Dive Brief:
- OpenAI is introducing a new approach for measuring the value of AI investments as organizations shift from experimenting with the technology to scaling its use while demonstrating business impact.
- In a Friday blog post, OpenAI CFO Sarah Friar outlined a framework that measures AI based on outcomes, including the number of tasks performed successfully and the total cost of completing them.
- “For years, software was measured through adoption: seats, active users, renewals,” Friar said in a separate post on LinkedIn. “AI is different — it needs to be measured by work accomplished.”
Dive Insight:
The push comes as companies face mounting pressure to show that AI investments are delivering measurable business value.
Worldwide spending on AI is forecast to total $2.59 trillion in 2026, a 47% increase year-over-year, according to Gartner.
In a study released by PwC in January, only 12% of CEOs said AI had delivered both cost and revenue benefits. Overall, 33% of respondents reported gains in either cost or revenue, while 56% said they had so far seen no significant financial benefit.
The issue has come into sharper focus amid growing scrutiny of AI token consumption, or the amount of AI processing companies pay for when using large language models.
Palantir CEO Alex Karp recently argued that many business leaders are growing increasingly frustrated with the economics of LLMs, questioning whether rising token costs are translating into ROI.
“The enterprises are just tired of it,” Karp said in a CNBC interview.
Karp made the comments while promoting Palantir’s own technology, acknowledging that his company has a commercial interest in the debate.
In her blog post, Friar introduced what she called a “useful-intelligence-per-dollar” scorecard for AI centered on four factors: whether the technology performs meaningful work, the cost of each successful task, the reliability of its output and whether each AI dollar produces more value as usage grows.
“Tokens create value when they transform into work people can use,” Friar wrote. “As models become more capable, they can take on longer and more complex tasks: maintaining context, reasoning through multiple steps, working across tools, and adapting as they go.”
