Rethinking Automation Risk: AI Applicability and Occupational Outcomes, 2019-24 (WP 2026-12)
B · Useful, limited Chicago Fed working paper by Broady, Dunson and Barr that checks Frey-Osborne risk scores and Microsoft Copilot applicability scores against BLS outcomes; transparent but descriptive group averages over a short, pandemic-affected window.
Summary based on the full text·Original: about 40 minutes to read
In short
Federal Reserve Bank of Chicago Working Paper 2026-12 by Kristen Broady, Caleb Dunson and Anthony Barr. It looks back at 2019-2024 US occupational outcomes using Frey and Osborne's 2017 computerisation risk scores and Tomlinson et al.'s 2025 AI applicability scores from Copilot usage. Occupations most applicable to AI grew in jobs and wages; high-risk occupations did worse on employment but saw wage gains. The authors read exposure as a sign of restructuring, not a forecast of job loss.
Key numbers
- 4.05%employment growth 2019-2024 in the 38 occupations with the highest AI applicability scores
- 21.7%wage growth over the same period in those high-applicability occupations
- -3%average employment change in occupations Frey and Osborne rated high risk of computerisation
- -51%employment change for telemarketers, the steepest decline among high-risk occupations
As reported by the source.
English·Web page·opens doi.org
What it reports
- High-applicability occupations (38 matched): employment from about 8.3 million to 8.62 million, up 4.05%; average wages from $63,363 to $77,138, up 21.7%.
- Low-applicability occupations (35 matched): employment fell 9.5% in 2020 and was still slightly below 2019 by 2024; wages rose about 25%.
- Frey-Osborne high-risk occupations: employment down 3% on average, wages up 26%. Telemarketers -51%, switchboard operators -47%; models +131%.
- Moderate-risk: employment down 6%, wages up 21%. Low-risk: employment up 8%, wages up 22%.
- Reviews how robotics, LLMs and generative AI have eroded Frey and Osborne's bottlenecks in perception and manipulation, creative and social intelligence.
How they got it
Matches both scores to O*NET codes and BLS employment and wage data for 2019 to 2024, then compares totals and averages across score groups.
Limits
- No controls or counterfactual; the COVID shock dominates 2020-2021.
- The window ends in 2024, before agentic tools spread.
- Small matched samples (35 and 38 occupations) for the applicability comparison.
- Wage gains in shrinking occupations may reflect composition change.
Why it matters for companies
A useful corrective for workforce plans built on exposure scores: high exposure has so far gone with growth in many roles. Exposure scores tell you where tasks will change, so they should drive role redesign and reskilling priorities rather than headcount cuts on their own.
Summary based on the full text. Read the original for the full argument and data.Read the original on doi.org
- Tags
- Occupations, Labour market