There are 3 different types of responses that your Chatbot will produce; high-confidence, low-confidence, and no-confidence responses.
High-Confidence Responses
Your Chatbot will produce a high-confidence response when it is more than 90% confident of its response based on the training data used. This will produce an answer from your Knowledge Sources or a Complex Response.
Low-Confidence Responses
Your Chatbot will produce a low-confidence response when your knowledge sources contain information about the question/topic asked, but the bot doesn’t have enough information for the AI to generate an efficient response. The bot will provide the user with a list of URLs, or other knowledge sources, that are in your bot’s brain that may contain the correct information.
No-Confidence Responses
Your Chatbot will produce a no-confidence response when it is less than 50% confident of its response based on the training data used. Your Chatbot will produce a fallback response, which enables the user to submit a ticket or connect to a Live Chat Agent, depending on the your fallback preferences.
Response Logic
There are 3 “layers” in the bot’s response logic. The bot will first look for custom built responses, or responses built by a human, and send it with high confidence. If there is no custom response, the bot will use a semantic and lexical search within its brain (your knowledge sources). If the bot finds an answer with high confidence, a humanized response will be sent to the user using GPT. If the bot is unable to find an answer, the user will receive a no-confidence response based on the institution’s fallback preferences.
If a high confidence response from GPT needs adjustment, you can edit it in the Inbox tab (as discussed in the Bot Performance Article.)
Next Article in Testing Process: Bot Limitations
Previous Article in Testing Process: Submitting Bot Questions