Ramp Tracks 21,559 Firms: Heavy AI Adopters Grew Headcount 10%, Entry-Level Roles Up 12%
A new paper from Ramp Economics Lab, co-authored with Revelio Labs, offers one of the cleaner looks yet at how AI adoption affects firm-level employment. The dataset is transaction-level: Ramp observes which companies are actually spending on AI tools through card and bill-pay data, not self-reported survey responses. That spending data is linked to Revelio’s workforce records for 21,559 US firms.
The headline result runs against the dominant narrative. Companies making the largest AI investments grew total headcount 10.2% over the two years following adoption. Entry-level positions grew 12%. Low-intensity adopters — companies that signed up for AI tools but did not substantially increase their AI spend — showed no statistically significant employment change.
How the Split Works
The study separates firms into high-intensity and low-intensity AI adopters based on their observed AI spending levels through Ramp. The effect size difference between the two groups is stark. High-intensity adopters see employment begin rising around months four to five post-adoption and continue building through month 24. The low-intensity group flatlines throughout the measurement window.
Employment gains for high-intensity adopters are broad: engineering, sales, administration, and customer service all show positive effects. The study does not report displacement within firms — it tracks net headcount, so within-role substitution and offsetting new hires are netted out.
Sector concentration is real. Gains in the aggregate are driven heavily by the Information sector. Whether that reflects where AI productivity gains are largest or where AI-intensive firms happen to be concentrated is not resolved in the paper.
The Selection Bias Problem
The study is direct about its limits. AI adopters are not a random draw from the US firm population. Ramp’s customer base skews toward venture-backed, engineering-intensive, faster-growing companies. Before adoption, high-intensity AI adopters were already larger and growing faster than comparable non-adopters.
That selection effect means the employment gains cannot be attributed to AI adoption alone. Firms that were going to grow quickly are the ones that also invested heavily in AI tools. Controlling for pre-adoption growth trajectories reduces the estimated effect but does not eliminate it — the post-adoption acceleration is statistically distinguishable from the pre-adoption trend.
What the data rules out clearly is broad net job destruction at adopting firms over a two-year window. The hypothesis that companies quickly cut headcount after deploying AI is not supported in this sample. The alternative hypothesis — that AI investment is associated with growth, not contraction — fits the observed data better.
Context
The finding sits alongside conflicting signals on AI’s labor market effects. Challenger Gray’s data has shown AI cited as the primary reason for US job cuts for multiple consecutive months. The Verizon CEO publicly projected AI-driven unemployment reaching 20-30%. The Ramp paper does not contradict those signals; it is observing a different part of the distribution — the firms that chose to invest heavily in AI tools, not the firms cutting labor to replace it.
The two trends can coexist: firms making large AI investments may be expanding while firms in sectors more exposed to automation contract. The Ramp data does not capture the second group.
Authors: Kharazian, A., Simon, L., and Stevens, R. Published June 2026, Ramp Economics Lab.