Thanks for the feedback! I agree that it can be combined and perform well. I believe a hybrid model to be the best approach as it empowers the agent (giving them more time to focus on the client) while several bot agents work behind the scenes to perform simple or complex tasks. Having a combined hybrid model also lowers the error rate of the bot as there is someone who can oversee the responses (and adjust for correctness).
Even in day-to-day usage, we have examples of hybrid sequences. Think, Alexa — Voice & Messaging, plus the owner giving a specific command and acting upon the request. i.e. Asking Alexa to recommend a book to read and me, ordering said book.
I do like how you guys are training your AI bots based on PAST supports giving specific scenario examples (the past is often the predictor of the future). Do you guys see your system able to handle more than level 1 support? I suppose that is where the hybrid model comes into play.