Measure the starting point.
Record how often the process runs, how long the manual steps take and how much time is spent correcting mistakes. Use a representative period and distinguish active work from waiting time.
Pick a small set of outcomes.
Useful measures can include time spent per case, response time, rework and completion rate. Choose measures your team can collect consistently and agree on what improvement would be meaningful.
Include the ongoing effort.
Count software costs, review time and maintenance alongside the time saved. An automation that needs frequent manual correction should be assessed with that effort included.
Review and refine.
Compare similar cases before and after launch. Discuss the results with the people using the workflow and record changes in volume or scope that might explain the difference.