Repeatability is a better signal than annoyance
People often want to automate a task because it is boring. Boredom is a useful clue, but not enough. A stronger candidate is work that happens frequently, follows recognizable rules, uses structured or predictable inputs, and produces an output that can be checked. Repetition gives automation enough volume to justify its setup and maintenance.
Automate stable decisions, not unresolved disagreements.
Do not automate confusion
If people disagree about the process, exceptions are constantly negotiated, or source data is unreliable, automation can make the problem move faster without making it better. Stabilize the decision logic first. The goal is not a perfect process, but a process whose normal path and exception path are sufficiently understood.
Match automation to consequence
Low-risk tasks can tolerate more autonomy. High-impact actions should require stronger validation, approval, logging, and rollback. Sending a reminder is different from changing a customer account, approving payment, or shutting down infrastructure. The right automation level depends on the cost of being wrong, not only the technical ability to automate.
Automate an outcome, then measure it
Define what improvement should appear after automation: shorter lead time, fewer manual touches, lower error rate, faster routing, or more consistent data. If the only measure is we automated it, the project can succeed technically while adding hidden maintenance and exception work. Useful automation leaves the total system simpler.
Key idea
The best candidates combine frequency, clear rules, reliable inputs, measurable value, and manageable consequences when something goes wrong.