Most AI talent reviews I sit in count the wrong thing.
Executives walk through headcount growth and retention like it is the whole story. The market is pricing something else.
The dashboard that looked healthy
A CHRO at a European B2B SaaS firm pulled up her talent slide for me last month. Forty AI hires in the year. Retention above the sector average. The chart climbed the way she wanted it to climb.
I asked one question. Where did the forty come from, and where did the twelve leavers go.
She paused. Then she opened two more tabs, scrolled, and admitted she did not know. Not firm by firm. Not even cluster by cluster. Her dashboard tracked the stock. The flow was invisible.
Every talent board I see is built this way. A count on the vertical axis, a date on the horizontal, nothing about where any of these people arrived from or where any of them went. Position in the hire-flow is the number I never see. It is also the number sophisticated capital is already watching.
What the research says
A new MIT-led preprint reconstructs the full interfirm AI talent flow network from Revelio Labs online professional profiles. About 535 million employment records across 58 countries. 23,596 listed firms, 15,391 matched to Orbis financials. The regression panel is 20,413 firm-years from 3,329 firms, 2010 to 2022.
The headline, within firm, with firm and year fixed effects and controls for workforce size and assets.
A one standard-deviation gain in standardized log AI PageRank is "associated with 0.023 log points (about 2.3%) higher enterprise value in the following year."
Standard error 0.007. Significant at the 1% level. The EBITDA coefficient is +0.013, p=0.078. Not significant at 5%. Those two numbers tell the same story twice. The market is pricing position before earnings show it.
A second finding cuts against intuition. The firms at the AI network core change hands far more often than non-AI leaders. US core retention is 72.6% for AI versus 90.5% for non-AI. In Southeast Asia it drops to 32.0% versus 79.8%. The ladder is still being climbed.
Caveats I would name out loud. The preprint is four days old and has no peer review. The +2.3% is associational within firm. The event-study is the closest causal cut. The EBITDA null I am reading as a signal of forward-looking pricing rather than a quiet failure.
Three moves I would make this quarter. Report a flow metric next to the headcount metric in the talent pack. Give it to investor relations as well as HR. In regions where core leadership turns over fast, treat the incumbents as a window into the top rung rather than a durable moat.
The AI talent dashboard most firms built in 2024 answers a question investors stopped asking.