Useful Intelligence Per Dollar: The New AI ROI Metric Every Enterprise Needs to Know

Enterprise leaders have poured billions into AI tools, but many still struggle to answer a simple question: Is it actually paying off? On July 17, 2026, OpenAI CFO Sarah Friar introduced a new measurement framework built around the concept of "useful intelligence per dollar" — a metric designed to evaluate the economic value AI tasks generate relative to their total cost, rather than relying on legacy software proxies like seats purchased, active users, or simple token pricing .

"For years, software was measured through adoption: seats, active users, renewals," Friar said in a LinkedIn post. "AI is different — it needs to be measured by work accomplished" . The framework explicitly argues that the lowest cost per token does not always produce the lowest cost per outcome, and that enterprises should optimize for total value creation rather than unit price .

The Four-Question AI ROI Scorecard

Friar's methodology replaces vague adoption metrics with four concrete questions that form a scorecard for AI ROI :

1. Is AI completing work that matters?

Measure the volume of real, outcome-based tasks — customer issues resolved, code changes shipped, contracts reviewed — rather than raw usage or prompt counts . "Tokens create value when they transform into work people can use," Friar wrote . This shifts focus from activity metrics (how many queries were made) to impact metrics (what got done).

2. What does each successful task cost?

Calculate the full cost per task that meets quality standards, including AI inference, retries, human review, and rework . A cheap model that fails often can cost more per successful outcome than a premium model that gets it right in one pass . This question forces enterprises to look beyond the per-token price and account for the hidden costs of errors and corrections.

3. How often does AI get the work right?

Track dependability — the rate at which outputs are "ready to use" versus requiring correction or escalation . Fewer human interventions mean lower total cost and higher trust. This is essentially a quality-assurance metric for AI outputs, recognizing that reliability is as important as raw capability.

4. Does each AI dollar produce more value as usage grows?

Monitor whether AI completes more high-quality work over time without exponential cost growth — a concept also called return on compute (ROC), borrowed from semiconductor and data center finance . This question addresses whether AI investments have compounding returns or diminishing marginal value, a critical consideration for scaling decisions.

How to Implement the Framework

The methodology is designed to be auditable with existing tooling . Teams define what "done" means for a specific workflow, count successful completions, sum the full cost (including human overhead), and track pass rates and quality trends .

Friar's approach treats AI as a productive asset rather than a software license, and provides a structured way for CFOs and technology leaders to answer the question she hears most often from her peers: "How do we get more value from our AI spend?" .

Why This Framework Matters Now

The proposal comes as enterprise AI spending has surged, with OpenAI itself reporting $1 billion in monthly revenue and over 500 million enterprise seats sold . Yet many organizations lack clear ROI frameworks, creating a "mismatch between AI's abilities and the value companies are capturing" . Friar previously described this disconnect as "capability overhang" — the gap between what AI can do and what organizations are actually able to harness for business value .

By grounding the four questions in auditable cost and quality data, the "useful intelligence per dollar" framework gives enterprises a practical tool to move from faith-based AI investment to evidence-based decision making.