Workers subject to algorithmic management report higher stress and worse wellbeing across the EU, with the gap markedly larger in Slovenia; unexpectedly, exposure is not associated with reduced autonomy.

Main Finding

Workers exposed to algorithmic management in the EU report significantly higher levels of work-related stress and lower wellbeing than non-exposed workers; these negative differences are more pronounced in Slovenia than the EU average. Contrary to much prior empirical work, the thesis does not find a loss of worker autonomy associated with algorithmic management.

Key Points

• Scope and data: original analysis of Eurofound’s European Working Conditions Survey (EWCS) 2024 microdata — 36,644 respondents across 35 countries — with particular attention to Slovenia.
• Exposure mapping: the thesis maps prevalence and composition of algorithmic management exposure across countries, sectors, and worker characteristics.
• Core job-quality outcomes examined: work-related stress, working-time sufficiency, job autonomy (autonomy index), and subjective wellbeing (WHO‑5).
• Main empirical contrasts:
  • Exposed vs non-exposed workers: higher stress and lower WHO‑5 wellbeing among the exposed group EU‑wide and in Slovenia.
  • Autonomy: no measurable reduction in autonomy for exposed workers (a result that departs from much qualitative evidence).
  • Heterogeneity: differences vary by sector, by EU‑15 vs newer member states, and by worker characteristics — with Slovenia showing larger negative gaps on stress and wellbeing than the EU average.
• Illustrative case: Uber is used as a concrete example to connect algorithmic management practices to the job‑quality findings and to contextualize regulatory relevance.
• Policy context: findings are interpreted in light of recent EU regulation (EU AI Act and the Platform Work Directive, both enacted in 2024) and ongoing debates over transparency, employment status, and worker protections.

Data & Methods

• Data source: Eurofound EWCS 2024 microdata; sample size 36,644 respondents from 35 countries. National focus: Slovenia contrasted with EU aggregates (EU‑27 / EU‑15 comparisons are also reported in the thesis tables).
• Key variables:
  • Exposure to algorithmic management: constructed from EWCS items identifying use of algorithmic practices/controls (the thesis reports exposure by individual indicators and combinations).
  • Job-quality indicators: work-related stress, WHO‑5 wellbeing index, working-time sufficiency, and an autonomy index derived from EWCS items.
• Analytical approach:
  • Descriptive and comparative analysis mapping (1) prevalence of algorithmic management, (2) how algorithmic practices are implemented, and (3) differences in job-quality indicators by exposure status.
  • Comparisons are carried out EU‑wide, by member state (with special attention to Slovenia), across sectors, and by worker subgroups (age, education, gender, etc.). A case study (Uber) illustrates mechanisms and regulatory relevance.
• Limitations and assumptions (summarized from thesis discussion):
  • Cross-sectional survey data: limits causal inference; results are associations rather than proven causal effects.
  • Measurement limits: the EWCS exposure indicators are survey-based and may not fully distinguish AI‑driven vs rule‑based algorithmic systems or capture all dimensions of algorithmic control.
  • Potential selection and unobserved confounding: workers self-select into jobs/platforms; omitted variables may influence both exposure and job-quality outcomes.
  • Country- and sector-level institutional differences may mediate effects; heterogeneity complicates generalization.

Implications for AI Economics

• Welfare and productivity trade-offs: consistent associations of algorithmic management with higher stress and lower wellbeing imply potential negative worker welfare externalities (health costs, absenteeism, turnover) that must be weighed against any productivity gains from algorithmic coordination.
• Distributional concerns: heterogeneous impacts across countries, sectors, and worker groups indicate that algorithmic management may exacerbate inequalities (by age, education, region), affecting labor supply responses and bargaining power in some segments.
• Regulation and policy design: empirical evidence of elevated stress/worse wellbeing (and the mixed result on autonomy) supports EU policy attention (AI Act, Platform Work Directive). Key policy levers include transparency/contestability of algorithmic decisions, stronger enforcement of employment status rules, psychosocial risk protections, and measures to preserve effective worker voice.
• Measurement and modeling in macro/market analysis:
  • Researchers and policymakers should not treat “algorithmic management” as a homogeneous technology; models should distinguish AI vs non‑AI systems, types of managerial functions automated, and intensity of monitoring.
  • Macroeconomic or labor‑market models that evaluate the impact of AI on employment, wages, or productivity need to incorporate non‑pecuniary job‑quality effects (stress, wellbeing, autonomy) and heterogeneity across sectors and institutions.
• Research priorities for AI economics:
  • Causal evidence: panel data, natural experiments, firm-level administrative data, or randomized interventions to establish causal links between algorithmic management and outcomes (productivity, health, turnover, wages).
  • Employer-side and system-level data: to measure algorithm design, decision rules, and adaptive behavior of AI systems (distinguish rule‑based from learning systems).
  • Cost–benefit and distributional analyses: quantify productivity gains vs worker welfare losses, and the implications for aggregate labor supply and social welfare.
  • Policy evaluation: empirical assessment of the effects of the EU AI Act and Platform Work Directive on firms’ algorithmic practices and worker outcomes.
• Practical takeaway for economists: algorithmic management matters for both micro (firm/worker) and macro (labor market, welfare) outcomes. Empirical work should combine careful measurement of algorithmic exposure with designs that can address selection and causality, and policy analysis should incorporate worker wellbeing and distributional consequences alongside efficiency gains.