The debate over whether AI delivers real productivity gains has shifted from theoretical to empirical. New analysis from the Federal Reserve Bank of St Louis, combined with academic research examining data since 2017, shows measurable productivity effects in industries adopting AI most aggressively. For UK business leaders, the findings present both a warning and an opportunity.
The numbers are starting to tell a story
Headline labour productivity growth in the UK and US looks encouraging. But dig beneath the surface and the picture becomes more nuanced—and more interesting for strategic planning.
Strategic Reality: Strong productivity numbers in both economies could reflect factors unrelated to AI: tariff uncertainty suppressing US hiring, and higher minimum wages clearing out low-productivity jobs in the UK. The real signal lies in industry-level data.
| Metric | US Evidence | UK Evidence |
|---|---|---|
| AI-adopting industries vs others | Correlation with higher productivity growth | No clear correlation found |
| Self-reported time savings | Correlates with productivity gains | Data not yet available |
| Software investment contribution | Up to 50% of productivity growth increase | Analysis not replicated |
| Business AI adoption rate | Below 20% (end of 2025) | Comparable or lower |
The Federal Reserve Bank of St Louis analysis represents a methodological improvement over simple adoption metrics. Economists asked workers to estimate time saved using AI tools, then compared this against productivity growth relative to 2015-2019 trends. Industries where workers saved the most time—information services, professional and technical services—showed the strongest productivity acceleration.
What the research actually reveals
Jonathan Haskel’s research team identified 2017 as the inflection point, when the transformer architecture paper introduced the foundation for modern generative AI. Comparing 2017-2024 against 2012-2017, they estimated software investment contributed as much as half of the productivity growth increase.
Critical Context: The Goldman Sachs figure of 32% average productivity boost from AI comes from corporate anecdotes—not systematic measurement. Treat it with the same scepticism you would apply to playground boasting about prodigy children.
The St Louis Fed analysis shows correlation strengthening through Q3 2025, suggesting the AI-productivity relationship is becoming more robust as adoption matures. But correlation still is not causation: more innovative industries may simply be both more likely to adopt AI and more productive for unrelated reasons.
What makes the evidence compelling is the convergence across multiple approaches:
- Industry-level adoption correlations
- Individual time-savings linked to sectoral productivity
- Software investment contribution analysis
Each method has limitations. Together, they point in the same direction.
The UK productivity puzzle
When Tera Allas, senior McKinsey adviser, examined British data, she found no evidence that AI-adopting industries were experiencing unusually high productivity growth. This gap matters.
Competitive Reality: If US productivity acceleration is partly AI-driven whilst UK gains are not, British businesses face a widening competitiveness gap that will compound over time.
Several explanations exist for the UK-US divergence:
Adoption timing: UK businesses may lag US counterparts by 6-18 months, meaning effects have not yet appeared in data. This would be the optimistic interpretation.
Implementation quality: Adoption metrics do not capture whether AI is being used effectively. UK businesses may be experimenting without integration into core workflows.
Measurement gaps: The UK lacks equivalent granular data linking individual time savings to sectoral productivity, making the relationship harder to detect.
Structural differences: UK industrial composition may mean fewer high-impact AI applications in dominant sectors.
For UK managing directors, the uncertainty itself demands action. Waiting for clearer UK data means falling further behind competitors who are learning by doing.
Strategic implications for business leaders
The research carries several actionable implications for UK businesses.
Success Factor: Industries seeing the strongest AI-productivity correlation—information services and professional/scientific/technical services—share a common characteristic: knowledge work with high language and analysis components. These are precisely the tasks where generative AI excels.
1. Time savings as leading indicator
The St Louis Fed methodology suggests individual time savings precede measurable productivity gains. Tracking where your team saves time using AI provides an early signal of where broader productivity effects may emerge.
2. Integration depth matters more than adoption
With US business adoption still below 20%, the productivity effects are coming from relatively intensive users. Casual experimentation does not show up in the data. Systematic integration into workflows does.
3. Sector-specific opportunities
If your business operates in information services, professional services, or technical consulting, you are in sectors where AI-productivity effects are already measurable. The window for competitive advantage through early effective adoption may be narrowing.
| Business maturity level | Recommended priority |
|---|---|
| No AI strategy | Commission rapid use-case assessment to identify 3-5 opportunities |
| Experimenting informally | Establish governance and integration frameworks to move from trials to systematic use |
| Integrated in some workflows | Expand to adjacent processes, track time savings systematically |
| Mature AI operations | Benchmark against sector productivity data, identify remaining gaps |
Implementation Note: Self-reported time savings correlate with productivity gains when redeployed to higher-value work. Time saved and spent on “perfecting passive-aggressive emails” (as Keynes notes) does not move the needle.
Hidden challenges in the productivity thesis
Four non-obvious risks warrant attention:
Measurement lag creates strategy risk
Productivity data takes 12-18 months to become reliable. Businesses making investment decisions based on current data are looking in the rear-view mirror whilst driving forward. The choice is between acting on incomplete evidence or waiting whilst competitors build capability.
Warning: ⚠️ Waiting for “definitive proof” of AI productivity gains means accepting a 12-18 month information disadvantage against competitors willing to act on probabilistic evidence.
Quality effects may not appear in productivity metrics
AI might improve output quality—better analysis, fewer errors, more personalised customer interactions—without changing the quantity of output per hour. Traditional productivity measures would miss this entirely.
The second-order effects take longer
Initial time savings from AI often accrue to individuals. Organisational productivity gains require workflow redesign, role evolution, and management adaptation. The Fed data may be capturing early adopters who have progressed through this curve.
Adoption inequality compounds
If AI-productivity gains are real, the gap between adopting and non-adopting businesses will widen faster than historical technology transitions. Digital divides that took decades may emerge in years.
What success looks like
For UK business leaders, the emerging evidence supports a clear direction even amid uncertainty.
Strategic Insight: The question has shifted from “does AI improve productivity?” to “are we positioned to capture the gains that others are already measuring?”
Three success factors
- Track individual time savings systematically: Create lightweight mechanisms to understand where AI is saving time and whether that time converts to higher-value activities
- Move from experimentation to integration: Ensure AI tools connect to core business workflows rather than operating as standalone novelties
- Benchmark against sector-specific data: Monitor emerging productivity research in your specific industry to calibrate investment levels
Next steps checklist
- Audit current AI tool usage across the organisation
- Identify which roles report meaningful time savings
- Map time savings to business outcomes (not just activity completion)
- Compare your integration depth against sector benchmarks
- Establish governance to enable systematic rather than ad-hoc adoption
Source and attribution
This analysis draws on Soumaya Keynes’ reporting in the Financial Times, incorporating research from the Federal Reserve Bank of St Louis examining AI time savings and productivity correlations, and academic work by Jonathan Haskel and colleagues comparing software investment contributions across the 2012-2017 and 2017-2024 periods.
Original source: Financial Times - “Where is AI showing up in the productivity data?” by Soumaya Keynes
Need help identifying where AI can drive productivity in your organisation?