Automating a bad process gives you a fast bad process, permanently, in software that's expensive to change. The cheapest improvement available is almost always removing steps rather than accelerating them.
Most processes have accumulated approvals and checks that were added in response to a specific incident years ago. Nobody removes them, because nobody is quite sure why they're there.
The real path work takes, gathered by watching it happen rather than by reading the documented procedure, which is usually aspirational.
Cycle time, wait time, error rate, and rework, captured before anything changes — otherwise improvement is a matter of opinion afterwards.
Finding the actual constraint. Effort spent anywhere else in a process produces no improvement at all, which is counter-intuitive enough to be routinely ignored.
Steps eliminated, combined, or resequenced, with each remaining approval justified by someone willing to say what it's for.
Designed with the team who runs it, because a process improvement they didn't shape gets quietly worked around within a month.
Trialled in one team, measured against the baseline, adjusted, and only then rolled out more widely.
This is the sequence that gets reversed most often, usually because automation has a budget line and simplification doesn't.
Once the process is worth keeping, automating it is covered under business process automation.
Name the process that frustrates everyone. We'll map it, measure it, and show you how much of it could simply stop happening.