OpenAI CFO's Four-Question Framework for Measuring AI ROI

OpenAI CFO Sarah Friar has released a framework for evaluating whether artificial intelligence investments deliver genuine economic returns, marking a shift from how enterprises have traditionally measured software value.

In a blog post, Friar contends that AI demands different metrics than conventional software, where success has historically been gauged through seat licenses, active users, and renewal rates. Instead, she proposes measuring AI by the actual work it completes.

The useful intelligence per dollar framework

Friar's approach centers on what she calls "useful intelligence per dollar," built on four core questions:

• Is the AI completing work that genuinely matters to the business?
• What does each successful task actually cost?
• Can users reliably depend on the results?
• Does each dollar invested produce increasing value as usage scales?

To implement this framework, Friar recommends organizations track the volume of AI-completed work that meets defined quality standards, calculate the total cost of producing that work, and divide by the number of successful tasks. The critical test is whether high-quality output grows faster than total costs while maintaining or improving quality standards.

"The basic economic question facing CFOs and other business leaders is whether the value of the work AI completes grows faster than the cost of producing it," Friar wrote.

Why it matters

This framework arrives as CFOs increasingly shape AI strategy alongside CEOs, not just manage budgets. McKinsey's recent Global CFO Forum revealed that roughly two-thirds of finance chiefs now oversee strategy functions—up from less than one-third five years ago. As enterprises move from AI experimentation to transformation, CFOs need concrete methods to justify infrastructure investments that can reach hundreds of billions of dollars.

Compute as strategic asset

For OpenAI, compute infrastructure represents a strategic asset rather than merely an operating expense. The company's Stargate initiative, announced in January 2025, outlined plans to invest up to $500 billion over approximately four years building AI infrastructure in the United States. The initial phase targeted roughly $100 billion, with broader buildout accelerating toward a 10-gigawatt capacity goal by 2029.

Just over a year into the initiative, OpenAI has already exceeded that milestone as demand continues accelerating. The company, currently valued at $852 billion and approaching the $1 trillion threshold, could pursue an IPO as early as this summer or as late as 2027.

Andy West, a senior partner at McKinsey who co-leads the firm's Strategy and Corporate Finance practice, told Fortune that CFO conversations about AI have evolved dramatically. While finance leaders were still experimenting with AI last year, this year's discussions have shifted decisively toward enterprise-wide transformation.

The details were first reported by Fortune.