Women across Europe are substantially underrepresented in advanced digital tasks — about a 15 percentage-point average gap that becomes a sharp 'digital glass ceiling' at the highest digital-intensity jobs. Observable characteristics explain only roughly 30% of the gap, implying workplace organisation, task allocation and promotion practices — not just STEM supply — must be addressed (Ireland shows an especially large male concentration in advanced digital roles).

Main Finding

Women are substantially underrepresented in performing advanced digital tasks (programming, AI/machine learning, complex IT systems) at work. Across Europe women are ~15 percentage points less likely than men to do advanced digital tasks; in Ireland the gap is much larger (~26 percentage points: ~44% of men vs ~18% of women). Observable factors (education, field of study, occupation, sector, age) explain only about 30% of the European gap on average; the majority remains unexplained, suggesting organisational, cultural or other unobserved mechanisms (a “digital glass ceiling”), particularly concentrated at the top tail of digital-intensity jobs and among younger cohorts.

Key Points

• Task-based approach: The report measures digital skill use by tasks rather than job titles, distinguishing basic (internet, word processing, spreadsheets) from advanced tasks (programming, AI/ML, IT systems management).
• Job Digital Intensity Index (JDII): A composite index capturing the overall digital intensity of jobs based on the range of digital tasks performed; gender gaps are largest at the upper quantiles of the JDII distribution.
• Magnitude:
  • Europe: women ≈ 15 percentage points less likely to perform advanced tasks.
  • Ireland: gap ≈ 26 percentage points (44% of men vs 18% of women do advanced tasks); Ireland’s large gap stems mainly from higher advanced-task participation by men, not lower participation by Irish women relative to European peers.
• Explanation decomposition:
  • Observable characteristics (education level, STEM field of study, occupation, sector, age, experience) account for ~30% of the average gap across Europe. In Ireland occupational sorting explains somewhat more than elsewhere but still leaves a large unexplained share.
  • Much of the remaining gap is concentrated among the most digitally intensive jobs — evidence consistent with a “digital glass ceiling” where women are underexposed to the highest-value digital tasks.
• Cohort pattern: Gaps are pronounced (and less explained by observables) among workers aged under 35, indicating the issue emerges early in careers rather than being solely a legacy effect.
• Policy implication emphasized by authors: Education alone is insufficient; interventions must target within-firm task allocation, progression, sponsorship, and organisational practices.

Data & Methods

• Data: Cedefop European Skills and Jobs Survey (ESJS), Wave 2 (2021). Harmonised, cross-country employee survey with detailed task, education and job characteristics.
• Key variables:
  • Binary indicators for performing intermediate/advanced digital tasks.
  • Job Digital Intensity Index (JDII) built from task items to summarize overall digital intensity.
• Empirical methods:
  • Regression analysis (probit and OLS) to estimate gender differences controlling for demographics, education, field of study, occupation, sector, country.
  • Blinder–Oaxaca decomposition to partition the gender gap into explained (observable characteristics) and unexplained components.
  • Unconditional quantile (RIF) decomposition to examine distributional heterogeneity and identify where gaps are largest (especially upper tail).
• Robustness: Cross-country comparisons, age-cohort stratification, and decomposition of JDII by occupational/sectoral categories.
• Limitations noted: cross-sectional self-reported data (limiting causal inference), lack of firm-level controls for task allocation and promotions, possible measurement error in task reporting, and missing variables capturing workplace culture, sponsorship or management practices.

Implications for AI Economics

• Human capital allocation and AI diffusion: Advanced digital tasks include AI/ML work. Underutilisation of women’s advanced digital skills reduces the effective supply of talent for AI adoption and development, potentially slowing firm-level and economy-wide productivity gains from AI.
• Innovation and bias risk: If AI development is concentrated in homogeneous (male-dominated) teams, models, systems and products may reflect narrower perspectives—raising risks of biased AI, weaker product-market fit, and missed innovations that require diverse inputs.
• Labor market returns and inequality: Access to advanced digital tasks is a stepping stone to higher wages, leadership and career progression. Persistent gaps in task exposure contribute to gender wage gaps and unequal gains from AI-driven productivity improvements.
• Policy levers for effective AI-enabled growth:
  • Within-firm reforms: transparent task allocation, rotation into advanced projects, sponsorship and promotion pipelines, gender-aware assignment to AI/ML and high-value technical work.
  • Measurement & monitoring: adopt task-based metrics (e.g., JDII) in firm and national monitoring to track equitable access to AI-related tasks and to evaluate interventions.
  • Skills policies beyond supply: combine education/STEM initiatives with interventions targeting organisational practices, apprenticeships, and employer incentives (procurement, funding criteria) that reward inclusive deployment of AI talent.
  • Early-career focus: interventions aimed at career entry and early progression (mentorship, project assignment, internships) are critical because gaps appear early and persist.
• Macro and policy research priorities:
  • Quantify productivity losses from underutilised advanced digital skills and model counterfactuals where task access is equalised—important for cost–benefit analysis of policy interventions.
  • Link task-based measures to wages, promotion, firm AI adoption and performance using matched employer–employee or panel data to establish causal pathways.
  • Evaluate policies (e.g., mandated reporting, gender-equity targets for project assignment, subsidised rotation/training for advanced tasks) with randomized or quasi-experimental methods to identify effective instruments for inclusive AI growth.

Caveats: findings are descriptive and decompositional from cross-sectional survey data; unexplained components signal the need for richer employer-level and longitudinal data to identify causal mechanisms (task allocation, bias in promotion/sponsorship, differential training access).