
INSKILL measures which firms adopt AI, and the consequences of adoption for their workforce and productivity, rather than which occupations
are potentially exposed to it.
INSKILL constructs a firm-level database for Belgium, France, Germany and Luxembourg from three sources that had not previously been linked. These are the text of firms’ own websites, 120 million online job advertisements, and individual employment histories from professional-network profile data. Linking them to administrative firm data makes it possible to estimate not only which firms adopt AI but the subsequent effects on their employment, workforce composition and productivity.

MAP-AI, the project’s first output, is a firm-level indicator of realised AI adoption. It is obtained by applying a large language model to the text of firm websites and classifying each firm against the European Commission’s definition of AI. The share of AI-active firms rose from about 1% in 2016 to nearly 12% in 2024, with acceleration after 2022. Luxembourg has the highest share, at 23%, much of which reflects its large financial sector. Adjusted for sector composition the share is about 14%, close to that of Germany. Realised adoption is only weakly correlated with the exposure indices used in policy analysis. Workforce skills are the strongest correlate of adoption. Data skills form a necessary base, and a small set of specialised AI skills act as strong complements.
A second paper, in preparation, moves from description to causal analysis. It links the adoption measure to French administrative employer–employee and balance-sheet records and applies an event-study design around the year of a firm’s first observed adoption, estimating the effects on employment, workforce composition and productivity.
Firms covered
3M+
Years observed
2016–2024
Accuracy against human coding
94%
Correlation with Eurostat
r = 0.94

Project
Funded by the LISER Innovation Call 2024 and led by Terry Gregory, running from December 2024 to December 2026.
Data
Website text for more than three million firms in Belgium, France, Germany and Luxembourg, 2016–2024, together with 120 million online job advertisements and professional-network profile data on individual employment histories, linked to business registers and to French administrative data.
Method
A large language model classifies each firm’s website text for realised AI use and assigns an ecosystem role and a technology type. The causal analysis applies an event-study design around the timing of first adoption.


Explore key topics driving this project, addressing challenges like ai and skills, digital transformation, with innovative solutions.
A passionate team of experts collaborates to deliver meaningful research and innovative solutions. Their dedication ensures this project achieves impactful outcomes.




United in purpose. This department contributes with its expertise to drive research and strategies that create tangible societal change.