AI Adoption Velocity, Productivity and Labour Outcomes in Europe: A Country-Sector Dataset with the AI Attribution Wedge, 2021–2025
Description
This dataset accompanies the study “Too Fast to Become Productive? AI Adoption Velocity and the Organisational Limits of Technological Realisation in Europe.” It provides a harmonised country-sector panel designed to examine the relationship between enterprise artificial intelligence (AI) adoption, the speed of AI diffusion, labour productivity, employment, working time and the compensation share across European economies. The dataset contains 733 country-sector-year observations covering 27 European Union countries, seven market sectors and four AI survey waves (2021, 2023, 2024 and 2025). The sectoral coverage includes manufacturing and utilities, construction, wholesale and retail trade, transport and storage, accommodation and food services, information and communication, and real estate. The data are constructed primarily from harmonised Eurostat enterprise ICT surveys and national accounts. Enterprise AI adoption is measured as the percentage of enterprises with at least ten persons employed reporting the use of at least one AI technology. The dataset combines this measure with real gross value added, employment, total hours worked, compensation of employees, GDP growth, unemployment and inflation. It additionally contains constructed forward-looking outcomes for labour-productivity growth, employment growth, hours-per-worker growth and changes in the compensation share. A central feature of the dataset is AI adoption velocity (dai), defined as the annualised percentage-point change in enterprise AI adoption between available survey waves. It also includes the study's descriptive AI Attribution Wedge (aaw), which combines standardised AI adoption, subsequent productivity performance and hours-per-worker dynamics to identify country-sector observations where relatively rapid or high AI diffusion coexists with comparatively weak realised productivity and non-declining labour time. The AAW is intended as a descriptive screening measure rather than a causal indicator. The deposited materials also include a README file providing documentation, variable definitions, data-structure information and instructions for using and reproducing the dataset. The accompanying R Codes file contains the R scripts used for data processing, variable construction, econometric analyses, robustness checks, and the production of the tables and figures reported in the study.