Most of the AI adoption dashboards I've seen measure activity: logins, prompts sent, seats licensed, hours theoretically saved based on some assumed baseline. Every one of those numbers can go up while the organization gets no faster at anything that matters.
None of the standard ones ask the harder question: did more finished value reach a customer or a decision point because of this?
I've started sketching a different scorecard, and it's a lot shorter than the activity version, mostly because the honest metrics are harder to fake and harder to game. Constraint cycle time. Decision-to-action time. Completed outcome rate, not draft or intermediate output. Queue growth — whether AI is quietly creating more work-in-process downstream than it's clearing.
None of these are metrics you can improve just by using the tool more. They require actually tracing whether the work moved, which is exactly why most organizations default to the easier, activity-based version instead.
I don't think that's cynicism about AI. I think it's an honest description of why good measurement is harder to build than good software, and why most companies end up with the software before they have the measurement.
Current Hypothesis
A throughput-honest scorecard has fewer, harder-to-game metrics than an activity-honest one, which is exactly why organizations avoid building it.
Not a conclusion — a working idea, revisited as evidence comes in.