OpenAI proposes an AI scorecard for measuring useful work

OpenAI has published a practical scorecard for evaluating AI returns, introduced by CFO Sarah Friar. The framework focuses on useful work, cost per successful task, dependability, and return on compute.

The emphasis is notable because many AI deployments still rely on broad productivity claims rather than operational metrics. A scorecard built around completed tasks and reliability can help teams compare pilots with production systems.

As AI spending grows, buyers are likely to demand more evidence that models and agents are producing measurable business value rather than just usage.