A New Look at AI’s Impact on Jobs
Ramp – Revelio: Firm-Level AI Spending and Workforce Adjustment
Abstract
We study how employment changes when firms adopt generative AI using observed AI spending from Ramp card and bill pay data linked to Revelio Labs workforce records for 21,559 firms in the United States. We find that companies that adopt AI tend to grow faster following adoption, but the relationship is driven almost entirely by high-intensity adopters. Firms making the largest AI investments grow employment by roughly 10% following adoption, while low-intensity adopters see no statistically significant change. Entry-level headcount rises 12% for high-intensity adopters. Gains emerge gradually and are broad across roles, including engineering, sales, administration, and customer service. They are also uneven: adopters are already larger, more technical, faster-growing firms, and sector-level gains are concentrated in Information. The results counter predictions that AI adoption will lead to broad job loss.
Key Takeaways
- **Firms that adopt AI grow headcount 10.2% over the two years following adoption**, but these gains are entirely driven by high-intensity adopters. Low-intensity adopters see no statistically significant change.
- **Entry-level headcount grew even faster.** At the companies making the largest AI investments, entry-level headcount grew 12% over the two years following adoption.
- **AI adoption and the associated gains are unevenly distributed**. AI adopters are already larger, more engineering-intensive, more likely to be venture-backed, and faster-growing than non-adopters. These firms then grow faster upon adoption.
Change in headcount after AI adoption: Total Headcount
Effect of AI adoption on firm-level employment by relative month since adoption
Job Functions
Job function
Education
Education
High AI intensity Low AI intensity
-20
0
20
40
60
80
View chart data as a table
Change in headcount after AI adoption: Total Headcount: estimates and 95% confidence intervals
| Month Relative to Adoption | Effect on High-Intensity Firms (log points × 100) | Low-End Confidence Interval on High-Intensity Firms (log points × 100) | High-End Confidence Interval on High-Intensity Firms (log points × 100) | Effect on Low-Intensity Firms (log points × 100) | Low-End Confidence Interval on Low-Intensity Firms (log points × 100) | High-End Confidence Interval on Low-Intensity Firms (log points × 100) | | --- | --- | --- | --- | --- | --- | --- | | -12 | -2.74 | -7.11 | 1.64 | -0.39 | -3.40 | 2.63 | | -11 | -2.07 | -6.07 | 1.94 | -1.07 | -3.97 | 1.83 | | -10 | -1.97 | -5.82 | 1.88 | -0.67 | -3.66 | 2.32 | | -9 | -3.50 | -6.93 | -0.07 | -0.62 | -3.18 | 1.94 | | -8 | -3.07 | -6.60 | 0.47 | -0.29 | -2.55 | 1.97 | | -7 | -2.17 | -5.30 | 0.96 | -0.18 | -2.16 | 1.80 | | -6 | -3.83 | -6.56 | -1.10 | -0.35 | -2.16 | 1.46 | | -5 | -3.01 | -5.38 | -0.65 | -0.18 | -1.72 | 1.36 | | -4 | -1.19 | -3.43 | 1.05 | -0.12 | -1.29 | 1.04 | | -3 | -1.50 | -3.09 | 0.08 | -0.12 | -0.93 | 0.69 | | -2 | -0.84 | -1.93 | 0.24 | -0.09 | -0.56 | 0.38 | | -1 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | | 0 | 0.32 | -0.85 | 1.49 | -0.13 | -0.58 | 0.32 | | 1 | 0.97 | -0.73 | 2.68 | 0.12 | -0.61 | 0.86 | | 2 | 1.45 | -0.88 | 3.78 | -0.18 | -1.24 | 0.88 | | 3 | 2.01 | -0.83 | 4.86 | -1.03 | -2.47 | 0.42 | | 4 | 3.66 | 0.30 | 7.02 | -1.32 | -3.13 | 0.50 | | 5 | 5.83 | 1.65 | 10.01 | -1.19 | -3.37 | 0.99 | | 6 | 7.05 | 2.10 | 11.99 | -1.56 | -4.14 | 1.03 | | 7 | 8.79 | 3.52 | 14.06 | -1.83 | -4.90 | 1.24 | | 8 | 9.71 | 3.63 | 15.79 | -2.21 | -5.59 | 1.16 | | 9 | 11.11 | 4.66 | 17.55 | -2.20 | -5.60 | 1.21 | | 10 | 13.76 | 6.50 | 21.02 | -1.60 | -6.21 | 3.00 | | 11 | 17.53 | 10.14 | 24.92 | -0.65 | -5.49 | 4.19 | | 12 | 18.79 | 10.43 | 27.15 | -0.99 | -6.20 | 4.21 | | 13 | 19.02 | 10.18 | 27.86 | -1.75 | -7.58 | 4.08 | | 14 | 20.94 | 11.27 | 30.60 | -1.42 | -7.84 | 5.01 | | 15 | 20.05 | 10.21 | 29.89 | -0.42 | -6.82 | 5.98 | | 16 | 20.88 | 9.44 | 32.33 | -0.64 | -8.50 | 7.22 | | 17 | 22.62 | 10.70 | 34.54 | -0.14 | -9.20 | 8.92 | | 18 | 27.70 | 14.70 | 40.69 | 0.88 | -9.29 | 11.05 | | 19 | 30.66 | 15.67 | 45.65 | 2.26 | -8.37 | 12.88 | | 20 | 31.96 | 14.57 | 49.34 | 1.49 | -10.68 | 13.67 | | 21 | 35.25 | 16.48 | 54.02 | 3.96 | -9.27 | 17.18 | | 22 | 35.99 | 16.15 | 55.82 | 5.32 | -10.84 | 21.47 | | 23 | 36.47 | 15.45 | 57.49 | 4.60 | -11.90 | 21.11 | | 24 | 45.24 | 21.67 | 68.82 | 3.68 | -13.32 | 20.67 |
Source: Kharazian, A., Simon, L., & Stevens, R. (2026). A New Look at AI’s Impact on Jobs: Firm-Level AI Spending and Workforce Adjustment. Ramp Economics Lab. https://ramp.com/data/ai-jobs-impact
Methodology
We link firm-level Ramp transaction data to Revelio Labs workforce records to study employment changes following paid generative AI adoption. Ramp card and bill pay records identify monthly AI vendor spend, including foundation models, GPU cloud, model serving, coding agents, API tools, and AI image, video, search, and research products. Revelio provides monthly firm-level headcount and workforce composition. The linked panel covers 21,559 U.S. firms observed monthly from January 2021 through February 2026.
We define AI adoption as the first month of the earliest three-consecutive-month spell in which a firm records at least $100 of AI vendor spend in every month. This rule screens out one-off experiments and captures sustained organizational use. Among treated firms, adoption intensity is measured as AI spend per baseline employee over the first three post-adoption months, using headcount in the month before adoption as the denominator. High-intensity adopters are firms in the top tercile of this measure; lower-intensity adopters are firms in the bottom two terciles.
Our preferred design is a Callaway-Sant’Anna staggered difference-in-differences estimator. Because permanent non-adopters differ sharply from adopters before treatment, the main comparison is not adopter versus never-adopter. Instead, each adopter is compared to firms in the same eventual intensity group that adopt later but have not yet adopted in that calendar month. Specifications include NAICS sector fixed effects, estimate dynamic effects from 12 months before adoption through 24 months after adoption, and use pre-adoption event-study coefficients as the main diagnostic for parallel trends.
Frequently asked questions
We identify AI adoption from Ramp card and bill pay transactions to AI vendors, including OpenAI and Anthropic and other AI infrastructure companies and software firms. A firm is classified as adopting AI in the first month of its earliest three-consecutive-month spell with at least $100 of AI vendor spend in each month. High-intensity adopters generally spend much more. This screens out one-off purchases and captures sustained paid AI use.
Among adopters, intensity is measured as AI spend per baseline employee over the first three months after adoption begins. Baseline headcount is measured in the month before adoption. High-intensity adopters are firms in the top tercile of this measure; lower-intensity adopters are firms in the bottom two terciles. In the estimating sample, lower-intensity adopters average about $2.78 per employee per month in the first 3 months of adoption, while high-intensity adopters average about $33.67. In both groups, per employee per month spend rises as firms advance along the adoption curve.
The preferred comparison group is not-yet-treated firms: companies in the same eventual AI-intensity group that adopt AI later but have not yet adopted in a given event month. The specification also includes NAICS sector fixed effects. As a robustness check, comparing adopters to never-adopters still shows larger employment gains, about 12.6% for high-intensity adopters; however, the non-adopter control group is less reliable. Adopters were already growing headcount faster than non-adopters pre-adoption.
Our firm-level results show no broad job loss — high-intensity adopters grew headcount, including entry-level roles. We cannot rule out reallocation within firms over a longer horizon; the 24-month window may be too early to detect large changes in workforce mix. We plan to update our results regularly to track firms beyond the first 24 months of adoption.
No. The sample consists of firms using Ramp that can be linked to Revelio Labs workforce records and meet the paper’s activity filters. It likely skews toward tech-forward firms and workers in knowledge-work. It is best interpreted as evidence from a selected, business-spend-active firm population, not a nationally representative sample of all U.S. businesses. In any case, we also report results sector-by-sector.
For press and media inquiries, contact press@ramp.com. For questions about the research, methodology, or data, reach out to the corresponding author, Ara Kharazian, at ara.kharazian@ramp.com.
Cite this work
APA
Kharazian, A., Simon, L., & Stevens, R. (2026). A New Look at AI’s Impact on Jobs: Firm-Level AI Spending and Workforce Adjustment. Ramp Economics Lab. https://ramp.com/data/ai-jobs-impact
BibTeX
@techreport{kharazian2026ainewlook, author = {Kharazian, Ara and Simon, Lisa and Stevens, Ryan}, title = {A New Look at AI's Impact on Jobs: Firm-Level AI Spending and Workforce Adjustment}, year = {2026}, month = {June}, institution = {Ramp Economics Lab}, url = {https://ramp.com/data/ai-jobs-impact} }
Don't miss key shifts in business spend from Ramp Economics Lab.
You might also like
