The OECD, IMF, Goldman Sachs, and McKinsey are remarkably in agreement on one point: the large-scale macroeconomic impact of AI is not yet visible in statistics, and this was also the case with electricity and the PC. This says nothing about whether it works. It says something about the time it takes for organizations and sectors to fundamentally change their way of working.
In five years, companies making this transition now will have a head start that will be difficult to overcome. The companies that wait until the statistics prove it will be doing the same thing as the manufacturers in 1900 who waited for proof that electricity was worthwhile.
How Augmentic views this
AI is not an IT project. It's a catalyst for taking work seriously – to examine what work truly is, what it costs, what it yields, and how it can be fundamentally improved.
This doesn't mean providing tools and hoping organizations will do something with them. It means working with organizations to look at the work itself: which processes are suitable for redesign, where the real potential lies, and how to structure the collaboration between humans and AI so that people are freed up to do the work that truly matters.
Our approach is called Work Design. Not AI adoption, not implementation, not change management — but work redesign. Because the question is not how to get employees to use an AI tool. The question is what work they could do if they were no longer trapped in the work that AI can now do for them.
The results we are seeing align with what the European BIS study shows: a four percent productivity gain as a starting point, increasing as Work Designs mature and employees grow in their AI skills. Not spectacular as a number. But real, measurable, and structural—because the work itself has changed, not just the tool.
The question that the FD doesn't ask, but that matters, isn't whether AI will yield results. That question has already been answered by enough robust research to be certain. The question is whether your organization is prepared to do the work necessary to make it work.
That's a different question. And the answer to it determines whether you will be using AI in five years – or actually working with it.
Augmentic helps organizations redesign work with AI. Through our platform Augmentic_OS, we translate this approach into daily work practices — so employees don't have more tools, but better workdays. Want to learn more? Contact us at hello@augmentic.nl
#augmentic #betterworkdays #AI #workdesign #agenticai
Resources
Reason
- Financial Times, March 14, 2026 — “Productivity gains from AI? Most companies don't see it yet” https://fd.nl/politiek/1589631/productiviteitswinst-door-ai-de-meeste-bedrijven-zien-het-nog-niet?gift=wTpht&utm_medium=social&utm_source=link&utm_campaign=earned&utm_content=20260316
Research and scientific sources
- Bloom, N., Davis, S., et al. (2026) — Firm Data on AI — NBER Working Paper 34836 https://www.nber.org/papers/w34836
- Bloom, N., Davis, S., et al. (2026) — Firms predict an AI productivity boom is coming — VoxEU / CEPR https://cepr.org/voxeu/columns/firms-predict-ai-productivity-boom-coming
- BIS / EIB / CEPR (2026) — AI adoption, productivity, and employment: evidence from European firms — BIS Working Paper 1325 https://www.bis.org/publ/work1325.htm
- Dell’Acqua, F., Mollick, E., et al. (2023) — Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality — Harvard/BCG, via SSRN https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4573321
- Brynjolfsson, E., Rock, D., Syverson, C. (2021) — The Productivity J-Curve: How Intangibles Complement General Purpose Technologies American Economic Journal: Macroeconomics https://www.aeaweb.org/articles?id=10.1257/mac.20180386
- David, P. (1990) — The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox — American Economic Review https://ideas.repec.org/p/wrk/warwec/339.html
- McKinsey Global Institute (2025) — The State of AI: How Organizations Are Rewiring to Capture Value https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
- IMF (2025) — AI and Productivity in Europe https://www.imf.org/en/publications/wp/issues/2025/04/04/ai-and-productivity-in-europe-565924
- UWV (2026) — Strong increase in AI use in the workplace Employer survey 2025 https://www.uwv.nl/nl/nieuws/sterke-toename-ai-gebruik-op-werkvloer
- Wolters Kluwer (2026) — Future Ready Lawyer Survey 2026 https://www.wolterskluwer.com/en/expert-insights/future-ready-lawyer-2026-webinar-series-scaling-ai-across-organizations
News articles and analyses
- Fortune, February 15, 2026 — One of Stanford's original AI gurus says productivity liftoff has begun https://fortune.com/2026/02/15/ai-productivity-liftoff-doubling-2025-jobs-report-transition-harvest-phase-j-curve/
- Fortune, February 17, 2026 — Thousands of CEOs just admitted AI had no impact on employment or productivity https://fortune.com/2026/02/17/ai-productivity-paradox-ceo-study-robert-solow-information-technology-age/
- Fortune, March 3, 2026 — Goldman finds ‘no meaningful relationship between AI and productivity at the economy-wide level,’ but a 30% boost for 2 specific use cases https://fortune.com/2026/03/03/goldman-earnings-ai-anxiety-no-meaningful-impact-productivity-economy-30-percent-in-2-areas/
