OpenAI's Sarah Friar publishes an 'AI scorecard' for measuring ROI

OpenAI CFO Sarah Friar published 'A Scorecard for the AI Age,' proposing a practical framework for organizations to measure AI's return on investment. The scorecard tracks four dimensions: useful work delivered, cost per successful task, dependability, and return on compute — an explicit push to reframe how enterprises judge AI value.

The substance addresses a real pain point: as AI spend balloons, boards and CFOs struggle to justify it against vanity metrics like tokens consumed or seats deployed. By centering 'cost per successful task' and 'dependability,' Friar is steering the conversation toward outcomes an accountant can defend, which conveniently favors capable frontier models that complete tasks reliably over cheaper models that fail more often — a subtle argument for OpenAI's premium positioning against low-cost rivals like DeepSeek.

The framing echoes a broader industry moment: Boris Cherny, creator of Claude Code, offered his own four-step framework this week for measuring enterprise AI success beyond raw token burn, and NVIDIA is marketing 'intelligence per dollar.' The whole sector is converging on ROI storytelling because the era of blank-check AI budgets is ending and buyers demand accountability.

Skeptics will note the obvious incentive: a scorecard authored by OpenAI's CFO that rewards 'return on compute' and 'successful tasks' is not a neutral yardstick — it's a sales instrument dressed as thought leadership. Still, the metrics are genuinely useful and likely to be adopted or adapted by enterprises. Watch whether independent analysts or competitors publish rival frameworks, and whether 'cost per successful task' becomes a standard procurement line item.