U.S. Needs Better Data Infrastructure to Track AI Adoption

The measurement gap

Researchers and policymakers cannot answer basic questions about artificial intelligence's economic footprint: How many companies have adopted AI tools? What tasks are they automating? How many employees work with AI systems daily, and in what capacity?

These aren't academic curiosities. Without reliable data on AI adoption and use, government officials are crafting policy in the dark—at precisely the moment when AI is reshaping labor markets and competitive dynamics across industries.

Why it matters

Partial or anecdotal evidence about AI's economic impact creates real policy risk. Legislators and regulators need comprehensive, granular data to design workforce programs, update tax frameworks, and anticipate displacement effects. Guesswork leads to misdirected resources and unintended consequences when the technology is moving this fast.

Infrastructure exists but needs expansion

The United States already maintains statistical agencies capable of tracking economic transformation at scale. The challenge isn't building measurement systems from scratch—it's upgrading existing infrastructure to capture AI-specific metrics before the policy response falls further behind the technology curve.

According to Nathan Goldschlag at the Economic Innovation Group, the framework for collecting detailed, economy-wide data is in place. What's missing is the targeted investment to adapt that framework for AI measurement. Federal statistical agencies need resources to add AI-focused questions to existing surveys, develop new classification systems for AI-enabled work, and link firm-level adoption data with worker outcomes.

What comprehensive measurement requires

Effective AI tracking demands more than asking companies whether they use machine learning. Researchers need to distinguish between firms experimenting with AI tools and those deploying them at production scale. They need to separate workers who occasionally use AI assistants from those whose core responsibilities have been restructured around AI systems.

That level of detail requires survey design expertise, coordination across multiple data collection programs, and sustained funding—not one-time appropriations but ongoing capacity building within agencies like the Bureau of Labor Statistics and Census Bureau.

The alternative is making consequential economic policy based on incomplete snapshots and vendor surveys with obvious selection bias. As Goldschlag argues in his essay, published by the Economic Innovation Group, the statistical infrastructure investment needed is modest compared to the cost of getting the policy response wrong.

The details were first reported by the Economic Innovation Group.