Skip to content
Subscribe
Rafael Cordeiro, colunista

AI ROI and the work that used to never happen

Rafael Cordeiro · 2 min read
Share

Part 2 of 2 on AI ROI. Part 1 covers counting tokens and proving value.

We may be measuring only the least interesting part of AI's ROI.

The obvious metrics

When companies try to justify their AI investments, they usually start with the most obvious metrics. Hours saved, costs cut, productivity per employee, tokens consumed.

All of that matters.

The work now within reach

But a recent OpenAI article, The Work Now Within Reach, raises a more interesting question. What happens when AI does not just make the work we already did cheaper, but makes economically possible work that used to never happen?

That difference matters.

Picture a company that can now analyze every contract before renewal. Personalize every sales proposal. Investigate thousands of cost-cutting opportunities. Follow every customer interaction. Simulate scenarios before major decisions.

Perhaps none of this was impossible before AI. It was simply too expensive to do at scale.

Editor's context. The article, written by Sarah Friar and published on September 8, 2026, argues that better models and more efficient compute make feasible work that used to require too much time, cost or expertise.

Two ways to create value

And that changes the ROI question. Not just how much work AI automated, but what result came into existence because we now have AI.

There are two very different ways to create value.

  1. Doing what we already did better, faster and cheaper.
  2. Doing something that simply was not worth doing before.

The second strikes me as far more interesting.

When agents can research, analyze and execute at a much lower marginal cost, a huge amount of work that used to sit below the line of economic viability can cross it.

And that changes the role of AI inside the company. It stops being just a productivity tool and starts expanding the frontier of what the organization can do.

Measuring outcomes and incrementality

That is why I believe the next conversation about AI ROI needs to go beyond adoption, tokens and hours saved.

We need to measure outcomes and incrementality. Not just how much the intelligence cost, but what value only came into existence because of it.

In short

  • We may be measuring only the least interesting part of AI's ROI.
  • There is value in doing better what we already did and in doing what was not worth doing before.
  • We need to measure what value only came into existence because of AI.

Further reading

Originally published on LinkedIn on September 11, 2026. See the original post. This version has been revised and expanded with context and sources. The views are those of the author, a co-founder of OperOS AI.

Read next