Ramp Studied 21,559 Companies. The Ones Spending Most on AI Are Hiring More, Not Less.

For two years, the standard forecast has been that AI adoption shrinks payrolls. A working paper published in June by Ramp Economics Lab and Revelio Labs offers some of the first large-scale evidence to the contrary, and it comes from an unusually solid place: transaction records.

Because Ramp processes corporate card and bill payments, its researchers could see exactly which companies were paying AI vendors, how much, and when. The tracked spend covers foundation models, GPU cloud, model serving, coding agents, API tools, and AI image, video, search, and research products. Revelio Labs supplied monthly firm-level headcount and workforce composition for the same companies. Linking the two produced a panel of 21,559 US firms observed monthly from January 2021 through February 2026, where "AI adoption" means money actually leaving the building, not a checkbox on a survey.

What the researchers found is that companies making serious, sustained AI investments grew their workforces in the two years after adopting. Total headcount rose 10.2%, and entry-level headcount, the segment most often predicted to disappear first, rose 12%.

A threshold effect, not a gradual one

The more interesting finding sits below the headline. The employment gains don't scale smoothly with spending. They behave like a threshold: below a certain level of investment, nothing happens at all.

The study counts a firm as an adopter once it records at least $100 in AI vendor spend for three consecutive months, a rule designed to screen out one-off experiments, then sorts adopters by how much they spend per employee in their first three months. The bottom two-thirds average $2.78 per employee per month, and their employment trajectories are statistically indistinguishable from firms that hadn't adopted yet. All of the measured hiring growth comes from the top third, which averages $33.67 per employee per month.

The full working paper offers a plausible explanation for why such a modest dollar difference matters so much. Spend at that level usually indicates a different kind of usage: coding agents, API access, and multiple models or vendors, rather than a stack of chat subscriptions. The money is a proxy for how deeply the tools have been folded into daily operations.

Seen that way, the study isn't really measuring spending. It's measuring commitment, with spending as the visible trace.

Patience turns out to be part of the investment

The employment effects also arrive slowly. Adopting firms look no different from their peers in the first few months. The divergence begins somewhere between month six and month twelve, then widens through the rest of the two-year window.

Ramp's economist frames this as a learning curve: organizations need time to restructure work around the tools before the gains materialize, and once they do, the gains compound over time.

This has an awkward implication for how AI initiatives typically get evaluated. A pilot that gets reviewed at the six-month mark, right when the data says effects are only beginning to surface, will usually look like a failure. Budgets cut at that point kill programs that were on schedule, not behind it.

The growth is broader than expected, but not evenly spread

Hiring gains among the top-tier adopters weren't confined to technical teams. The paper documents increases across engineering, sales, administration, and customer service. Sector patterns are another matter: Information shows the strongest effects, coverage of the study notes that finance and insurance follow, while construction, healthcare, arts and entertainment, and food service lag.

The authors are careful about what they can and can't claim. Firms that adopted AI were already larger, more engineering-intensive, more likely to be venture-backed, and faster-growing before adoption, so the study compares adopters against similar firms that hadn't adopted yet, using a staggered difference-in-differences design, rather than against the economy at large. It's rigorous correlational work, not proof of causation, and the authors acknowledge that the 24-month window may be too early to detect larger changes in workforce mix. The sample also tilts toward tech-forward firms that use Ramp, not a nationally representative cross-section.

Those caveats narrow the claim without changing its shape. Within this population, depth of adoption is what distinguishes the companies that grew from the companies that didn't.

Two strategies wearing the same label

Nearly every company now describes itself as using AI. The data suggests that label covers two very different postures.

One treats AI as a cost lever. The tools get purchased, a productivity mandate goes out, and headcount plans get trimmed on the assumption that software will absorb the work. The other treats AI as an expansion of what the company can attempt: more products shipped, more customers served, more functions brought in-house. Firms in the second camp appear in the data as the top-tier adopters, and they respond to their new capacity by hiring into it, including at the junior levels where AI-assisted employees can now do work that once required years of experience. Ramp's economist makes a version of this argument directly, suggesting that AI lowers the fixed cost of building software, handling administrative work, and improving customer support, which unlocks growth that previously required new salaries.

The same commentary spells out the practical takeaway for job seekers: between two otherwise similar employers, a young person should choose the one using AI, because it's more likely to grow faster.

That flips the usual anxiety on its head. In this data, the employment risk doesn't come from AI arriving at your company. It comes from AI arriving in name only.

Why this is a brand story

The State of Brand covers this study because adoption depth is about to become part of how companies are perceived, whether they manage that perception or not.

A firm's position in this data is legible from the outside. Entry-level job postings, the language leadership uses when discussing AI (growth versus savings), and the growth rate itself all point to which posture a company has taken. As findings like Ramp's circulate, candidates and customers will start reading those signals deliberately. Claiming an AI strategy will be easy. Demonstrating the hiring pattern of a top-tier adopter will not.

The companies that treated the last two years as a genuine operational transition, rather than a procurement exercise, are about to have the numbers to show for it.

The research cited throughout is 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 and Revelio Labs, available at ramp.com/data/ai-jobs-impact.