Artificial Intelligence, Productivity, and the Workforce: Evidence from Corporate Executives (w34984)
A · Strong evidence NBER paper by Federal Reserve Bank of Atlanta and Richmond economists with Duke's John Graham, using the established CFO Survey panel to ask executives for AI-attributed outcomes; method, weighting and response rates documented.
Summary based on the full text·Original: about 75 minutes to read
In short
NBER Working Paper 34984 (March 2026) by Salomé Baslandze, Zachary Edwards, John Graham, Ty McClure, Brent H. Meyer, Michael Sparks, Sonya R. Waddell and Daniel Weitz. CFOs report AI productivity gains larger than the gains implied by their own revenue and staffing data, a "productivity paradox" the authors link to delayed revenue. Near-term job effects are small overall: large firms expect AI-driven cuts, small firms modest growth, and work shifts from routine clerical to technical roles.
Key numbers
- nearly 750financial executives surveyed (603 CFO Survey panel responses plus 145 supplemental)
- 1.8%mean CFO-reported labour productivity gain from AI in 2025, against 0.6% implied by revenue and employment changes
- 3.0%mean labour productivity gain from AI that CFOs expect in 2026
- less than 0.4%expected decline in weighted aggregate employment due to AI in 2026
- more than 2 percentage pointsexpected fall in the routine clerical share of employment over three years, mostly at large firms
As reported by the source.
English·Web page·opens nber.org
What it reports
- More than half of firms have invested in AI. Smaller firms are only starting in 2026. Barriers: lack of training, immature technology, privacy concerns.
- Reported AI productivity gain: 1.8% in 2025 and 3.0% expected in 2026. Implied revenue-based gain: 0.6% in 2025 and 1.8% expected in 2026. Reported 2025 gains roughly match 2026 implied gains, suggesting a one-year lag.
- High-skill services and finance gain most: about 0.8% implied in 2025, over 2% expected in 2026. Low-skill services, manufacturing and construction about 0.4%.
- Gains come mostly from revenue-based TFP, not capital deepening, and are tied to new products and serving customers better rather than cost cutting.
- Weighted aggregate employment is expected to fall by less than 0.4% from AI in 2026.
- A Negative Exposure Index from open-ended answers ranks office and administrative support as the most replaced function; professional, technical and sales roles are more often described as enhanced.
How they got it
Survey fielded 11 November to 16 December 2025 to panel members of The CFO Survey (Duke University with the Richmond and Atlanta Feds), giving 603 responses, plus 145 responses from FEI, NASDAQ and Duke alumni contacts to mid-January 2026. Response rate 24% (45% among active CFOs). Results use representativeness and employment weights. A public dashboard shows de-identified data.
Limits
- AI effects are attributed by executives themselves, so they rest on judgement.
- The median firm is small (revenue $46 million), with a few very large firms driving weighted results.
- 2026 numbers are expectations.
Why it matters for companies
The gap between perceived and measured gains is a useful internal test: if finance cannot see AI productivity in revenue per person, check whether the gains are lagged or overstated before scaling spend. The firm-size split matters for planning: large firms expect cuts while smaller ones expect to grow, so redeployment and external hiring markets will move in different directions. The clerical-to-technical shift tells HR where reskilling demand will come from.
Summary based on the full text. Read the original for the full argument and data.Read the original on nber.org
- Tags
- Productivity, Workforce