AI everywhere, productivity nowhere? Why the new Solow moment forces a rethink
Almost every company uses AI, yet little has shown up in productivity so far. What the Solow paradox has to do with it – and why the lever lies in implementation, not in the next tool.
In strategy meetings and annual reports, AI is everywhere. Ask about its measurable effect on productivity and results, though, and the room often goes quiet. This contradiction is not new. Economists recognise a familiar pattern from the history of computing: the Solow paradox.
Déjà vu from IT history
From the 1970s onwards, computers spread rapidly through companies. A productivity boost nevertheless failed to materialise for a long time; US productivity growth even slowed. In 1987, the economist Robert Solow summed it up: you can see the computer age everywhere – except in the productivity statistics.
One explanation was as mundane as it was important. New technology first produced more data, more reports and more coordination. The organisational benefit came later, once companies reorganised their processes around the technology.
What executives report about AI today
Today, a similar picture emerges. A research group around Nicholas Bloom and Steven J. Davis surveyed nearly 9,000 CFOs, CEOs and other senior executives in the US, the UK, Germany, Australia, Canada, Mexico and South Africa for the National Bureau of Economic Research (survey in summer 2026, version of October 2026):
- 78 percent of firms use AI.
- 80 percent of executives use AI regularly – on average around 2.1 hours per week.
- In their assessment, AI had essentially no net effect on employment over the last three years.
- By their own account, AI raised labour productivity by 0.2 percent per year. For the next three years, respondents expect 0.6 percent per year.
So AI has arrived – in the tools. In the results, hardly yet. Expectations nevertheless remain high.
Why the breakthrough is missing
Between technical availability and measurable impact lies a gap that no model bridges on its own. Three points stand out:
- Depth of use, not the existence of tools: an AI account does not change a process. Impact only appears when workflows, responsibilities and quality checks are redesigned.
- Complexity eats the gains: many parallel tools without priorities create friction, switching and rework. Time saved in the individual task is lost in the whole.
- Trust and skills: if employees do not trust AI results or do not know when to check them, work is done twice – or not at all.
Mixed evidence is normal in such phases. Early signals in individual teams respond faster than macroeconomic figures.
The J-curve effect: first friction, then leverage
Economists around Erik Brynjolfsson describe this pattern as the productivity J-curve: investment and conversion effort rise first, before the impact becomes visible. We know this from earlier technology waves. When processes, roles and management logic follow, selective AI use can turn into a system-wide effect.
How large that effect turns out to be depends less on the model than on implementation. Value is created where AI is embedded in decision paths and customer processes – in sales, for example, where AI agents in sales prepare cases, follow up and let people decide at the right points.
What you can do now
- Choose a few measurable use cases instead of a tool collection without priorities.
- Define impact upfront: which outcome metric should move – revenue, lead time, error rate, margin? And what does the baseline look like?
- Recut the work: design processes, roles and decision rights for agents and people from the outcome – this is what we mean by AI-first.
- Build skills close to daily work: with clear standards for when a person checks and approves.
- Lead the change: AI success is an implementation and leadership task, not a pure IT project. Our article on change management for AI adoption describes how it works.
Conclusion: AI is a lever, not a shortcut
The new Solow paradox is no proof that AI does not work. It shows that impact does not come with the tool but with the design. If you apply the lever properly, the apparent paradox becomes a reliable contribution to results.
That is exactly why REVAGO starts not with technology but with your goals. In AI-first coaching you find out whether and how AI agents can secure and grow your revenue. How we then make business value assessable is described in Value: How we make the business value of AI agents assessable.
