Heymarket's AI Agent feedback loop lets your team flag agent responses that miss the mark and automatically turns that feedback into actionable knowledge base improvements. Over time, this closes gaps in your AI Agent's performance without manual auditing.
How It Works
The feedback loop moves through four stages: user feedback, issue categorization, AI summarization, and admin review.
1. Rate Agent Responses in Supervisor Mode
When your AI Agent is running in Supervised Mode, any user with access to the conversation can rate a suggested reply. Use the thumbs up icon to confirm a good response, or thumbs down to flag one that needs improvement.
2. Submit Feedback Details
Clicking thumbs down opens the Submit Agent Feedback modal. Select one or more accuracy categories:
- Incorrect information — the agent stated something factually wrong
- Missing important information — the agent's response was incomplete
Optionally, add a short explanation describing what was missing or wrong. Specific feedback produces stronger recommendations downstream.
3. Daily AI Summarization
On a daily cadence, Heymarket's AI reviews all feedback submitted across your workspace and identifies patterns. It groups recurring issues, detects knowledge gaps, and generates specific, actionable recommendations for your knowledge base.
4. Review Recommendations in Agent Enhancements
Admins and owners can review the generated recommendations under Settings → AI → Agent Enhancements. Each recommendation includes a severity rating, impact summary, and the proposed knowledge base update.
Click any recommendation to see the full details, including the source of the gap and the suggested article content.
From here, you can either Reject Suggestion or Mark as To Do to add it to your team's queue. Status updates are confirmed with a toast notification.