"Should we automate this?" is usually the wrong first question, because almost anything can be automated with enough effort. The better question is whether automating it is worth the effort, and that depends far more on the shape of the task than on the technology behind it.
The tasks worth automating share a pattern
Automation earns its cost when a task is high-volume, well-defined, and repetitive, with inputs and outputs that don't change much from one instance to the next. Categorizing incoming support requests into a handful of known types. Reconciling records between two systems that should agree but drift apart. Pulling numbers out of a spreadsheet each week to generate the same report. These tasks are boring precisely because they're consistent, and that consistency is what makes them automatable.
The test is simple: if you handed this task to five different competent people, would they all produce roughly the same result? If yes, it's a strong candidate. If the "right" answer depends heavily on judgment, context, or a relationship with the person on the other end, that's a sign the task needs a human doing the deciding, even if a tool helps them do it faster.
Where it falls apart
Low-volume tasks rarely justify automation, even when they're repetitive in theory. If something happens twice a month, the time spent building and maintaining an automated process can easily exceed the time it would have taken to just do it by hand. Automation has a setup cost and an ongoing maintenance cost, and both need to be smaller than what they're replacing.
Ambiguous judgment calls are the other place automation struggles, especially when the cost of a wrong call is high. A system that auto-approves routine requests can save real time. The same system making judgment calls on exceptions, refunds, disputes, anything where the "correct" answer depends on context a rule can't fully capture, tends to create more cleanup work than it saves, plus a worse experience for whoever's on the receiving end of a bad automated decision.
What this looks like in practice
The businesses that get real value from automation usually start narrow: one specific, well-understood task, with a clear definition of correct and a person still in the loop to handle the exceptions the system wasn't built for. That's a very different starting point from "let's use AI to handle customer service" or "automate our operations," which sound like plans but aren't specific enough to actually build or evaluate.
If you're not sure whether a process is a good candidate, the useful exercise is writing down the rule a human currently follows to do it. If that rule is short and consistent, you likely have something worth automating. If it takes a page to explain and still has exceptions, you have a process that needs custom software to support the people doing it, not a system meant to replace their judgment entirely.