Measuring AI ROI & why you shouldn't do it

Most AI strategy work starts with the same question: which use cases have the highest ROI? That's a reasonable procurement question. It's a bad transformation question.

Think about how a founder would approach this: she would not rank AI opportunities by payback period. She would work backward from what the business needs to look like in three to five years — and figure out what it takes to get there, even when several steps along the way cost more than they return.

The ROI-first framing has a specific failure mode. You approve the top use cases, capture some efficiency gains, and then nothing compounds. This list was built on how the business works today. It says nothing about where it needs to go.

The founder question is different: what does this business need to become, and what would have to be true to get there? Some of those steps will be negative ROI in isolation. That's expected when you're building a capability rather than optimizing a process. A strategy team transitioning from quarterly research cycles to continuous competitive intelligence isn't evaluating the ROI of each component separately. The intermediate investments — tooling, workflow redesign, new skills — are part of the path.

Process optimization and transformation are different exercises. Ranking use cases by ROI works fine for the first one. For the second, you need a coherent theory of where the business is going and a plan for getting there.

The organizations that look different in five years aren't the ones that approved the highest-ROI use cases in 2026. They're the ones that decided what they needed to become and started building — even when the early steps didn't justify themselves on their own.