Application teams often arrive with twenty charts and leave our workshops with four or five. The cut is rarely about sophistication; it is about which numbers still matter when everyone is tired on a Friday afternoon.
Lead time from ready to production
Measure the elapsed time from “ready for release” to production for the applications you actually own. Exclude speculative backlog age. When lead time stretches, the conversation stays concrete: batch size, environment contention, or approval queues.
Change failure rate
Count releases that required a hotfix, rollback, or emergency patch within a defined window. Argue about the window once, write it down, and stop re-litigating it mid-review.
Priority-one incident hours
Hours spent in P1 response reveal load that ticket counts hide. Pair the figure with a short narrative of the dominant failure mode that month.
Feature adoption at fourteen days
For product-facing applications, track whether a released capability was used by the intended cohort within fourteen days. It keeps delivery conversations honest when “shipped” and “used” diverge.
These four are a starting pattern, not a prescription. A warehouse control system may swap adoption for inventory reconcile lag. The principle holds: if the KPI cannot change a decision in the room, it does not belong on the scorecard.