Good experiences are hard to come by. A lot of times it'll be client driven and that requires a lot of education.
Really the motto should be:
If a problem is better solved by a human, user should be directed to a human.
But what tends to happen is the client thinks a bot will automagically solve a user's problem. That's not the case.
I think one of the first things to understand in customer service (while not always true) is that the customer is always right.
In saying as much, that should mean, in general, that you can encourage a customer to avail themselves of certain options, but by and large, you shouldn't 'force' it.
This isn't just important to avoid backlash; its important because it allows you to test an idea/tech/product with people who are willing adopters, and probably better suited to it on average than those who will avoid it, given a choice.
This, in turn allows you to refine the product w/those who are interested in it; while avoid needless complaints from those that may be hard to please.
****
Bots as gatekeepers are the same problem in many ways as the infernal 22 options when you phone customer service at some company or another and get :
Press one for Bill Payment
Press two for Account Balance
Press three to change your password
Press four for store locations/hours
Press five for bad hold musak that never stops
Press six for ......
Press nine to hear this useless menu that didn't include the option you needed all over again.
Which then also omits press Zero to bypass this garbage and talk to a human.
There is a great value in AI and quick-answer tech via desktop/phone/text etc. But it must be properly deployed, which it rarely is.....