Boards asked for AI ROI in mid-2026; teams that only reported “prompts run” lost credibility. A good dashboard ties each AI step to a business outcome.

For your main workflow three numbers are enough: volume completed, average handle time, and the share of outputs humans still edit. If human edits stay high, the model helps but has not replaced the step.

Avoid vanity metrics. “Hours saved” only counts when those hours became other work or lower overtime.

Review the dashboard weekly; a monthly cadence lets drift and quiet waste grow unnoticed.

Manager action: For your main AI workflow, log weekly: volume, average handle time, and % of outputs edited by humans.