Most ML in support implementations I've seen, including the one in my team [1], focus more on providing solutions directly to the user so they never need to talk to an agent in the first place, which (if successful) provides far large efficiency & user satisfaction gains.
I posted a ticket saying i was overcharged unfairly for a trip, and a bot resolved the issue and sent a reply apologising for the incovenience and hoping i use uber again soon.
I demanded to speak to a human, but got the same automated reply again!
Boooooo!
My Uber incident involved my driver running out of gas on the way to the airport. Like as a PROFESSIONAL driver, knowing how much gas you have is pretty important. Even asking me to stop for some would’ve been better.
Uber support told me they were sorry and only refunded that ride, not the one I had to hail and pick us up on the side of the highway at 4am(which kept directing the driver to a road along side since they don’t expect you to need a pickup on the highway, the first car gave up after driving in a circle and the second I had to explain to what was going on and he was cool though.) in freezing Seattle weather while having to rush through the airport to get our flight. Pretty much told me tough luck and it’s not their fault.
I hope Uber fails.
Why? It's a low margin business and stuff like this drives up cost for everyone. I'd rather have low cost rides than talking to human support
Is there any data on whether people prefer talking to a human being and having a less successful outcome vs talking to an AI and having a more successful outcome?
To be picky, it is a negative margin business (at least until now).
Still this is not a reason to have not decent customer support.
The usual approach (not necessarily by Uber, almost everyone is on this - IMHO flawed - track) is the periodical announcement roughly on the lines of "our automated, automagical AI/whatever is so perfect that the service/product was so bettered that the need for support was reduced to 0.0000000000001% of customers". (which is good)
Then, IF it was actually true (it usually isn't) that single request every zillion transactions/sales could be actually dealt with properly with a marginal cost by a human (possibly someone actually capable and with the powers to actually provide some form of remediation).