I'm not entirely sure whether it would be achievable with the way CloseBot actually works. I used to work with a chatbot system built on NLP referencing a knowledge base for it's answers (no LLM) and the stand out feature was a heatmap of what customers had asked (using the "AI" to group questions into tiles based on the intent of the question, so that the many different ways of asking the same thing were "accurately" grouped). Being able to say that the majority of questions were about a particular thing was often useful in a number of ways; Firstly to identify "gaps" in the customer journey on the website - e.g. if a majority of people are asking "when are you open?" on a store's website, that information needs to be more readily available. Secondly, being able to see what the bot is having an impact on in a quantified way was really helpful, both for the website owner but also for us with the business case for the bot. Thirdly, being able to see conversations that the bot could not "resolve" on its own and needed to escalate to a human led interaction.
