When you should NOT automate a process
Automation can make a bad process fail faster. Check stability, input quality, judgement and ownership first.
TL;DR: Do not automate a process merely because it feels repetitive. First check whether it is stable, understood, frequent enough, based on clean inputs, governed by rules and owned after handover.
Seven reasons to pause
Pause when the process changes every week, nobody can perform it manually with confidence, the annual volume is too low to justify build and maintenance, source data is inconsistent, the decision depends on judgement rather than rules, nobody owns maintenance, or the project exists only because another company uses automation.
For low-frequency work, measure the actual hours spent. A process taking fewer than 10 hours a year is usually not worth automating unless its errors are expensive or risky.
What to do first
Stabilise the process, write the manual steps down, standardise the source of truth, and separate data preparation from the final human decision. Automation should prepare evidence and execute clear rules; it should not hide an undefined judgement behind a button.
Ready for implementation?
Proceed when the workflow is stable, inputs are standardised, rules are explicit, repetition creates measurable cost, and an owner can maintain and replay it. Start with one bounded scope and define what success and rollback mean before expanding.