A University of Chicago / University of Copenhagen working paper (April 2025) links two large-scale surveys (≈25,000 workers, 7,000 workplaces, 11 white-collar occupations) to Danish administrative earnings and hours records through June 2024, roughly 18 months after ChatGPT's launch. Despite widespread adoption (64% ever-used at encouraged workplaces), substantial firm investments (38% with in-house models, 30% of workers trained), and self-reported time savings averaging 2.8% of work hours, difference-in-differences and quasi-experimental estimates rule out earnings or hours effects larger than 1% at the individual level and 1% at the workplace level. The gap between controlled-experiment productivity gains (15–50% in RCTs) and real-economy outcomes is attributed to occupational heterogeneity in task fit, weak wage pass-through (3–7%), and incomplete firm-level complementary reorganization. For financial institutions, this is the most methodologically rigorous population-level data point to date on the timeline between AI adoption and measurable labor-cost or productivity impacts, directly bearing on whether enterprise ROI claims will translate into durable economic outcomes in the near term.
343 other tracked developments touching the same claims.