Chatbot Sensitivity determines that how your Chabot is sensitive to visitors’ questions during the live chat. The sensitivity score range is from 0 to 100 and can be manually adjusted. This will change the score range of the returned “High Confidence Answers”, “Possible Answers” and “No Match Score Answers”.
Once a visitor asks a question, the system will then calculate the Matching Score of the question from the pre-defined Intents base with in the system and sort out the Matching Score from high to low. Each question in the pre-defined Intents base will have a real-time Matching Score.
By default, the High Confidence Answer Score is 40 and No Answer Score is 10. Hence the questions in the pre-defined base which have Matching Score equal to or greater than 40 will include corresponding answers in this range.
If the score of one question is equal to or greater than the High Confidence Answer Score 40, the Chatbot will send the preset response of this question to visitors as the High Confidence Answer.
If multiple questions in your pre-defined Intents base have the score equal or greater than High Confidence Answer Score 40, or there are one or more answers whose Matching Score is less than the High Confidence Score but greater than No Answer Score, Chatbot will send 3 questions that have the highest scores to visitors to choose from and then provide a Possible Answer.
The question above “What is Chatbot” does not get a High Confidence Answer. However, it is in the Possible Answer range (scores from 11 to 39) and so it is why Chatbot returns a list of suggested available questions that could be more relevant to their query and help the Chatbot to answer it in a timely manner.
Note: If there is only 1 or 2 Possible Answers, Chatbot will give a list of 1 or 2 possible relevant questions that the visitor can select to get their answers and be more specific.
No Answer Score defines the Matching Score range in which the Chatbot returns an alternative solution to the visitor since the original question asked by the visitor returned as being not within the operating score range and so the Chatbot was not able to find any pre-defined solution.
The above example question “Where can I find the email address of a customer?” gets a very low Matching Score which is under 10. So Chatbot returns No Answer. If you go to Bot Learning page, you can find these questions as Unrecognized questions, and scores will be given.
In the following example, Chatbot delivers 2 Possible Answers’ Standard Questions for the question “How to change payment info?”, even the visitor’s asked question is quite close to the questions listed in the Chatbot answer. This sample question should get a very high Matching Score but it is still under the High Confidence Answer Score 40.
Now if you change the High Confidence Answer Score to 35 and send the same question again, Chatbot might send a High Confidence Answer directly as only one question’s Matching Score is equal or greater than 35. Also, there will also be scenarios that Chabot returns Possible Answers to one visitor question, but returns a High Confidence Answer directly to the same question if you increase the High Confidence Answer Score.
If your Bot Intents is well equipped with all the question types of your business and it is “smart” enough to answer the visitor questions directly, you may increase the High Confidence Answer Score. Thus, there is a higher chance that your visitor gets the High Confidence Answer directly instead of the possible question list.
In the case of No Answer Score, if your Chatbot returns no answer to most of your visitor questions, you may decrease the No Answer Score. So there is a higher chance for the Chatbot to deliver a selection of possible questions which should help instead of No Answer.
Each live chat Campaign will have its own Chatbot, which means the Chatbot sensitivity can be varied.
Helping your Chatbot to learn is always the best policy to improve its sensitivity. The more questions you give Chatbot to learn, the more accurate it will be in fetching the right answers.