Measure the current state honestly
Two weeks of observed timings across the real distribution of cases — including exceptions — produces a baseline that survives challenge. Averages taken from a manager's estimate do not.
Three benefit categories
Separate them; they have different confidence levels and different owners.
- Labour capacity released, with a named redeployment plan
- Error and rework cost avoided, priced from actual incident history
- Cycle-time value, counted only where a real constraint moves
Cost the whole life
Include build, integration, change management, and ongoing supervision and maintenance. Automation that nobody owns after go-live degrades quietly and takes the business case with it.