Ramp study: AI-heavy spenders grew headcount 10.2% in two years

For two years the implied story has been that AI capex would thin out payrolls. A new joint study from corporate card firm Ramp and workforce analytics firm Revelio Labs, as covered by the Financial Times, pulls in the other direction. Linking actual spend records to headcount data for 21,559 U.S. firms from January 2021 through February 2026, the researchers find that firms which adopt AI grew headcount 10.2% over the two years following adoption, and that the entire gain was driven by the heaviest spenders. Low-intensity adopters saw no statistically significant change.

The split inside that headline is where it gets interesting. Among high-intensity adopters, entry-level headcount grew 12% and managers and officers grew 6.7%, a pattern that runs against the assumption that AI would eat the junior roles first. The effect did not show up immediately either. Hiring gains emerged six to twelve months after adoption, consistent with firms needing integration time before a productivity dividend translates into more bodies on the org chart. Ramp's own writeup defines high-intensity adoption as roughly the top third of per-employee AI spending, anchored around thirty dollars per employee per month in the initial three months after adoption.

Where it concentrates matters as much as the topline. The heaviest adoption, and the headcount expansion that follows, sits in information, finance, and insurance, while construction, healthcare, arts, entertainment, and food services lag. The researchers also note that firms which adopted AI were already larger, more engineering-intensive, more often venture-backed, and growing faster before they ever bought a model subscription. The honest caveat is that this is a correlation inside an already-thriving subset of American firms, not a clean causal claim that buying more model seats produces hires at a steel fabricator in Ohio.

What the reporting does not give you is whether those hires are durable past the two-year window, what share of the new roles are AI-adjacent rather than ordinary backfill, or how the curve looks once you strip out venture-funded growth. The forward look is for traditional employers in the lagging sectors. If their integration windows arrive on the same six-to-twelve-month lag once they spend at the threshold the paper identifies, the next vintage of this dataset is where the displacement-versus-augmentation question actually gets answered.