OpenAI CFO Sarah Friar's AI scorecard measures ROI on compute

OpenAI CFO Sarah Friar has unveiled a pragmatic AI scorecard designed to help enterprises measure the real return on their AI investments. The framework moves beyond vague benchmarks and focuses on four core metrics: useful work accomplished (e.g., number of tasks completed or time saved), cost per successful task (including compute, API calls, and human oversight), dependability (accuracy, consistency, and uptime), and return on compute (value generated per unit of compute spent). By standardizing these metrics, the scorecard aims to answer the CFO's perennial question: is our AI spend actually generating productivity gains or just adding cost? Friar's proposal comes as enterprises pour billions into AI, yet many struggle to quantify its business impact.

The scorecard is particularly timely given the rising costs of large-scale AI inference. For example, a company deploying a customer support chatbot could track tasks resolved per dollar, human escalations avoided, and uptime reliability. Return on compute forces teams to optimize for efficiency—choosing smaller models or caching responses when possible. Friar emphasizes that the scorecard should be adapted per use case but insists on consistency across deployments. This framework could become a de facto standard as CFOs demand accountability, much like SaaS metrics became the norm for cloud spending. By tying AI spend directly to business outcomes, the scorecard helps companies avoid the hype trap and focus on high-ROI applications.