OpenAI CFO Sarah Friar proposes a practical AI scorecard centered on "Useful Intelligence per Dollar" to measure ROI beyond simple adoption metrics. The framework answers four questions: First, how much useful work gets done—tracking resolved issues, shipped code, or reviewed contracts per workflow. Second, what each successful task actually costs, factoring compute, human review, retries, and rework; GPT‑5.6's three tiers (Sol, Terra, Luna) let customers optimize this equation, with Sol achieving 72.7% on the Artificial Analysis Coding Agent Index using 54% fewer output tokens at 36.2% lower estimated API cost than a leading rival.

Third, dependability—measuring outcomes as ready-to-use, needs-correction, or needs-escalation to gauge whether AI genuinely reduces workload. ChatGPT Work builds on ChatGPT Enterprise security to enable deeper workflow access with oversight. Fourth, whether each AI dollar buys more work as usage scales—tracking cost per successful task over time. Friar argues compute efficiency and model capability must improve together so organizations complete more valuable work while per-task costs fall.