DECARThe instinct is to automate anything repetitive. The better filter is frequency times consequence: a task done fifty times a day with low stakes each time is worth automating even if a mistake is cheap, because the time saved compounds fast.
A task done twice a month with a costly mistake if it's wrong is a worse automation candidate than it looks — the volume is too low to justify the build, and a rare failure in an automated process is harder to catch than a rare failure a human was already double-checking.
Before we recommend automating anything, we ask what happens the day it silently breaks. If the answer is 'someone notices within the hour,' build it. If the answer is 'nobody would know for weeks,' it needs monitoring built in before it needs anything else.