More than just Plug and Play: Early Evidence on Organisational Capital and AI Adoption (ESCoE DP 2026-13)
A · Strong evidence Diane Coyle and colleagues at ESCoE and Cambridge's Bennett School link ONS management-survey panels from 2020 to AI adoption in 2023, with matching, placebo technologies and robustness checks shown.
Summary based on the full text·Original: about 60 minutes to read
In short
ESCoE Discussion Paper 2026-13 (September 2026) by Diane Coyle, Nghi Nguyen, John Lourenze Poquiz and Rebecca Riley. UK firms with better structured management in 2020 were more likely to be using AI by 2023. The same management scores do not predict adoption of robotics, specialised equipment or specialised software, and predict cloud only weakly. Performance monitoring (KPIs and targets) carries most of the link, and multi-site firms that let local teams run product development adopt AI more often.
Key numbers
- 1,292UK firms in the matched 2020-2023 Management and Expectations Survey panel
- roughly 10 percentage pointshigher AI adoption in the top management-score decile vs the bottom decile
- 93.3%AI adopters that also use cloud computing
- 10.7%cloud computing adopters that also use AI
- £268,800 versus £193,5002020 labour productivity per worker, AI adopters vs non-adopters
As reported by the source.
English·Web page·opens d.repec.org
What it reports
- Baseline management score (2020) significantly predicts AI use in 2023 and plans to adopt in 2024, after controls for size, industry, age, region, productivity, ownership and intangible investment.
- The top management decile adopts AI at rates roughly 10 points above the bottom decile.
- The relationship is specific to AI: no link for robotics, specialised software or equipment; a smaller positive one for cloud.
- Of the four management components, KPI tracking and target-setting matter most; continuous improvement and employment practices show weaker or insignificant effects.
- Among multi-site firms, decentralised product development predicts AI adoption even with management controlled.
- Complementarities are asymmetric: 93.3% of AI users use cloud, but only 10.7% of cloud users use AI. AI adopters spent six times more on advertising and four times more on software and databases in 2021.
How they got it
Linear probability models on the ONS Management and Expectations Survey, accessed through the Secure Research Service, with management measured in 2020 and technology use in 2023 to set temporal order. The same specification runs for four other technologies as a placebo check. Robustness: 2023 cross-section, winsorising, leave-one-industry-out, propensity score and other matching, and a broader adoption definition including pilots.
Limits
- The authors do not claim causal effects; unobserved firm traits could still drive both management and AI use.
- Panel attrition favours larger, better-managed firms; the cross-sectional check helps but does not remove this.
- AI adoption is self-reported and measured in 2023, early in LLM diffusion.
- The decentralisation result applies only to multi-site firms.
Why it matters for companies
AI tools do not drop into weakly managed firms and work; the firms that adopt are the ones already tracking performance and setting targets. That points to measurement discipline as a prerequisite for getting value from AI, ahead of tooling. The decentralisation finding suggests use cases are found by teams close to the process, so purely central AI programmes may miss them. Cloud and software foundations are near-universal among adopters, but having them does not make firms adopt.
Summary based on the full text. Read the original for the full argument and data.Read the original on d.repec.org