Why this gap matters now
Generative AI moved from novelty to everyday tool within two years. Official statistics capture the spread: in the EU, the share of enterprises using AI rose from 13.48% in 2024 to 19.95% in 2025 (Eurostat, 2025), and an OECD survey found generative AI in use in 31% of SMEs across seven countries (OECD, 2025). Governments and business bodies now talk about generative AI as a way for small firms to close productivity gaps with large ones.
The research base for that claim is narrow. The best-known causal estimates come from an online experiment with professionals (Noy & Zhang, 2023) and from agents in one large customer-support operation (Brynjolfsson et al., 2025). National register data, meanwhile, show no measurable change in earnings or hours so far (Humlum & Vestergaard, 2025). None of these isolates the small firm, where there is rarely an IT department, a data team or a formal roll-out.
Industry is measuring adoption; research has not yet measured effects
The imbalance is visible in the type of evidence available. Statistical offices and international organisations publish large, comparable surveys every year. Peer-reviewed reviews of AI in SMEs, by contrast, describe barriers and applications: one review of 71 papers found most work centred on operations, with knowledge and cost the main obstacles (Oldemeyer et al., 2024), and another of 50 studies listed long-term impacts and workforce issues among the areas still under-researched (Tran et al., 2025).
| Source type | What it measures | Limitation for SMEs |
|---|---|---|
| Official surveys | Adoption rates, reported benefits | Self-reported; no outcome data |
| Experiments | Task speed and quality | Individuals, not firms |
| Large-firm field data | Worker productivity | One firm; scale and IT support |
| Register studies | Earnings and hours | Early period; not SME-specific |
Is this feasible for a doctorate?
Yes, if the project starts with a data partner rather than a data wish-list. A focused thesis might follow 100 to 300 SMEs in one region or sector, combining firm records with a short survey and a set of interviews. The method is well established in management and economics; the novelty lies in the population and the outcome measures.
- Record which tools firms use and when, because tools change quickly.
- Separate informal use by staff from planned adoption by the firm.
- Report null or small effects with the same care as large ones.
A useful test of scope is whether the project could still deliver if adoption turns out to be shallow. Designs that measure who uses which tools for which tasks will produce findings either way, and a well-measured small or null effect is itself a contribution, given how much current policy rests on reported rather than measured benefits (OECD, 2025; Humlum & Vestergaard, 2025).
