Using the Eurostat Labour Cost Index across 34 European economies, we examine how AI-driven productivity gains are diverging from wage growth — and which occupation groups are most affected.
Key Findings
Our analysis of the Eurostat LCI (Labour Cost Index, base year 2020 = 100) reveals significant divergence across European economies.
High-Growth Wage Economies (LCI > 140)
- Bulgaria: LCI 180.8 (2025) — 80.8% above 2020 baseline
- Romania: LCI 170.5
- Croatia: LCI 164.2
- Hungary: LCI 176.0
- Poland: LCI 165.0
These economies have experienced rapid nominal wage growth, driven largely by catch-up effects, EU structural funds, and tight labour markets in manual and service sectors. However, our risk model shows that the professional services layer in these economies faces among the highest risk scores in our dataset.
The paradox: aggregate wages are rising, but the specific occupations most exposed to AI are seeing wage growth decelerate relative to peers.
Moderate-Growth Economies (LCI 110-140)
- Germany: LCI 117.3
- France: LCI 112.7
- Netherlands: LCI 123.2
- Finland: LCI 119.5
These are the mature Western European economies where AI adoption is most advanced but institutional buffers slow the wage compression effect. Our data shows that employment change in these economies is lagging the risk signals.
Low-Growth Economies (LCI < 110)
- Italy: LCI 105.5
- Spain: LCI 113.7
Italy stands out as an outlier with the lowest LCI growth in the EU. Italy may be less exposed to the AI displacement spiral precisely because its wages were already low enough that the automation cost-benefit calculus is less compelling.
Occupation-Level Patterns
Highest risk scores (avg > 65): Clerical support workers, Business and administration associate professionals, General and keyboard clerks, Numerical and material recording clerks
Lowest risk scores (avg < 50): Protective services workers, Personal care workers, Building and related trades workers, Food preparation assistants
The dividing line is clear: occupations involving routine cognitive tasks are at highest risk. Occupations involving physical presence, manual dexterity, or unpredictable human interaction are at lowest risk.
The Demand Signal
Our vacancy data from Eurostat JVS covers 29 European countries. The most striking finding: total vacancy counts have remained roughly stable, but the composition of vacancies is shifting.
- Postings for "AI/ML Engineer", "Data Scientist", and "Automation Specialist" have increased 40-60% YoY in DE, FR, NL
- Postings for "Administrative Assistant", "Data Entry", and "Junior Analyst" have declined 15-25% in the same period
- The net vacancy count is approximately flat — masking a fundamental compositional shift
This is the "ghost demand" phenomenon. The aggregate statistics look healthy. The occupation-level data tells a different story.
Implications
For policymakers: wage compression in the professional services layer is a leading indicator of consumer demand weakness.
For workers: the data supports retraining toward roles that combine cognitive skill with physical presence or high-trust human interaction.
For investors: the European labour market is pricing in AI displacement slowly due to institutional buffers. This creates a lag between when the risk becomes measurable and when it shows up in corporate earnings.
Data: Eurostat LFSA_EGAI2D (employment by ISCO-08), Eurostat lc_lci_r2_q (Labour Cost Index), Eurostat JVS_Q_NACE2 (Job Vacancy Statistics), Adzuna API (16 countries). All data publicly available. Updated monthly.