Payrolls to Prompts: Firm-Level Evidence on the Substitution of Labor for AI

Introduction. AI is profoundly reshaping the nature of work. Every day, stories abound about startups disrupting industries, new models being released, and the anxiety of legacy companies getting left behind in the new AI era. An important question in this sector is whether firms are substituting labor for AI. Given its potential implications for labor markets, productivity, and income inequality, the mag- nitude and type of substitution is critical to understand. While this question is often asked, there is limited evidence on how businesses are spending on AI, and how that is interacting with the labor force. This study provides some of the first direct evidence on how firms are spending on AI tools, and how that is interacting with the labor force. We focus on two key questions: Are businesses substituting labor for AI? If so, what is the rate of substitution between AI and labor? Put differently, what are the cost savings from AI adoption? This study focuses on a specific sector of workers most impacted by AI: jobs contracted through online labor marketplaces.

Discussion / Conclusion. This study sheds light on how businesses are substituting labor for AI; however, the magnitude of the substitution is not uniform. Increasing exposure to AI shocks causes businesses to substitute away from labor marketplaces faster and at a higher rate. To our knowledge, this is the first study to show firm-level differences in substitution patterns. Substitution from labor to AI is not simply faster but also appears to be happening at a lower cost. Higher AI-exposed firms substitute relatively more labor for AI at a lower cost. While this study provides evidence, we cannot determine the mechanism behind the substitution. There are many different theories that could explain this pattern, such as returns to scale in building internal AI capabilities or having a natural advantage in AI adoption relative to less exposed firms. Exploring these mechanisms is beyond the scope of this study but is an important area for future research.