Research
Research
Working papers
This paper examines how offshoring reshapes firms’ task structure and skill composition. I distinguish between two margins of skill adjustment: reallocating workers through external labor markets and retraining incumbent workers into new occupations within the same firm. The analysis shows that offshoring firms use these two margins on different segments of the task structure: they retrain manual and production workers who move to other jobs within the organization, while hiring younger and more educated workers for cognitive tasks. The results suggest that internal labor markets and retraining can facilitate structural transformation, though they may not fully address skill gaps arising from changes in task demand.
Who Adopts AI? Evidence on Firms, Technologies and Workers with Mariola Pytlikova, Sarah Schroeder, Magnus Lodefalk
Using two waves of nationally representative Danish firm surveys linked to employer--employee administrative registers, we study how adoption varies across artificial intelligence (AI) and related advanced technologies. We show that AI adoption is highly technology-specific. While firm size and digital infrastructure predict adoption broadly, workforce composition operates through distinct channels: STEM-educated workforces predict core AI adoption, whereas non-STEM university-educated workforces are associated with generative AI adoption, indicating different human capital complementarities. The factors associated with adoption differ from those predicting deployment breadth: firm size and digital maturity matter for both, whereas workforce composition primarily predicts adoption alone. Machine learning and natural language processing are deployed across multiple business functions, whereas other advanced technologies remain concentrated in specific operational domains. Individual-level evidence provides a foundation for these patterns, with awareness of workplace AI usage concentrated among managers and high-skilled workers. Self-reported AI knowledge is higher among younger and more educated individuals. Finally, commonly used occupational AI exposure measures vary substantially in their ability to predict observed adoption, with benchmark-based measures outperforming patent-based and LLM-focused alternatives. These findings show that treating AI as a monolithic category obscures economically meaningful variation in who adopts, what they deploy, and how well existing measures capture it.
This paper provides new evidence on how automation transforms firms’ demand for skills, not by changing the occupational composition, but by reshaping what existing jobs require. Using matched data on firm-level automation investments and detailed job ads from Denmark, we extract multidimensional skill profiles through natural language processing. Guided by a task-based framework, we decompose changes in skill demand into within- and between-occupation components and find that within-occupation adjustments dominate. Automation increases the demand for soft skills in professional and managerial roles and reduces the reliance on routine hard skills in production jobs. Register data confirm parallel shifts in workforce composition: increased experience in high-skill occupations and intensified retraining among lower-skilled workers. Our findings highlight that automation reconfigures work from within, with important implications for training policy and labor market resilience.
IZA DP No. 16717; RFBerlin DP 14/24
Despite widespread concern, the labour-market impacts of artificial intelligence (AI) remain hard to quantify. We construct a Dynamic AI Occupational Exposure (DAIOE) index—tracking annual frontier gains across nine AI subdomains (2010–2023)—and link it to employer–employee registers in Denmark, Portugal, and Sweden. While AI exposure has no net effect on total firm employment, it is consistently tied to compositional up-skilling: firms with higher DAIOE scores increase their high-to-low skill employment ratios. Firms with higher DAIOE scores tend to expand high-skill white-collar positions and reduce low-skill clerical roles. Blue-collar effects average out but vary sharply by subdomain. Our findings underscore the need to unpack “AI” into its component technologies when evaluating labour-market outcomes.
Selected Work in Progress
The Employment Effects of AI Adoption with Mariola Pytlikova, Sarah Schroeder