Is it frequent enough?
Something done daily is worth automating sooner than something done quarterly, even if the quarterly task is bigger. Frequency is what turns a small saving into a real one.
Are the rules stable?
A process whose rules change weekly is a poor first candidate. Automation encodes the rules as they are today; a moving target means constant rework.
Is the data reachable?
If the inputs live somewhere with no API and no export, the cost is in getting at them, not in the automation. Check access before you scope anything.
What is the cost of a wrong answer?
Drafting is forgiving; posting to the ledger is not. Start where an error is cheap to catch, and keep a person approving anything that is expensive to reverse.
Can you measure it?
Time a handful of real runs by hand before you build. Without a baseline you cannot tell whether the result was worth it. Measure yours, and ignore anyone else's average.
Related