The ROI Question That Won't Go Away
OpenAI published a 69-page report on August 11 detailing enterprise adoption of ChatGPT. The headline narrative was bullish: exponential usage growth across seniority levels, a 'frontier gap' opening between AI adopters and laggards, and a clear warning that companies not deploying AI agents are falling behind.
Then there is page 35.
In one small table, the researchers report no statistically significant correlation between revenue per employee and how intensively those employees use AI — measured in messages sent and tokens consumed. The report states it plainly: 'Revenue per employee is not meaningfully associated with output tokens per employee or messages per active user once other controls are included.'
The finding does not say AI is worthless. It says that companies with already-high revenue per employee tend to be early ChatGPT adopters, and that heavier AI users tend to come from more lucrative firms in general. What the study does not establish is that using AI more produces more revenue. Correlation runs in one direction; causation is absent in both.
Who Is Actually Using It — and Who Is Judging the Returns
A second buried data point compounds the problem. A graph on page 29 shows that the most senior employees are the least intensive users, logging the fewest weekly messages per user. Early-career employees dominate usage by a wide margin.
OpenAI CFO Sarah Friar acknowledged the gap in a LinkedIn post about the report: 'For leaders, that's a reminder that competitive advantage comes from the people closest to the work. Listen to them, learn from them, and help the rest of the organization catch up.'
The implication is uncomfortable for the enterprise sales pitch: the executives signing the contracts are the employees least equipped to measure what those contracts are delivering.
A Flat Line, Then a Sprint
The report's usage graph adds another wrinkle. Total output token growth across OpenAI's enterprise base went nearly flat from approximately October 2025 through December 2025 — a period when Anthropic's Claude Code was gaining significant traction in corporate environments. Growth resumed sharply in January 2026, and one venture capitalist quoted in Fortune described OpenAI's 2026 run rate as 'pretty incredible.' The graph ends at March 2026.
On the same day the report drew scrutiny, OpenAI announced it had hired Dali Rajic as its new Chief Revenue Officer, replacing Denise Dresser, who had held the role for less than one year. Rajic's mandate, per the company, is accelerating customer adoption and helping businesses measure impact — precisely the gap the report exposes — as OpenAI moves toward its IPO.
A credibility footnote also warrants attention. Two of the report's five authors — David Holtz of Columbia Business School and Prasanna Tambe of Wharton — carry academic affiliations on the cover page. A footnote clarifies that both 'contributed to this work in their capacity as paid contractors for OpenAI.' The arrangement is not hidden, but it blurs the independence that academic co-authorship typically signals.
The Market Has Not Yet Voted on ROI
Capital rewards clear rules, and right now the rules around AI's return on investment are anything but clear. Enterprises are spending; the productivity gains are not yet legible in the numbers that matter — revenue per employee. OpenAI deserves credit for publishing the finding at all rather than burying it entirely. But the same-day CRO replacement signals that the company knows the measurement problem is existential, not academic. Until enterprise buyers can point to a line on their own income statement that moves with AI spend, the trillion-dollar valuations circling this industry rest on a foundation that one inconvenient table just made a little less solid.
