Measuring AI ROI beyond the hype cycle
Fritz DesirManaging Partner · Strategic Design
AI spend is easy to grow and hard to justify. The fix isn’t a better dashboard — it’s a better question.
- Most AI ROI claims are stories, not numbers — and stories don’t survive a budget review.
- Separate what AI spent from what it changed; never claim a causal line you can’t draw.
- Start with a reconciled spend baseline — it surfaces 15–25% recoverable waste and gives every future claim a denominator.
- Track three figures that hold up: recoverable spend, real adoption, and goal contribution.
Every leadership team has been asked the same question in the last eighteen months: what is our AI actually worth? And most of them answer with a story — a demo that wowed someone, a team that “feels faster,” a vendor’s case study. Stories don’t survive a budget review.
The problem isn’t that AI has no return. It’s that the way most companies measure it is built for the hype cycle, not for accountability. Here’s a more honest way to think about it.
The attribution trap
The most common mistake is crediting AI with outcomes it merely touched. Revenue went up, AI was in the building, therefore AI drove revenue. That logic won’t hold when a CFO pushes on it — and it shouldn’t.
Honest ROI starts by separating what AI spent from what it changed, and refusing to claim a causal line you can’t draw. The goal isn’t a flattering number. It’s a defensible one.
Start with a baseline, not a guess
You can’t measure return without a starting point. Before attributing a single win to AI, reconcile what you’re spending across every tool, seat and cloud — most teams are off by a wide margin because spend hides in personal cards and shadow tools.
That baseline does two things: it surfaces waste you can cut today (a guaranteed return), and it gives every future claim a denominator. ROI with no denominator is just a press release.
“You can’t manage — or defend — a number no one can actually see.”
Three numbers that actually hold up
Skip the vanity metrics. Three figures survive scrutiny: recoverable spend (idle licenses, duplicate tools), adoption (real usage vs. what you pay for), and goal contribution (spend mapped to a KPI with a baseline and a target). The first pays for the program; the other two tell you whether to keep investing.
If you can’t state your savings as a range, name the methodology behind it, and point to the baseline it’s measured from — it isn’t a number yet, it’s a hope. Ranges with methodology beat precise-looking guesses every time.
From measurement to governance
Measurement is the beginning, not the end. The teams that win don’t just see their AI spend — they govern it: owners on every tool, renewals tied to usage, and spend steered toward the goals that move the business.
That’s the whole point of measuring honestly. Not to produce a number for a slide, but to earn the right to make the next AI investment with confidence.
