Configure

Surveys & CSAT

Ask visitors a short survey when a conversation ends, then track customer satisfaction over time — average rating, resolution rate, how many conversations actually leave feedback, and a per-persona breakdown.

Turn on the survey

1. Open Analytics → Customer feedback from the dashboard.
2. Pick the widget you want (if you have more than one).
3. Flip the CSAT survey toggle on.

That's it — visitors now see a quick survey at the end of their chat. Everything below is optional fine-tuning.

Choose what to ask

With the survey on, a small builder appears under the toggle. In the panel below, the CSAT survey switch at the top is the on/off from the step above; once it's on, the When to ask dropdown and the three question rows unfold. The Rating question is always on (it drives your Avg rating), while “Was it resolved?” and Open feedback each have their own toggle so you can turn them off — flip a toggle off and its wording field disappears. Edit any question's text inline, then press Save survey.

When to ask

• When the conversation ends — the survey appears as the chat closes. Best for most sites.
• After a human hands the chat back to the AI — only ask after a teammate has taken over and handed back, so you measure the assisted experience.

The questions

• Rating (1–5 stars) — always on. This drives your Average rating.
• “Was it resolved?” (yes / partially / no) — optional. Drives your Resolution rate.
• Open feedback (free text) — optional. Non-English answers are auto-translated to English for you in the dashboard, with the original kept too.

Edit the wording of any question, toggle the optional ones off, then press Save survey.

Read your results

The Customer feedback tab shows four headline numbers:

• Avg CSAT rating — the mean of all star ratings in the period. If only a handful of people have rated, we flag it as directional so a couple of responses never mislead you.
• Resolution rate — the share of responders who said their question was answered.
• Response rate — how many of your conversations actually left feedback. A great average from 2% of conversations means something very different from a great average from 40%.
• Total responses — the raw count in the period.

Below that you'll find the rating distribution (how many 1s through 5s), a responses-over-time trend, a per-persona breakdown (which persona earns the best scores), and your most recent feedback with the visitor's comment.

Real-time sentiment

Surveys tell you how a conversation ended; sentiment tells you how it felt the whole way through — even when the visitor never fills in a survey.

Turn it on in Analytics → Sentiment (per widget, off by default). Sentiment scoring is on the Scale plan.

In the panel below, the Sentiment scoring switch at the top is that per-widget toggle. Once it's on and messages start flowing, the tab fills in with your headline numbers and a Needs attention list — the conversations with the longest run of negative messages, each tagged with how many negatives in a row, so you can jump on a frustrated visitor.

• Every visitor message is classified — positive, negative, a question, or neutral — by a lightweight model, off the chat's critical path so it never slows a reply.
• The Sentiment tab shows your message-sentiment mix, average sentiment over time, and a “Needs attention” list — the conversations with the longest run of negative messages, so you can step in on a frustrated visitor.
• It works in any language (the classifier reads the visitor's own words; nothing is translated first).

Escalate when a chat turns negative

Don't just measure frustration after the fact — act on it live. When a conversation racks up several negative messages in a row, Quincer AI can automatically escalate it so a human picks it up before the visitor gives up.

Configure it per persona: open Personas → (a persona) → Channels and, under Where your team gets notified, open the row titled Escalate on negative sentiment. It needs real-time sentiment switched on for the widget (above). Like every row on that tab it starts collapsed with its saved setting on one line — “Off — negative conversations are not auto-escalated” until you turn it on. Expand it and tick Escalate on sustained negative sentiment to reveal the rest: Escalate to picks the destination, Negative streak threshold sets how many negatives in a row it takes, and Cooldown (minutes) keeps one rough chat from re-escalating every turn.

• Negative streak threshold — how many consecutive negative messages trigger an escalation (default 3). One grumpy message won't fire it; a sustained run will.
• Escalate to — pick where it goes:
  • Flag for a human (Live conversations) — posts nowhere; it marks the conversation in Inbox → Live with an “Escalated” badge so your team can take over.
  • Slack / Teams / Telegram / Google Chat — posts to that platform's Escalations go to target — the channel, chat or space set in its own row under Where your team gets notified, the same target the persona uses when it can't answer a question — with a Take over button.
  • Choosing Google Chat escalation space also reveals a space picker inside this row, so you can set that space without opening the Google Chat row. Only spaces you've added the Quincer app to appear.
• Cooldown — the minimum gap before the same conversation can escalate again (default 60 minutes), so one rough chat doesn't spam your channel every turn.
• If it's on but no target is chosen — or the platform you picked has no escalation channel, chat or space saved — the row's pill turns amber (Needs attention) and its one-line summary says so (“On, but no escalation target chosen — nothing will be sent”), because nothing is actually being sent.
• An amber row can also open itself the first time you load the Channels tab — the one exception to every row starting collapsed — but only ever one row does. The tab checks Slack, Microsoft Teams, Google Chat, this Escalate on negative sentiment row and finally Phone, in that order, and opens the first one that needs attention. So if Slack is amber too, Slack opens and this row stays collapsed behind its amber pill until you click it. Once you've opened or closed anything yourself, your choice wins for the rest of the visit.

Tips

• Keep it short. The rating plus one optional question gets the highest response rates.
• Use the per-persona table to spot a persona that's underperforming and tune its tone or knowledge.
• Watch the response rate, not just the average — a rising response rate means more people are willing to tell you how it went.