I think the key issue here is that the people deciding how long the escalation path is isn't the humans (a fair few people do opt into search company FAQs for their answer before dialing a hotline - you're robbing those people in particular of their time by forcing them back through the same FAQ steps and discouraging the usage of those opt-in low cost resources) and, right now, consumer protection and rights are at an all time low so a fair number of AI rollouts have been downright customer malicious.

It absolutely has a place in the system - but that place (in the companies that do it well) starts by giving call center employees access to the AI as a fallback when they don't know an answer and reduces the amount of information and product specialization needed. Assuming it is ranked highly by internal teams then you can consider shifting it from being an internal tool to one exposed externally - instead, in a lot of cases, companies have just switched off the ability to dial in without going through the AI hoops and, in the worst cases, if there's a tech issue where the call center disconnects from the customer, the customer is forced to go through all those hoops again.

I like to emphasize that AI is a tool - it can be applied well in a considered and thoughtful manner - or it can be rolled out to every conceivable usage with reckless abandon... we're in a place where number two is the dominant approach.

The escalation path is always too arduous though, because most people still prefer to talk to a human when they’ve got to the point of opening a chat window. You’ve always got to jump through a bunch of hoops which are basically answering yes when asked whether you’ve tried reading the website.