I recently observed that AI customer support is usually judged wrongly. ‎Companies talk about feedback, swiftness, automation rates, knowledge bases, and how many prompts their AI can give response to. The big issue is what really goes on when the AI is stuck. Is it able to understand that it has failed? Can it tell when it doesn’t know something? Does it collect the information a human agent will need? Can it handover the conversation without making the customer explain everything again?
What sets AI support apart isn’t how much it answers, but it’s ability to know when to remain silent. Personally, I’d rather have an AI that admits it needs help than confidently send a customer off track.
What do you think is more important to you when judging an AI support tool: its automation rate or how it handles problems that are difficult for it?