chapter one
1 Introduction to improving conversational AI
This chapter covers
- How to identify and improve the top failure modes of conversational AI
- How to assess where generative AI can help you in your conversational AI
- How to safely use generative AI
- How to continuously improve your AI (from those top failure modes) towards a defined target
Perhaps you have a conversational AI solution in production. Maybe you built it yourself, maybe you inherited it, or maybe it's now just a part of your daily life. Or maybe you use a chatbot and want to learn more about it. That AI solution was deployed with some primary goal in mind: to deflect users from a call-center, to provide speedier resolution to issues, or to provide more self-service capabilities to users. We’ve seen various kinds of conversational AI:
- Question-answering: Also known as FAQ bots, these AI solutions deliver a response directly to a user’s question, usually without any follow-up.
- Process-oriented or transactional solutions: The user is guided by an AI to achieve some goal through a series of questions from the bot. For instance: checking an account balance, booking an appointment, or checking the status of an insurance claim.
- Routing agents: The bot’s only job is to figure out where to redirect the user next. The redirection may be to a different specialist bot or a human agent.