Exceptions are the process
The happy path is easy to automate. What decides whether an automation survives is what happens to the cases that do not fit.
Every process map has a clean central line and a cloud of arrows around it labelled "unless". The central line is a few days of engineering. The cloud is the project.
Count them first
Before automating, log a month of exceptions: what they were, who handled them, how long they took. In most operations a handful of exception types account for most of the effort — and half of those are rules nobody had written down.
Route, do not block
An automation that stops when it meets an unknown case trains people to bypass it. One that routes the case to the right person, with everything they need on one screen, trains people to trust it.
A good automation is judged by its worst day, not its best.
Feed the exceptions back
Every handled exception is a candidate rule. Review them monthly; promote the stable ones into the system, and leave genuine judgement with people.