In April, the Federal Reserve published a working paper focused on quantifying and tracking AI adoption rates among US businesses. The fact itself is noteworthy: when monetary policymakers begin formally studying AI adoption as an economic indicator, it signals that AI's impact on the macroeconomy has become significant enough to warrant policy attention.
But the most interesting part of the paper is not its conclusions—it is the finding that measuring the same thing with three different methods yields dramatically different numbers.
Three Data Sets, Three Worlds
The Fed research team compared three major survey sources:
BTOS (Business Technology Survey): 18%
Asked at the enterprise level: Does your company use AI? About 18% of businesses said yes.
RPS (Randomized Productivity Survey): 41%
Asked at the individual level: Have you used generative AI at work? 41% of respondents said yes.
SBU (Employment-Weighted): 78%
After weighting for large companies: 78% of workers are employed at firms that have deployed AI.
The same "US AI adoption rate": 18%, 41%, 78%—all three numbers are correct, but they answer different questions.
How you define "adoption" directly determines the world you see.
Ask "Has the company formally deployed an AI system?" and the answer is 18%. Ask "Are employees using it?" and the answer is 41%. Weight for large-company employees (which have higher AI deployment rates) and the answer jumps to 78%.
Who Uses It Most
Breaking it down by industry reveals wide gaps:
| Industry | Enterprise Adoption Rate | Individual Usage Rate |
|---|---|---|
| Professional Services (law, consulting, accounting) | 33% | 62% |
| Financial Services | 30% | 63% |
| Manufacturing | Lower | Lower |
| Wholesale Trade | Lower | Lower |
Financial and professional services have the highest individual usage rates, with 63% of financial workers using generative AI on the job. This is unsurprising—both industries deal heavily with text, data, and analysis, where AI can be directly embedded.
By contrast, manufacturing and wholesale trade show significantly lower adoption, indicating that AI has not yet penetrated deeply into physical production.
Growth Rate of 68%, but from a Small Base
One figure in the paper stands out: at the end of 2025, the year-over-year growth rate of enterprise-level AI adoption was 68%.
But note—even with rapid growth, the base is only 18%. Growing 68% from 18% brings it to just over 30%.
More interesting is daily usage frequency:
- Employees using work-related AI daily: 12%
- Using it at least once a week: 35.2%
- Employment-weighted adoption rate for large language models (LLMs): 54%
At 12% daily usage, that would be considered "low frequency" for any other enterprise software. But by analogy: Excel's daily active user numbers in its early days were not like this either. The "stickiness" of AI tools is still in its early cultivation phase.
Why the Fed Cares
The Fed's interest in AI adoption rates is driven by very practical policy considerations.
AI's impact on labor productivity directly affects three things:
- Economic growth trends: Productivity growth is a core driver of GDP growth
- Inflation path: Productivity gains can suppress supply-side prices, allowing the economy to run faster without generating inflation
- Labor market structure: Which jobs are being replaced by AI, and what employment policy support is needed
In other words, the Fed is trying to determine: Are we in an AI-driven productivity revolution, or just a digital bubble?
This judgment directly influences interest rate decisions. If AI is systematically improving labor productivity, the economy can sustain growth for longer without overheating.
This paper does not provide a final answer, but it makes the question quantifiable and trackable.
Measurement Itself Is Progress
The paper's honesty is refreshing: it does not endorse "AI changes everything," but instead says, "We have three measurement methods, each with biases, and we are still learning how to observe this phenomenon more accurately."
This precisely illustrates that the quantification of AI's economic impact is still in its early stages. MIT, Stanford, and McKinsey each have their own methods; the Fed has its own. The numbers they produce conflict—not because the data is false, but because they are not measuring the same thing.
From policymakers beginning to seriously quantify AI's economic impact to these data actually entering monetary policy models may take several years. But the direction is set: AI in macroeconomics is moving from "topic" to "indicator."
The day the Fed starts citing AI adoption rates in its interest rate decision statements will be the truly historic milestone. For now, we are probably about one-third of the way down that road.
Sources: Monitoring AI Adoption in the US Economy (Federal Reserve Board, FEDS Notes, CocoLoop, April 2026); Enterprise AI Adoption Curve Now Past the Internet at Year 3 (Asanify AI News Digest, April 20, 2026)