Improvement AI

Use Improvement AI to improve your chatbot's content coverage and accuracy more effectively.

With Improvement AI, you work by reviewing how your chatbot performed over a certain period, marking whether it responded to users with the appropriate passage, and where it didn't have the content it needed. This way, you can clearly prioritise just the content changes your chatbot needs and avoid spending time on content your chatbot users aren't asking for.

Improvement AI also takes the guesswork out of improving your chatbot's question accuracy by generating recommended changes to your NLP training phrases that you can publish straight away.

You should try to run a review of your chatbot on a regular basis. If there are particular conversations, passages, or questions you want to review at a different frequency to the rest of your chatbot content, you can create:

• Filtered reviews that exclude specific NLP questions or response passages. 
• Targeted reviews that only include selected passages or conversations.

When you open your Reviews screen, you can see each of your reviews with:

• Two charts showing your chatbot's content coverage and accuracy, based on your last five or ten reviews. 
• The period the review includes.
• The date the review was created.
• How long until the chatbot user responses expire, according to your chatbot's data retention settings.
• The progress of the review.

You can:

• Create a new review.
• Click a current review to continue marking passages.
• Click a finalised review to see your results. 
  The chatbot user responses will be deleted after your data retention period has expired. See Data Settings. You can still see your review results and any comments or labels applied to the passages.

How often should you review?

How often you should perform reviews depends on how much content your chatbot contains, and how many interactions with users it processes on a regular basis. Chatbots that have extensive or very diverse content, or that interact very frequently with users, would benefit from more frequent reviews. Below are some general guidelines: