Overview
- OpenAI CFO Sarah Friar published the public scorecard, which was posted on Friday, July 17, 2026, and centers on a metric she calls useful intelligence per dollar.
- The metric rests on four clear questions: does AI complete work that matters, what is the full cost per successful task, can people depend on the outputs, and does value scale faster than cost.
- Friar recommends firms track the volume of AI‑completed work that meets a quality bar, include retries and human oversight in full cost accounting, and divide total cost by successful tasks to get cost‑per‑task.
- OpenAI frames compute as a strategic asset tied to the test and notes its large infrastructure push under the Stargate plan, while the scorecard could force vendors to prove cost‑per‑task and accuracy or lose enterprise business.
- If widely adopted the scorecard will push CFOs to expand AI FinOps and governance, reshape procurement choices between centralized and decentralized providers, and change how enterprises fund and measure AI projects.