The EU Needs Dedicated Capacity to Track AI’s Impact on Jobs – Social Europe

Warnings of widespread disruption to the economy, labour, and productivity driven by artificial intelligence (AI) are growing, even as immediate job displacement is not yet reflected in aggregate data. This gap between prediction and available evidence should compel the EU to build systematic evidence of AI’s impact on the labour market, preparing for drastic scenarios and for the “messy middle” of patchy transformation alike.

With 41 per cent of European workers using AI “somewhat” or “very” concerned about being displaced by AI at work, building this capacity will also be crucial for communicating with voters. The target should be systematic, harmonised monitoring: the foundation for policies that avoid both excessive hype and excessive caution. On the EU policy front, some measures are already beginning to appear.

The EU’s forthcoming Quality Jobs Act is expected to address, among other challenges, the governance of AI in the workplace. Workers’ organisations are already pressing for better training ecosystems and labour protections to counterbalance displacement and de-skilling, and for an extension of the Platform Work Directive’s algorithmic transparency requirements from the gig economy to other sectors.

In a recent stocktake of the European Pillar of Social Rights, the European Commission also committed to setting up a high-level expert group on AI’s impact on the labour market, tasked with developing scenarios of its “potential long-term trajectory” and identifying “enhanced analytical tools for monitoring, early warning, foresight and anticipation”.

But which indicators should such intelligence tools cover? As AI development and adoption accelerate, geographically or sectorally concentrated jumps in unemployment may carry very different economic and social consequences, calling for differentiated response strategies.

Framing the Measurement Challenge

To address this challenge, the European Policy Centre and the Open Markets Institute held a workshop in Brussels this June, bringing together economists, civil servants, and civil society. The goal was to rank labour market transformation metrics by priority, to inform a proposal for an EU-wide monitoring framework.

Part of the problem, as several participants observed and as a recent critical review paper discusses, is misrepresented figures. A given job’s estimated AI exposure, itself hard to define, is often treated as a proxy for displacement risk. Similarly, firms’ AI procurement spending and performance expectations are taken at face value rather than checked against real-world adoption rates and economic gains.

As the highest-priority indicators, participants identified changes in junior-level job vacancies; AI exposure and adoption across occupations; changes in firm structure attributable to AI; and the labour share of gross domestic product (GDP). They also pointed out that source reliability and attributional strength vary greatly even among these metrics, since data is often scattered across public and private actors, or simply inaccurate.

Another issue highlighted was the evidence on AI adoption, which is currently limited to self-reporting by individuals and businesses. More reliable, comprehensive usage-based data can only be gathered by AI providers, who benefit from assumptions about AI’s inevitability in the workplace rather than from transparency. Even while backpedalling on their “jobpocalypse” predictions, these firms actively shape the discourse, hiring top economists and publishing forecasts and policy prescriptions.

Counterbalancing this influence will require data-access obligations for researchers, allowing actual usage data to be cross-referenced with self-reports. This could mean taking a page out of the Digital Services Act, for example, though access would need to be continuous.

The macroeconomic data already available, as deployed in current monitoring frameworks, partially matches our participants’ recommendations. The Organisation for Economic Co-operation and Development (OECD), in particular, publishes nuanced reports on measuring task exposure to AI-driven displacement. Yet its proposed metrics do not yet validate how theoretical exposure affects employment and wages in practice.

The Yale Budget Lab’s monthly tracker, by contrast, effectively synthesises exposure estimates with real-life market developments, although it covers only the US market.

The European Centre for the Development of Vocational Training’s (Cedefop) Skills Intelligence dashboards offer a compelling alternative. These dashboards consolidate labour market data across the EU, including sectoral employment and online job vacancies, as well as automation displacement forecasts based on task and skill exposure. Exposure-based forecasts are not cross-referenced with actual employment shifts, however, and changes in firm structure remain outside Cedefop’s mandate. Moreover, the data is processed too slowly for AI’s rapid development, with some dashboards still referring to 2024 figures.

These delays highlight the challenge of processing data scattered across regional and sectoral monitoring institutions. The International Network of Observatories on AI and Work aims to address this problem. Launched at the Paris AI Action Summit, it coordinates EU, International Labour Organization (ILO), and OECD analysis with that of a dozen countries, to enable stronger collective monitoring and policy responses.

Rethinking the Transition Signals

Quantitative indicators of economic transition, however, may be insufficient to capture the full nuance of AI-driven workplace transformation, as emerged clearly in our session. Job quality and stability may suffer long before wage pressures manifest, as workers are more closely monitored or as work pace and pressure increase without productivity gains. It is thus essential to distinguish what kind of AI is used, with what intensity and regularity; to track workers’ AI literacy; and to establish clear correlations with changes in productivity and employment levels. Such analysis begins with qualitative research on the transforming workplace.

Nevertheless, there is under-utilised potential in deploying macroeconomic measures of technological change in the labour market. Tracking investment in automation technologies, for instance, reveals its connection to job displacement or growth when cross-referenced with changes in total labour expenditure.

Another such metric is job creation: where in the economy new positions appear, how quickly, and whether they can offset displacement. Surveys tracking aggregate worker movement between sectors, cross-referenced with economy-wide trends in employment, job titles, and task composition, would together help validate predictions from the technology-versus-labour-expenditure indicator.

While challenging to set up, a comprehensive public capacity for studying AI usage and its labour effects is crucial to effective EU policymaking. It would enable evidence-based planning and intervention, and help us steer the direction of technological change. Importantly, it would stand apart from AI industry-sponsored studies and their conflicts of interest.

This new body, an EU equivalent of the UK’s AI Economics Institute, belongs within the Commission’s Directorate-General for Employment, Social Affairs and Inclusion (DG EMPL), with its existing labour market policy mandate and expertise. It must also collaborate closely with Cedefop, the International Network of Observatories on AI and Work, and other established sources of relevant data, as well as with member states and the European AI Office. In particular, the AI Office’s technical expertise will be crucial for translating AI usage data into labour market insights.

Besides automation exposure and displacement, employment, and technology expenditure, the new capacity must track high-frequency indicators such as monthly job postings. Algorithmic management and job quality indicators should also be quantified. Data processing workflows should feed automatically updated dashboards rather than quarterly or annual static reports. Regular public-facing engagement at Cabinet level must also be prioritised, to prevent misinterpretation from triggering premature policy responses. Once built, this monitoring capacity may in turn offer a blueprint for lawmakers in other regions seeking to improve their own AI monitoring and preparedness.

European policymakers and politicians want to offer a positive vision of an AI-powered future and build broad support for the ambitious reform it requires. To realise such a vision, the EU must prepare to act, not react, and do so on the basis of timely, accurate, and accessible facts about the economy.