Making AI Productive: Evidence from the Danish Economy

This project combines large-scale surveys, administrative register data from Statistics Denmark,
anonymized AI usage logs from tool developers, and randomized “AI accelerator” experiments to
study how AI adoption affects the Danish economy.
We field nationwide surveys of firms and workers and, in collaboration with Anthropic and
Microsoft, link anonymized enterprise AI usage logs to the microdata at Statistics Denmark. The
resulting dataset measures AI use with unprecedented precision and links it directly to firm and
worker outcomes.
Using the linked survey, usage-log, and register data, we characterize patterns of AI adoption across
firms, workers, and managers, and apply difference-in-differences methods to study how adoption
relates to productivity, profitability, employment, wages, and worker mobility.
To move beyond observational evidence, we recruit firms into randomized “AI accelerator”
programs that provide access to enterprise AI tools, with or without complementary training. These
accelerators generate experimental variation in AI adoption, enabling us to identify key barriers to
adoption and estimate causal effects on firm and worker outcomes.