Customer satisfaction (CSAT)

CSAT asks customers to rate their experience from 1 to 5. Ratings of 4 or 5 count as positive in the CSAT charts.

Mimir collects it in two places.

In chat

Customers are shown a rating screen at the end of a chat. Nothing to configure — it's part of the widget.

The wording is yours, under Settings → Channels → your chat channel → Text, in the Rating & feedback section: the title, the subtitle, the label on each of the five stars, the feedback placeholder and the send button. Each region has its own set.

By email

Email CSAT is a survey sent after a conversation is resolved, configured per email channel under Settings → Channels → your email channel.

Turn on CSAT Survey and you get:

Delay (hours) — how long after resolution to send

Subject

Greeting

Low rating label and High rating label — the ends of the scale

Footer

The survey job runs hourly. A conversation gets one only if it's resolved, past the delay, has a customer email address, has had at least one reply sent, and hasn't already been rated.

Companies partway through a helpdesk migration are skipped, so historic imported conversations don't trigger a wave of surveys.

Choosing a delay

The default is a day. Short delays catch the experience while it's fresh; longer ones let the customer find out whether the answer actually worked — which matters when the resolution involved a delivery or a refund.

If your resolutions usually depend on something happening afterwards, a longer delay gives a truer picture.

Reading the results

CSAT Score is the share of rated conversations rated 4 or 5. CSAT Distribution breaks it down by rating, and CSAT Over Time shows the daily trend.

Two things to keep in mind:

Only rated conversations count. Response rates on satisfaction surveys are low, and unhappy customers respond more than happy ones. Your score is a sample with a known bias, not an average of everyone.

Ratings follow the whole experience. A customer rating a chat rates the outcome — including a delayed delivery or a policy they dislike — not just how well Mimir wrote. A dip is a prompt to read the conversations, not necessarily a problem with the AI.

Using the feedback

Filter conversations by Customer rating in the inbox filter menu and read the low ones. That's more useful than the score.

Patterns worth looking for:

Long waits in live chat. Usually availability or notifications rather than answer quality — see Notifications.

A wrong answer repeated across conversations. Expand Mimir worked for X seconds to find the source, and fix it once. See Why did Mimir answer that?

The right answer, badly received. A policy problem rather than a support problem — worth passing to whoever owns the policy.