Most automation projects fail because they start with the tool, not with the problem. Somebody sees a demo, buys a licence, and then looks for a place to apply it.
Before automating anything I ask three questions. What exactly hurts? Who feels it? What does “fixed” look like? If the answers are vague, the automation will be vague too.
In customer service, two things are worth doing before anything else.
The first is measuring time. If you cannot say how long a complaint takes from the first message to the resolution, you cannot improve it. We set up automatic alerts when a case sat too long at any stage, and that alone cut resolution time by half to two thirds. Nobody worked faster. People simply saw where things were stuck.
The second is routing. The biggest waste of time in service is the question “who handles this?”. Once cases were assigned automatically by category, priority and team load, that question disappeared from the chat.
Good automation is invisible. The team should not think about it. They should just notice, after a while, that things work.