"Success rate" is a tricky metric.
I had to use an Amazon chatbot a few weeks ago. It introduced itself as "Deepshikha". After that facepalm, I started down the path with its various preferred responses to get a refund on an item I bought that never arrived, had no tracking information, and was almost certainly a scam by a third-party seller. I eventually, after a few tries, selected the right combination of things to get the refund processed. But the chatbot wasn't helpful, it didn't make any decisions, and it simply served as a filter for scam refund requests.
I guarantee you that some middle-management PM and some VP at Amazon counted that interaction as a success. I'm sure that's how it ended up on their quarterly graphs and charts. After all, the customer (me) got what they wanted and the right decision was made. And, !bonus!, it used "AI", reduces cost, and had low latency. Raises and promotions are almost certainly incoming!
But the experience was abysmal and insulting, contibuting to the ongoing ensh*ttification of the Amazon experience.