AI adoption is reshaping EU labour markets, hitting routine office and administrative roles hardest; the Netherlands, France, Portugal, Italy and Malta appear best positioned to translate AI diffusion into favourable labour‑market transformation.

Artificial intelligence as a driver of economic growth: Challenges and prospects for the labor market

Svitlana Korobka, Nataliia Tilikina, T. Gorokhova, M. Kravchenko, M. Lukash · Fetched March 23, 2026 · International Journal of Data and Network Science

semantic_scholar correlational medium evidence 7/10 relevance DOI Source PDF

Using an input-oriented envelope model on EU countries, the paper finds that AI diffusion is associated with substantial restructuring of employment—most prominently reducing roles with routine analytical and administrative tasks—and identifies several countries (Netherlands, France, Portugal, Italy, Malta) as relatively well-positioned to convert AI diffusion into positive labor-market outcomes.

The article highlights the consequences and transformational changes in labor markets and labor resources due to the spread of artificial intelligence. The purpose of the article is to empirically study the processes of interaction between artificial intelligence and national labor markets, as well as to develop methodological tools for quantifying the impact of AI technologies on labor market parameters. The research methodology is based on the envelope model (“input” orientation) to assess the level of transformation of labor resources and labor markets as a result of the spread of artificial intelligence for a sample of European Union countries. The study found a significant transformation of the employment structure under the influence of artificial intelligence, where the most vulnerable occupational groups are office workers, data entry operators, call center workers, accountants, and administrative staff with routine analytical and administrative tasks. The study identified countries that can optimally transform AI diffusion (the Netherlands, France, Portugal, Italy, and Malta) for the domestic labor market into results when the trend of economic development and the realization of human capital potential is formed. The obtained results confirmed the existence of reserves for optimizing the interaction of artificial intelligence with the labor market, with an emphasis on the need to adapt AI to the specifics of economic models of national economies.

Summary

Main Finding

The paper empirically assesses how AI diffusion transforms national labor markets (27 EU countries) using an input-oriented DEA-CCR model. It finds substantial restructuring of employment: routine analytical and administrative occupations (office staff, data-entry, call-centre workers, accountants, certain managers) are most vulnerable, while demand for IT/digital skills and flexible work rises. A subset of countries (Netherlands, France, Portugal, Italy, Malta) appear relatively efficient at converting AI diffusion into favorable labor-market outcomes. The authors conclude there are untapped reserves for optimizing human‑AI interaction but emphasize the need to adapt AI adoption to national economic models and active policy responses (retraining, social protection, institutional design).

Key Points

Occupations most vulnerable to AI-driven automation: routine administrative and analytical roles (office administrators, data entry, call-centre operators, accountants, some managerial/admin staff).
Complementarity vs. displacement: Higher-qualified workers more likely to complement AI (gain productivity), whereas middle/low-skilled workers face greater displacement risk.
Structural shifts (reported comparisons 2019 → 2024):
- Manual-labor share: -20 percentage points
- IT specialists: +35 percentage points
- Freelance/remote work: +40 percentage points
Sectoral impact: manufacturing, retail, and transportation are singled out as most transformed (manufacturing has the highest automation/robotization effects).
Robotization snapshot: global industry average ~141 robots/10,000 production workers (end 2024); South Korea >>1,000/10k; Germany/Italy ~400–500/10k.
Policy-relevant observation: countries differ in their ability to translate AI adoption into productivity and employment stability; some EU countries are identified as “optimal transformers.”
Social and institutional concerns: rising inequality (skilled gains > unskilled losses), gig economy growth reducing traditional protections, algorithmic fairness risks in hiring/evaluation.

Data & Methods

Sample: 27 European Union countries (treated as decision-making units in DEA).
Method: Data Envelopment Analysis (DEA), Charnes‑Cooper‑Rhodes (CCR) specification, input‑oriented (estimates efficiency score θ ∈ (0,1], where θ=1 is on the frontier).
Inputs (x):
- x1: developers per 1,000 working‑age people
- x2: share of working‑age population with higher education (%)
- x3: ICT sub‑index rank (ordinal; lower = better)
Outputs (y):
- y1: readiness for frontline technologies (0–1)
- y2: projected productivity increase from AI (%)
- y3: share of jobs with potential for AI expansion (%)
- y4: employment stability (transformation of an unemployment‑impact indicator to a 0–4 scale)
Efficiency interpretation: θ* < 1 implies proportional needed reduction in inputs to reach the observed outputs; λ vectors identify peer/benchmark countries.
Data sources cited by authors include European Commission, UNCTAD, PwC, World Economic Forum and related reports (years primarily 2024–2025).
Important model caveats noted by authors: semantic choice of inputs/outputs is critical; DEA is scale-invariant but non-causal and relies on observed comparators.

Implications for AI Economics

Policy design:
- Prioritize targeted reskilling/upskilling (digital, analytical, creative and social skills) to increase complementarity and reduce displacement.
- Reform education and lifelong learning systems to align with rapid AI-driven job transformation.
- Update labor regulations and social protection to address gig work and platform-mediated employment.
- Monitor and mitigate algorithmic bias in hiring/performance evaluation.
Economic strategy:
- Countries should adapt AI adoption to their national economic model (industrial composition, human‑capital profile) rather than copy one-size-fits-all approaches.
- Invest in sectors where AI complements labor to maximize productivity gains while cushioning transitions in highly automatable sectors (e.g., manufacturing → robotics/automation management roles).
- Use frontier-country benchmarks (Netherlands, France, Portugal, Italy, Malta per study) to design peer-informed policies, but analyze context-specific drivers behind their performance.
Research directions for AI economics:
- Move beyond cross-sectional, frontier-efficiency measures to causal, dynamic analyses (panel methods, natural experiments, IVs) that quantify long-run employment, wage, and distributional effects.
- Incorporate occupation-level and firm-level microdata to capture heterogeneity in task-level automation risk and complementarities.
- Model interactions between AI adoption, institutional features (labor laws, education systems), and macro outcomes (productivity, inequality).
- Assess distributional and welfare impacts of AI adoption, including fiscal and social‑insurance responses required to manage transitions.
Limits to current inference: DEA indicates relative efficiency in converting inputs to desirable outputs but does not establish causal pathways, timing of effects, or within‑country distributional outcomes—so policy prescriptions should be tested with complementary methods and richer data.

Assessment

Paper Typecorrelational Evidence Strengthmedium — The paper provides systematic, quantitative cross-country evidence of associations between AI diffusion and changes in employment structure using an established efficiency modelling approach; however, it lacks exogenous variation or clear causal identification, relies on aggregate country-level measures and model assumptions, and likely faces omitted variable and measurement limitations that reduce confidence in causal interpretation. Methods Rigormedium — The envelope/input-oriented approach is a defensible method for comparing transformation/efficiency across units and can highlight relative performance and vulnerabilities, but the rigor is limited by (a) no plausibly exogenous shock or identification strategy to isolate causal effects, (b) potential sensitivity to variable selection and model specification, (c) aggregation at the national level and likely small sample (EU countries), and (d) limited information on robustness checks, timing, or alternative specifications. SampleCountry-level sample covering member states of the European Union (specific years and data sources not provided in the summary); uses national-level indicators for AI diffusion and labour-market/occupational employment structure with focus on occupational groups (office workers, data entry operators, call-centre workers, accountants, administrative staff) and country-level transformation/efficiency scores. Themeslabor_markets adoption IdentificationUses an envelope model (input-oriented efficiency analysis, similar to DEA) to compare levels of labor-resource transformation across European Union countries as AI diffuses; identification is comparative/correlational (efficiency scores and associations) rather than based on exogenous variation or quasi-experimental contrasts, so causal claims are not separately identified. GeneralizabilityLimited to European Union countries—results may not hold for low- and middle-income countries or non-EU institutional contexts, National-aggregate analysis masks within-country heterogeneity (regional, sector, firm-size, and worker-level differences), Findings depend on how 'AI diffusion' and occupational vulnerability are measured—different metrics could change conclusions, Cross-sectional/aggregate design cannot capture short-run dynamics or causal pathways (retraining, job creation, redeployment), Sample and time period unspecified—may not generalize to later phases of AI adoption

Claims (5)

Claim	Direction	Confidence	Outcome	Details
The study found a significant transformation of the employment structure under the influence of artificial intelligence. Employment	mixed	high	transformation of employment structure	0.3
The most vulnerable occupational groups to AI-driven transformation are office workers, data entry operators, call center workers, accountants, and administrative staff with routine analytical and administrative tasks. Job Displacement	negative	high	vulnerability / exposure to AI-driven job displacement	0.3
Certain countries can optimally transform AI diffusion into positive domestic labor-market outcomes (economic development and realization of human capital potential): the Netherlands, France, Portugal, Italy, and Malta. Adoption Rate	positive	high	capacity to translate AI diffusion into economic development and human capital realization	0.3
There exist reserves for optimizing the interaction of artificial intelligence with the labor market, and it is necessary to adapt AI to the specifics of national economic models. Governance And Regulation	positive	high	potential to optimize AI–labor-market interaction / need for policy adaptation	0.3
The research methodology is based on the envelope model ("input" orientation) to assess the level of transformation of labor resources and labor markets due to the spread of artificial intelligence. Other	null_result	high	method of measurement / assessment approach	0.15