OpenAI is moving to reframe how companies judge the value of AI just as finance chiefs start to push back on ballooning model costs and unclear returns.

Early concerns over AI’s runaway spending

In recent months, CFOs have discovered just how quickly AI experiments can turn into massive bills, including an accidental half‑billion‑dollar charge at one firm, sharpening scrutiny of AI spending and ROI.¹ OpenAI CEO Sam Altman has acknowledged that costs are now the second biggest concern customers raise, after the challenge of deploying AI across their organizations.¹

OpenAI’s scorecard: ‘Useful Intelligence per Dollar’

Against this backdrop, OpenAI CFO Sarah Friar published an essay outlining what she calls a “scorecard for the age of AI,” arguing that traditional software metrics like seats purchased or active users are inadequate and that “understanding the value of AI demands a more powerful measure: work accomplished.”²

Friar proposes a central metric, “Useful Intelligence per Dollar,” focused on whether “the value of the work AI completes grows faster than the cost of producing it.”² She urges leaders to look beyond cost per token and instead ask four questions: Is AI completing work that matters, what does each successful task really cost, can people depend on the result, and do AI economics improve as usage scales?²

Human vs. AI framing of value

Axios characterizes the framework as OpenAI’s response to an “enterprise AI cost reckoning,” noting that companies are increasingly routing work to cheaper models and demanding clearer returns.¹ The outlet emphasizes that, by shifting attention from sticker price to completed work, OpenAI can argue that its most capable — and more expensive — models “ultimately deliver better economics.”¹

Friar, meanwhile, stresses concrete outcomes: customer issues resolved, code shipped, contracts reviewed and “tokens [that] create value when they transform into work people can use,” positioning the metric as a way for skeptical CFOs to quantify whether AI output is truly outpacing its cost.²