We used boost.ai for customer support automation where we needed predictable answers, good handoff to live agents, and clear reporting. It helped reduce repeat tickets once we got the top intents trained and the content cleaned up. The best results came when we treated it like an ongoing program, not a one-time launch.

The fact that processes all language to English (both samples sentences and user input) before matching is both a strength and a weakness. The advantage is that we can easily add answers in new languages without adding new samples. Our question or intention hierachy remians the same. However, we have to do continual workarounds to account for the fact that a single word in our language may actually have two or more very different meanings, depending on context - but we can only assign it to a single English word. Translated language processing is not exactly a tool for precision of understanding. We use Boost for answering questions from the public about their rights and duties and about the laws and rules. It is challenging because there are a lot af nuances in what they ask. We find that Boost's "ASU" feature does help our chatbot to answer many more questions instead of going to fallback - however, since the answers are approximate: "You might mean this, or perhaps that?" The chatbot gives the impression of having an insecure grasp of the user input. Also users are then at risk of making an incorrect choice of alternative.