The Activity Trap

The Activity Trap

61% of leaders say they're under more pressure now to prove AI ROI than they were a year ago. Only 23% can. I don't find that surprising. I find it accurate.

I've been in the room with clients who have AI budgets, AI dashboards, and AI steering committees. What they don't have is a P&L line that moved because of any of it (other than the cost line going up). Demos that impressed the board. Pilots that impressed the executive. Metric dashboards that track how many people opened the chatbot. None of that is ROI. It's activity, and activity and outcomes aren't the same, even though we measure them interchangeably.

Most organizations end up here not because of laziness. Measuring activity is easy; measuring outcomes is hard. When someone approved the AI spend, they approved it on a promise about what it would produce. Defining the outcome metric would have forced a real conversation about what "producing that" actually means and who owns it. That conversation is uncomfortable enough that many AI programs have avoided it since the money showed up. Figuring out how to "defuzz" the blurry ROI after we've seen some things is post-hoc rationalization and not super helpful when you're trying to figure out true value.

The organizations I've seen actually land AI ROI did one thing differently early: they tied each initiative to a specific outcome with a dollar figure and a named owner. Not a committee. One person. Who is it? They didn't wait until the pilot was live and the costs were sunk to ask what the business was supposed to get out of it. Or figure out the multi-step path back to the original thing they hoped to do when they started. They asked first, built second, and then verified it did what it was supposed to.

The pressure to show AI ROI isn't going away. If anything, 2026 is the year the "we're still learning" answer stops working in board conversations. What does your AI ROI story sound like right now? If the honest answer is "we've had a lot of activity," that's the answer worth changing before someone else asks it for you.