The company selling AI productivity tools just published a 69-page report that can't prove they improve productivity.
The Summary
- OpenAI released research on enterprise ChatGPT adoption that found zero correlation between AI use and revenue per employee
- The 69-page report sidesteps the ROI question entirely, an existential issue for OpenAI's corporate business model
- If enterprise AI adoption can't move the needle on the metrics CFOs actually care about, the whole agent economy thesis has a measurement problem
The Signal
OpenAI just published a comprehensive study on how companies use ChatGPT Enterprise, and buried in 69 pages of adoption metrics and use cases is a stunning admission: they found no correlation between AI tool usage and revenue per employee. For a company betting its future on selling AI to the Fortune 500, this is the data equivalent of a shrug emoji.
The report focuses on everything except the thing that matters. Adoption rates, user engagement, prompt patterns, sentiment surveys. All valid metrics for a product team. None of them answer what a CFO buying enterprise software needs to know: does this make my company more money per person, or not?
"The AI lab grapples with a question that's existential for its future: Do ChatGPT corporate customers get a clear ROI?"
Here's what makes this notable: OpenAI didn't have to publish this. They could have released a marketing piece full of cherry-picked testimonials and vague productivity wins. Instead, they shipped research that inadvertently reveals the measurement crisis at the heart of enterprise AI adoption. Companies are deploying these tools at scale without knowing if they work.
This isn't necessarily proof that AI tools don't boost productivity. It might mean:
- The gains are real but diffuse across qualitative improvements that don't show up in revenue per employee
- The time horizon is wrong and measurable ROI takes longer than the study window
- Revenue per employee is the wrong metric for knowledge work output
- Early adopters are still figuring out how to use these tools effectively
Or it might mean that most enterprise AI use is theater. Employees use ChatGPT to rewrite emails and summarize documents, managers check the "AI transformation" box on their OKRs, and nothing fundamental changes about how the company makes money. The productivity paradox from the PC era all over again, just with better UX.
The Implication
If you're selling AI tools to enterprises, this research says measurement is now the product category to watch. Companies will pay for observability, attribution, and ROI tracking before they pay for another general-purpose chatbot. The winners in the next wave won't be the companies with the best models but the ones who can prove their models made the customer money.
For companies buying AI tools, stop measuring adoption and start measuring outcomes. Revenue per employee might not be the right metric, but something has to be. Figure out what output actually matters in your business, instrument it, and only pay for AI that moves that number. Otherwise you're just renting very expensive faith.