The Productivity J-Curve of Generative AI Adoption: Early Evidence from Service Sector Firms

This article investigates the impact of generative artificial intelligence (GenAI) tools on firm-level labor productivity and task composition. Leveraging a novel panel dataset of 2,500 U.S. firms in customer support, software development, and content creation, we implement an event-study methodology around GenAI adoption dates. Results indicate an initial 6–8 month “learning dip” where productivity declines by 4% due to workflow retooling, followed by a sustained 19% productivity gain at the 12-month mark. Gains concentrate in non-routine cognitive tasks (e.g., summarization, code debugging) but are absent for fully automatable roles. Crucially, firms that combine GenAI with worker retraining see twice the productivity lift of those that simply replace labor. The evidence supports a J-curve adoption pattern and suggests that AI’s economy-wide benefits depend on complementary human capital investments.