From Mainframes to Machine Learning: Skill Gaps over the Technology Life Cycle

Abstract

Despite the productivity gains associated with digital technologies, their diffusion across firms remains slow and uneven. A large literature attributes this pattern to skill gaps between firms and workers, yet most existing studies treat these gaps as an exogenous barrier to adoption. This paper shows that skill gaps instead evolve endogenously over the technology diffusion process. We develop a framework in which the gap between firms' skill requirements and workers' capabilities changes over the technology life cycle. To test the model, we construct granular measures of firm skill demand and worker skill supply using matched data on job postings and worker résumés. Our analysis yields three key findings. First, we document a systematic U-shaped pattern in skill gaps over the technology life cycle: skill gaps are high when new technologies are first adopted, decline as firms and workers adjust, and rise again as technologies mature into legacy systems. Second, we show that diffusion bottlenecks arise not only from shortfalls in technical expertise but also from shortages of complementary nontechnical capabilities, in particular managerial and coordination skills. Third, we demonstrate that skill gaps help explain the persistence of firms operating legacy systems with limited growth prospects. Together, these findings highlight skill gaps as a dynamic force shaping both the diffusion of new technologies and the persistence of older ones.