For me, the real test is after launch. In a demo every bot looks great, but once actual customers start asking messy, unexpected questions, it can easily create more work for support. That’s why I’d look beyond just ticket reduction and track real resolution rate, how often it hands off to a human, how many answers need correcting, and how much time we spend maintaining it. A 70% automation rate doesn’t mean much if your team has to clean up the other 30%.