If your AI pilot deck still measures "time saved," the model is ahead of the business case.
Time saved is a page-3 metric. It tells you a person clicked fewer buttons. It doesn't tell you whether a single decision reached the system of record faster, cheaper, or with better information than last quarter.
For CFOs and operators actually trying to write AI into the P&L, the useful questions look different:
- How many decisions per week now execute without a human in the loop?
- What's the cycle time from signal detected to state changed downstream?
- Where is the agent still waiting for data that should already exist?
- How often does reality drift from the decision the agent recorded — and how do you know?
These are operating metrics, not AI metrics. That's the point. The pilots that graduate to production are the ones that stop being measured as AI experiments and start being measured as business processes with a new kind of worker inside them.
The quiet truth of the current cycle: most "AI ROI" slides are measuring the activity of the tool, not the behaviour of the system. Until that shifts, boards will keep approving pilots and quietly wondering why the operating ratios don't move.