Scaling Laws for Economic Productivity: Experimental Evidence in LLM-Assisted Consulting, Data Analyst, and Management Tasks
Ali Merali
TL;DR
The paper provides experimental scaling laws linking LLM compute and algorithmic progress to real-world professional productivity across multiple domains. By conducting a large RCT with 13 models, it shows that frontier AI accelerates task completion and boosts output quality, with distinct gains for non-agentic versus agentic work. A decomposition into compute versus algorithmic progress reveals compute scaling accounts for roughly half of the observed improvements, while calendar-time progress captures the rest, implying durable, accelerating productivity effects if model scales continue. Applying these elasticities within an Acemoglu-style framework yields an estimated ~20% gain in U.S. productivity over the next decade, though benefits are uneven across task types and contingent on the degree of human–AI complementarity. The work offers a rigorous, economically grounded benchmark for the marginal productivity of labor in human–AI teams and highlights where further R&D or workflow redesign may be needed to realize full potential.
Abstract
This paper derives `Scaling Laws for Economic Impacts' -- empirical relationships between the training compute of Large Language Models (LLMs) and professional productivity. In a preregistered experiment, over 500 consultants, data analysts, and managers completed professional tasks using one of 13 LLMs. We find that each year of AI model progress reduced task time by 8%, with 56% of gains driven by increased compute and 44% by algorithmic progress. However, productivity gains were significantly larger for non-agentic analytical tasks compared to agentic workflows requiring tool use. These findings suggest continued model scaling could boost U.S. productivity by approximately 20% over the next decade.