How fallbacks work
In intent-based chatbots, the fallback fires when no intent scores above a confidence threshold. The classic reply, “Sorry, I didn’t get that,” repeated twice, is what makes customers hate bots. Better designs escalate after a second miss and log every fallback for review.
Language-model assistants rarely fail to understand, so the modern fallback problem is missing information. When the knowledge base doesn’t cover a question, the right fallback is “I don’t have that information” plus a next step. The wrong one is a confident guess, which is how AI hallucination reaches customers.
Example: a pet groomer asked about cats
A pet grooming salon’s website chat gets asked, “Do you groom Persian cats?” The salon’s knowledge only covers dogs. A good fallback says it doesn’t have information about cat grooming, offers to have the salon confirm, and takes the visitor’s contact details. The owner sees the question in a list of knowledge gaps and adds the answer. This scenario is illustrative.
Fallbacks in InsertChat
InsertChat’s assistant is set to say when information is not available, a setting you cannot turn off, and it is instructed never to invent business facts, prices, policies, or availability. By default it asks a person for help when an answer is uncertain, and after a customer says twice that an answer didn’t help.
Unanswered or weak questions from real conversations appear under Knowledge gaps in knowledge and answers, with how many times each was asked. Read how to set up human handoff, or see the pet grooming setup.
