The Knowledge Base agent has strong potential for turning support history into useful documentation but it has a lot of limitations that lead to inaccurate and not relevant articles.
We would like HubSpot to consider three improvements:
- A detailed feedback loop so the Knowledge Base agent can learn why a suggestion was rejected. How to report the poor result reasons? We can just provide a basic signal, “Not helpfull / thumbs down” and/or dismiss. How can AI improve future recommendations?
- A way to ask the AI to generate a basic article from a selected list of tickets. A team may already know that a particular group of tickets belongs together and should be able to select those tickets and ask the AI to create a basic article draft from that specific set.
- Support for attachments associated with ticket conversations (including PDFs, Word files, screenshots, test reports, drawings, and other technical documents) as article-generation sources.