Building a Customer Feedback Agent with Copilot Studio

Collecting customer feedback through a conversational agent is more natural than static forms. Here's how to build one in Copilot Studio that actually works well in production.

The Design

Our agent will:

• Greet the user and ask about their recent experience
• Classify feedback as positive, neutral, or negative
• Ask targeted follow-up questions based on sentiment
• Route critical issues to support
• Store all feedback in Dataverse for reporting

Setting Up the Agent

Create a new agent in Copilot Studio and configure a Feedback topic with a trigger phrase list that includes things like "feedback," "complaint," "suggestion," and "I want to share."

The Conversation Flow

Step 1: Opening question

Use an open-ended prompt: "Thanks for reaching out! Could you tell us about your recent experience with our product?"

Step 2: Sentiment classification

Feed the user's response to a generative AI node with this system prompt:

Classify the following customer message as "positive", "neutral", or "negative".
Respond with ONLY the classification word.

Store the result in a variable called sentiment.

Step 3: Conditional follow-up

• Positive: "That's great to hear! Would you be willing to leave a public review?"
• Neutral: "Thank you. Is there anything specific we could improve?"
• Negative: "I'm sorry to hear that. Let me connect you with someone who can help directly."

Step 4: Save to Dataverse

Use a Power Automate flow triggered by the agent to save the feedback record with sentiment, timestamp, and user details.

Results

In testing with a pilot group, this agent achieved a 73% completion rate - significantly higher than the 31% completion rate of the equivalent web form.