CSAT

What CSAT does

CSAT is a short satisfaction survey sent to the customer after a conversation has been closed. One question, five faces, an optional comment. It is the only number in the product that comes from the customer rather than from us, which is what makes it worth protecting.

You turn it on under Settings → CSAT.

What the customer sees

The survey is a single question with a five-point scale:

• 😠 Very dissatisfied
• ☹️ Dissatisfied
• 😐 Neutral
• 🙂 Satisfied
• 🤩 Very satisfied

They pick a face, can add a comment, and then see your thank-you message. That is the whole flow. There is no second page and nothing to log in to.

The settings

• Enable CSAT surveys ✅ The master switch. Everything else is hidden until this is on.
• Send email after (hours) ⏱ How long after the conversation closes the survey goes out. The default is 72 hours.
• Survey title 🏷 The heading above the question. Leave it empty and each customer gets the built-in title in their own language, which is usually the better choice.
• Survey question ❓ The question itself. The default is "How would you rate the support you received?".
• Thank you title and Thank you message 🙏 What is shown once they have rated.

Each of those text fields has a Translations action next to it. If you support several languages, use it. A Swedish customer rating a Danish question is a worse experience than no survey at all.

Why the delay matters

The default of 72 hours is not arbitrary. It is long enough that the thing you promised has usually happened, and short enough that the customer still remembers the conversation.

Send it too early and you are asking about a refund that has not landed yet, or a parcel that has not arrived. You will be rated on the outcome, not on the support, and the outcome has not happened. Send it too late and you are asking about something they have stopped thinking about, so the response rate falls and the answers get noisier.

If your typical case is resolved on the spot, shortening the delay is reasonable. If most of your cases end with "your replacement is on its way", leave it long or push it out further.

When a survey is deliberately not sent

This is the part people are most often surprised by, and almost every rule here exists to stop you from measuring something meaningless.

• The channel does not support it. Surveys go out on email, chat and connected helpdesk conversations. Other channels are skipped.
• The conversation was closed as noise. Anything the Auto Closer closed as out of office, spam, marketing, an auto-reply or a blacklist match gets no survey. Nobody wants to be asked how satisfied they were with the spam they sent you.
• It was closed as a duplicate. The customer already has the real conversation, and that one can be surveyed instead.
• A survey is already scheduled or sent. One per conversation, always.
• An earlier reply to this customer failed to deliver. If a previous outbound message bounced, hard-failed or was reported as spam, the survey is skipped. It would bounce too, and a bounced survey reopens the conversation, which can re-close and re-schedule in a loop. This is common with addresses like no-reply@ that never accept incoming mail.
• It is a carrier conversation. Automated traffic from a shipping partner is not a customer.

Note: A "thank you" closed by the Auto Closer as a courtesy message does still get a survey. The case was genuinely resolved, so the rating is real.

Some connected helpdesks are configured to only survey conversations the AI resolved by itself. If you see far fewer surveys than closes on one integration, that is the reason.

Reading the results

Open Insights → CSAT, or use the View Insights button on the settings page. You get the distribution of scores, the average over time, and how many responses came in. From there, Responses gives you the individual ratings with their comments.

Three things worth holding in mind when you look at it.

Read the comments before the average. The average tells you whether something changed. The comments tell you what. A score that dropped half a point is a puzzle; six people writing the same sentence is an instruction.

Watch the response count, not just the score. A great average from eleven people is not a result. If the count is small, look at a longer period before you conclude anything.

Low scores are usually about the outcome. Customers rate whether their problem got solved, not whether the writing was pleasant. A run of bad scores on a single topic is nearly always a policy or a process problem rather than a tone problem.

Good habits

Turn it on early, even if you are not ready to act on it. You cannot get last month's satisfaction back later, and the first weeks of a new setup are exactly the period you will want to compare against.

Translate the copy before you launch a new market. It takes five minutes and it is visible to every customer in that country.

Do not change the question often. Every change quietly breaks comparison with everything that came before it. Pick a wording you can live with and leave it alone.

Follow up on the very dissatisfied. A one-face rating with a comment is a customer telling you exactly what went wrong, in writing, unprompted. That is worth more than the number it is attached to.