What you will learn
- Provide four different types of feedback on agent responses.
- Understand how your feedback becomes permanent training data.
- Use the "Include in Training" toggle effectively.
- Understand how the automatic Scoring Skill measures conversation quality.
How to Give Feedback
Training an AI Agent is not a one-time event; it is a continuous process. As an Agent User, your feedback is the primary engine that drives the agent's improvement.
Whenever you are reviewing a conversation in Copilot, you will see thumb icons beneath every agent message. Clicking these icons opens the feedback modal, where you can provide four different types of input.
Type 01
Thumbs Up
A quick signal that the answer was accurate, complete, and had the right tone. Do not skip positive feedback — it reinforces good patterns the agent should keep.
Type 02
Thumbs Down
A quick signal that the answer needs work. Flags the response for Admin review.
Type 03
Written Comment
A plain-text explanation of exactly why the answer was good or bad (e.g., "The refund window is 30 days, not 14").
Type 04
Suggested Replacement
The most powerful feedback type. You write the exact response the agent should have given. The agent learns exceptionally well from high-quality rewrites.
From Feedback to Training
Feedback in Copilot does not just disappear into a void. It enters a structured continuous improvement cycle.
When you submit feedback, it is immediately logged to the conversation history. From there, an Administrator reviews the feedback in the dedicated Feedbacks tab.
If the feedback represents a permanent rule or correction (rather than just a one-off situational fix), the Admin can mark it to be Included in Training. Once added to the training memory, the agent will apply that improved behavior to all future conversations.
The Scoring Skill
While you provide manual feedback on specific messages, raia also measures quality automatically using the Scoring Skill.
If the Scoring Skill is enabled on your agent, it will automatically evaluate the quality of every conversation after it ends, assigning a score from 0 to 10 based on factors like engagement, accuracy, and resolution.
As an Agent User, you do not need to configure the Scoring Skill, but you should use it to prioritize your work. Instead of reading every conversation, you can filter your Copilot view to only show conversations that scored below a 6, focusing your manual feedback where it is needed most.
Frequently Asked Questions
If I give a thumbs down, does the agent immediately stop giving that answer?
No. A thumbs down flags the response for review. The agent's behavior changes permanently once an Admin reviews the feedback and adds the correction to the agent's training memory or instructions.
Should I write a comment or provide a suggested replacement?
A suggested replacement is always better. Showing the agent exactly what a perfect answer looks like is much more effective than just telling it what it did wrong.
Can end-users (customers) leave feedback?
Yes, if the Live Chat widget is configured to show feedback buttons, customers can click thumbs up/down on responses. This customer feedback appears in the exact same Feedbacks tab for Admin review.