Research gapBusiness & management

Generative AI in small firms: what does it actually do to productivity and jobs?

Official surveys show generative AI spreading quickly through small and medium-sized enterprises (OECD, 2025; Eurostat, 2025), yet the causal evidence on productivity and work comes almost entirely from large firms and controlled experiments (Brynjolfsson et al., 2025; Noy & Zhang, 2023). What happens inside the small firm is still largely unmeasured.

PhD Research Lab InsightsEditorial team, reviewed by a subject specialist
Posted
5 October 2026
Last checked
5 October 2026
Reading time
9 min
Sources
9
Research proposal support
The gap

We do not yet know how adopting generative AI changes measured productivity, task content and employment inside small and medium-sized enterprises, as opposed to what owners report in surveys. Official surveys record adoption and perceived benefits (OECD, 2025), while the rigorous causal studies examine single large firms, online experiments or national registers (Brynjolfsson et al., 2025; Noy & Zhang, 2023; Humlum & Vestergaard, 2025). Peer-reviewed reviews of AI in SMEs focus on adoption barriers and applications rather than workforce outcomes (Oldemeyer et al., 2024; Tran et al., 2025).

Evidence gap (self-report versus measured outcomes)Population gap (small and micro firms)Methodological gapContextual gap (sectors and regions)

What we know

  • Generative AI was in use in 31% of SMEs in a representative 2024 OECD survey of over 5,000 SMEs in Austria, Canada, Germany, Ireland, Japan, Korea and the United Kingdom; firms reported better performance and help with skill gaps and labour shortages (OECD, 2025).
  • In the EU, 19.95% of enterprises with 10 or more employees used AI in 2025, against 17% of small and 55.03% of large enterprises (Eurostat, 2025).
  • Among 5,172 customer-support agents, access to an AI assistant raised issues resolved per hour by 15% on average, with the largest gains for less experienced workers (Brynjolfsson et al., 2025).
  • In an experiment with 453 college-educated professionals, ChatGPT cut time on writing tasks by 40% and raised output quality by 18% (Noy & Zhang, 2023).
  • Danish register data show precise null effects on earnings and recorded hours two years after ChatGPT's launch, ruling out effects larger than 2%, while employers reorganised tasks (Humlum & Vestergaard, 2025).
  • A review of 71 peer-reviewed papers on AI in SMEs found lack of knowledge, cost and low organisational maturity to be the main barriers, with 44 papers centred on operations (Oldemeyer et al., 2024).

What we do not know yet

  • Whether self-reported performance gains in SMEs show up in objective measures such as sales per employee, output per hour or error rates.
  • Whether productivity gains found among workers in large firms carry over to small firms with no IT staff, thin data and owner-led management.
  • How generative AI changes hiring, job design and skill needs in micro and small firms over two or more years.
  • Which complementary investments (training, process redesign, data) explain why some SMEs gain and others do not.
  • How effects differ across sectors, countries and firms outside the high-income economies most studied so far.

Research questions a PhD could ask

  1. Does adoption of generative AI tools change labour productivity in SMEs when measured with firm accounts rather than owner perceptions, and how large is the effect?
  2. How does generative AI alter the task mix of SME employees, and which roles gain or lose work?
  3. Do the largest gains in SMEs accrue to less experienced workers, as reported in large-firm field data?
  4. Which managerial practices and skills investments moderate the productivity effects of generative AI in small firms?
  5. How do SME owners decide where to deploy generative AI, and what do they do with any time saved?
Methods that could close the gap
ApproachWhy it fits
Matched firm panel with difference-in-differencesCompares adopting and non-adopting SMEs over time using accounts or administrative data, in the spirit of the register design used by Humlum and Vestergaard (2025).
Randomised encouragement or training trialRandomly offers SMEs training or tool access to estimate causal effects, extending experimental logic (Noy & Zhang, 2023) to whole firms.
Within-firm field studyUses task logs or output records in a small number of SMEs, adapting the worker-level design of Brynjolfsson et al. (2025) to settings without large IT teams.
Linked survey and outcome dataPairs adoption surveys with objective firm outcomes to test whether reported benefits (OECD, 2025) match measured ones.
Comparative case studiesExplains mechanisms, such as how owners redesign jobs and skills, that panel data alone cannot reveal (Tran et al., 2025).
Gap typeIndustry problem, research still limited

The problem

SMEs are adopting generative AI with little evidence about what it does to their productivity, jobs and skills needs; what is known comes largely from surveys of perceived benefits (OECD, 2025).

Why it happens

Firm-level data on small businesses are scarce and costly to collect, adoption is often informal (staff using free tools), and most rigorous studies are run where large data sets already exist, in big firms or national registers (Brynjolfsson et al., 2025; Humlum & Vestergaard, 2025). Reviews of SME research show attention concentrated on adoption barriers and operations (Oldemeyer et al., 2024).

Where it is felt most

The adoption gap is widest for small firms: in the EU, AI use rises from 17% of small enterprises to 55.03% of large ones (Eurostat, 2025). Exposure is also uneven across countries, with a larger share of jobs exposed in high-income economies (Gmyrek et al., 2025).

How big is the problem

  • 31%Share of SMEs using generative AI, OECD survey of over 5,000 SMEs in seven countries, fielded in 2024 (OECD, 2025)
  • 19.95%EU enterprises with 10 or more employees using AI in 2025, up from 13.48% in 2024, Eurostat (Eurostat, 2025)
  • 17% vs 55.03%AI use among small versus large EU enterprises, 2025, Eurostat (Eurostat, 2025)
  • 1 in 4Workers worldwide in an occupation with some generative AI exposure, ILO Working Paper 140, 2025 (Gmyrek et al., 2025)
What studies say, and at what level of evidence
LevelWhat it shows
Systematic reviews and meta-analysesReviews of 71 and 50 studies map AI applications and adoption barriers in SMEs; long-term impacts and workforce challenges are flagged as under-researched (Oldemeyer et al., 2024; Tran et al., 2025).
Trials or field studiesStrong causal evidence of worker-level gains, but from an online experiment and a single large customer-support operation rather than SMEs (Noy & Zhang, 2023; Brynjolfsson et al., 2025).
Observational studiesRegister data from Denmark find no measurable change in earnings or hours so far, alongside task reorganisation; this is a working paper (Humlum & Vestergaard, 2025).
Industry and policy reportsOfficial surveys track adoption and self-reported benefits for SMEs at scale (OECD, 2025; Eurostat, 2025), and the ILO estimates exposure by occupation (Gmyrek et al., 2025).
What is missingPeer-reviewed, SME-specific studies that link adoption to measured productivity and employment outcomes over time, especially for micro firms and outside high-income economies.

Who should take this on

Scholar profile

  • A management or economics researcher with access to a regional SME network, chamber of commerce or accountancy partner who can share anonymised firm data.
  • A practitioner-researcher (for example a consultant or business adviser) who works directly with small firms adopting AI tools.
  • A labour economist able to link survey data to administrative or tax records.

Subjects and PhD specialisations

  • Strategic management
  • Entrepreneurship and small business
  • Human resource management
  • Labour economics
  • Information systems
  • Operations management

Countries where the problem is greatest

  • EU member states: AI use among small enterprises lags far behind large ones (Eurostat, 2025).
  • Austria, Canada, Germany, Ireland, Japan, Korea and the UK: the OECD survey offers a baseline that new outcome data could build on (OECD, 2025).
  • Denmark and other countries with linked registers: designs that track earnings and hours are feasible (Humlum & Vestergaard, 2025).
  • High-income economies generally, where a larger share of employment is exposed to generative AI (Gmyrek et al., 2025).

Scope for future research

  1. Firm-level panels that follow SMEs for two to three years after adoption.
  2. Sector studies in professional services, retail and light manufacturing, where SMEs dominate.
  3. Studies of micro firms (fewer than 10 employees), which official AI statistics often exclude (Eurostat, 2025).
  4. Comparative work in middle-income economies, where little SME evidence exists.
  5. Research on skills, training and job redesign as complements to adoption.

How a doctoral project could address it

  1. Define the outcome

    Choose measurable outcomes (output per employee, turnaround time, hiring) before choosing the method, and justify them against the existing literature (Oldemeyer et al., 2024).

  2. Secure a data partner

    Work with an SME association, bank, accountancy firm or software provider that can give anonymised firm-level data with consent.

  3. Build a comparison group

    Match adopting and non-adopting firms on size, sector and prior performance, or design a randomised training offer.

  4. Add a qualitative strand

    Interview owners and staff to explain how tasks and roles changed.

  5. Plan ethics and data protection

    Obtain approval for employee data, protect commercial confidentiality and agree publication terms with partners early.

  6. Plan the outputs

    Aim for three linked papers: a measurement study, a causal estimate and a mechanisms study.

Why it is worth solving

Small firms employ a large share of the workforce in most economies, and policy makers are already funding AI adoption programmes for them (OECD, 2025). Without measured evidence, those programmes rest on owner perceptions. Knowing whether and how generative AI raises SME productivity would help firms invest wisely, help governments target support, and tell workers which skills to build.

For science, SMEs are a strong test of whether gains found in large organisations (Brynjolfsson et al., 2025) are general or depend on scale, data and management capacity.

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).

Where current evidence on generative AI and work comes from
Source typeWhat it measuresLimitation for SMEs
Official surveysAdoption rates, reported benefitsSelf-reported; no outcome data
ExperimentsTask speed and qualityIndividuals, not firms
Large-firm field dataWorker productivityOne firm; scale and IT support
Register studiesEarnings and hoursEarly 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).

What to do now

  • Map recent studies by setting (lab, large firm, SME, register) and by outcome (reported or measured).
  • Approach an SME association or accountancy partner about anonymised data before fixing your questions.
  • Type your gap explicitly using an established typology (Müller-Bloch & Kranz, 2015).
  • Draft one measurable outcome and one mechanism question as the core of your proposal.
Your next step

Could this gap become your thesis?

  1. 1Read the evidence and the open questions
  2. 2Talk to a mentor about fit, data and method
  3. 3Build a defensible proposal
See proposal support

Questions scholars ask

Has the productivity question not already been answered?

For individual workers in some settings, partly (Noy & Zhang, 2023; Brynjolfsson et al., 2025). For SMEs as firms, measured with objective outcomes over time, the evidence is still thin (Oldemeyer et al., 2024; Tran et al., 2025).

Do I need administrative data?

Not necessarily. Firm accounts shared by a partner, platform usage logs or a carefully designed panel survey with outcome measures can all work if the comparison group is credible.

Will my findings date quickly?

Specific tools will change, so frame the contribution around mechanisms such as task redesign and skills, and record tool versions and dates.

Can PhD Research Lab write the study for me?

No. We mentor scholars on gap framing, design and methods; the research and writing remain your own.

Sources

  1. Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942. https://doi.org/10.1093/qje/qjae044
  2. Eurostat. (2025). Use of artificial intelligence in enterprises [Statistics Explained; data extracted December 2025]. European Commission. https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises
  3. Gmyrek, P., Berg, J., Kamiński, K., Konopczyński, F., Ładna, A., Nafradi, B., Rosłaniec, K., & Troszyński, M. (2025). Generative AI and jobs: A refined global index of occupational exposure (ILO Working Paper 140). International Labour Organization. https://doi.org/10.54394/HETP0387
  4. Humlum, A., & Vestergaard, E. (2025). Still waters, rapid currents: Early labor market transformation under generative AI (NBER Working Paper No. 33777). National Bureau of Economic Research. https://ideas.repec.org/p/nbr/nberwo/33777.html
  5. Müller-Bloch, C., & Kranz, J. (2015). A framework for rigorously identifying research gaps in qualitative literature reviews. In Proceedings of the Thirty Sixth International Conference on Information Systems (ICIS 2015), Fort Worth, TX, United States. https://aisel.aisnet.org/icis2015/proceedings/ResearchMethods/2/
  6. Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654), 187–192. https://doi.org/10.1126/science.adh2586
  7. OECD. (2025). Generative AI and the SME workforce: New survey evidence. OECD Publishing. https://www.oecd.org/en/publications/generative-ai-and-the-sme-workforce_2d08b99d-en/full-report.html
  8. Oldemeyer, L., Jede, A., & Teuteberg, F. (2024). Investigation of artificial intelligence in SMEs: A systematic review of the state of the art and the main implementation challenges. Management Review Quarterly, 75(2), 1185–1227. https://doi.org/10.1007/s11301-024-00405-4
  9. Tran, G. T. C., Le Dinh, T., & Vu, M.-C. (2025). Artificial intelligence in SMEs: Enhancing business functions through technologies and applications. Information, 16(5), 415. https://doi.org/10.3390/info16050415

How to cite this page

Choose your referencing style, then copy.

PhD Research Lab Insights. (2026, October 5). Generative AI in small firms: what does it actually do to productivity and jobs?. PhD Research Lab. https://www.phdassistance.com/insights/r-and-d/research-gap/generative-ai-sme-productivity-workforce/

PhD Research Lab Insights. “Generative AI in Small Firms: What Does It Actually Do to Productivity and Jobs?.” PhD Research Lab, 5 Oct. 2026, www.phdassistance.com/insights/r-and-d/research-gap/generative-ai-sme-productivity-workforce. Accessed [access date].

PhD Research Lab Insights (2026) Generative AI in small firms: what does it actually do to productivity and jobs?. PhD Research Lab. Available at: https://www.phdassistance.com/insights/r-and-d/research-gap/generative-ai-sme-productivity-workforce/ (Accessed: [access date]).

PhD Research Lab Insights. “Generative AI in Small Firms: What Does It Actually Do to Productivity and Jobs?.” PhD Research Lab. October 5, 2026. https://www.phdassistance.com/insights/r-and-d/research-gap/generative-ai-sme-productivity-workforce.

PhD Research Lab Insights, ‘Generative AI in Small Firms: What Does It Actually Do to Productivity and Jobs?’ (PhD Research Lab, 5 October 2026) <https://www.phdassistance.com/insights/r-and-d/research-gap/generative-ai-sme-productivity-workforce> accessed [access date].

PhD Research Lab Insights. Generative AI in small firms: what does it actually do to productivity and jobs? [Internet]. PhD Research Lab; 2026 Oct 5 [cited [access date]]. Available from: https://www.phdassistance.com/insights/r-and-d/research-gap/generative-ai-sme-productivity-workforce/

PhD Research Lab Insights, “Generative AI in small firms: what does it actually do to productivity and jobs?,” PhD Research Lab, Oct. 5, 2026. [Online]. Available: https://www.phdassistance.com/insights/r-and-d/research-gap/generative-ai-sme-productivity-workforce. [Accessed: [access date]].

@online{generative_ai_sme_productivity_workforce_2026,
  author = {{PhD Research Lab Insights}},
  title = {{Generative AI in small firms: what does it actually do to productivity and jobs?}},
  organization = {PhD Research Lab},
  date = {2026-10-05},
  url = {https://www.phdassistance.com/insights/r-and-d/research-gap/generative-ai-sme-productivity-workforce},
  urldate = {[access date]}
}

  • Generative AI
  • SMEs
  • Productivity
  • Future of work
  • Business & management
Generative AI in small firms: what does it actually do to productivity and jobs?Research gap · Business & management
Proposal support