OpenAI's own CFO says quit counting tokens and start counting the work AI finishes
Uprovd Take When the finance chief of the company selling the tokens tells buyers to judge AI by the work it completes per dollar rather than by usage or price-per-token, she is describing the exact scorecard Uprovd was built to keep.
Read the original at AxiosOpenAI’s finance chief, Sarah Friar, used a company post this week to argue that the way most enterprises price AI - by expense or by cost per token - measures the wrong thing entirely. Her proposed yardstick, which she frames as useful intelligence per dollar, asks a different set of questions: is the system finishing work that actually matters, what does each completed task cost, and can people trust the result enough to depend on it? Coverage in Axios, Fortune and CFO Dive all zeroed in on the same pivot.
What makes this notable is the source. The pushback against usage-as-value has been coming from CFOs and analysts for months; hearing it from the vendor’s own balance-sheet owner removes the last excuse to keep reporting adoption counts. A rising token bill can even be a good sign, she suggests, but only if the volume of high-quality finished work is climbing faster than the cost.
That is Uprovd’s thesis stated almost word for word. We baseline each initiative and tie its spend to the outcomes it actually produced, so the board sees value per dollar rather than activity per seat. When the industry’s biggest model vendor starts grading AI the way we do, the argument for measuring outcomes over usage stops being contrarian and starts being consensus.
This is Uprovd's analysis of third-party reporting. Original article linked above.