Frequency beats sophistication. A task performed forty times a week returns your investment long before an impressive project that runs twice a month.
The second test is whether the task needs an opinion. If two competent colleagues would produce the same result, software can produce it too. If they would disagree, keep a person in the loop and automate only the preparation around the decision.
The third test is what happens when it fails. Start where a mistake is visible and cheap to correct, so your team builds confidence before anything touches money or clients directly.
Write down the hours a task consumes before you change it. Without that number, nobody can tell six months later whether the automation was worth keeping.