Set Up CSAT Surveys on an Agent

In this tutorial you'll get an agent ready to ask for satisfaction at the end of every conversation and read the aggregates next week.

Prerequisites

You need:

• A live WhatsApp or email agent (with at least one test conversation).
• Edit permissions on the agent (admin role or owner).
• ~10 minutes.

Step 1 — Create the question set

1. Go to AI Agents → WhatsApp (or Email) → open the agent.
2. Tab Settings → CSAT → New question set.
3. Intro: "Can you help us with 2 quick questions? Anonymous, under 1 minute."
4. Question 1:
   • key: satisfaction
   • type: scale_1_5
   • prompt_to_ai: "Ask how satisfied they were with the service (1-5 scale)"
   • required: true
5. Question 2:
   • key: would_recommend
   • type: yes_no
   • prompt_to_ai: "Ask if they would recommend us to a colleague"
6. Thank-you: "Thanks! Your feedback helps us improve."
7. Save → now version 1.

Step 2 — Flip csat_enabled

1. Same Settings tab, toggle Enable CSAT.
2. Save.

From here on, any conversation idle for the agent's session_timeout_minutes (default 30 min) fires the survey.

Step 3 — Test the flow

1. Open a test conversation with the agent.
2. Send "hi" and leave it idle more than 30 minutes.
3. The agent sends intro + question 1. Reply "4".
4. The agent phrases question 2. Reply "yes".
5. The agent closes with the thank-you.

The response lands in _csat_responses — you can see it on the agent detail under the CSAT Responses tab.

Step 4 — Read aggregates after a week

Let the agent run ~1 week to collect 20-50 responses.

1. Dashboard Analytics → CSAT → filter by agent + channel.
2. Check:
   • Response rate: if under 20%, the intro is likely too long or there are too many questions.
   • Average score of question 1 (satisfaction).
   • Yes/no distribution of question 2 — % of users who would recommend you.

Step 5 — Iterate

If response rate is low or questions aren't capturing what you need:

1. Edit the question set → the system creates version 2, archives v1.
2. New responses point to v2; previous responses stay linked to v1, untouched.
3. In the dashboard, aggregates separate by version — compare evolutions without mixing data.

Limits worth remembering

• Triggered on idle timeout only, not on a user's "thanks, bye". If your conversations are short and closed, 30 min default is fine.
• Max 10 questions. Past that, response rate collapses.
• Per-channel. If the same team runs WhatsApp + email, the question set is per-channel — make one per channel if you want different questions.

See features/csat for the full reference and available question types.