Conversation Quality
Score resolved support conversations against your own rubric, with a dispute trail your team can trust
Every resolved conversation gets a quality score, judged against a bar your team defines.
Conversation quality reviews your resolved support tickets and scores each one from 0 to 100, so coaching conversations start from evidence instead of anecdotes. It lives in two places:
- The Quality tab of the Service hub shows the scores: what was scored, and any dispute about it.
- Service, Configuration, Quality holds the setup, in two sections: Rubrics (what tickets are judged on) and Lanes (which scoring lanes run). The Configure link beside Learn more on the Quality tab opens it. It opens for people who can view tickets and assign tickets; anyone else reads You need permission to view tickets and to assign tickets before you can configure conversation quality. An administrator in your workspace can grant it.
Resolved tickets are scored within about 30 minutes and appear on the Quality tab, newest first. A summary row shows the Average score, Open disputes and Overturned scores. You can narrow the list to All, the scores at or under your pass mark (for example 70 or lower), or Disputed. Opening a score shows the per-criterion breakdown behind the number, so a 62 is never just a 62: you can see which criteria earned it.
Before the first score
- With scoring on, the tab reads No scored conversations yet and Scores appear within about 30 minutes after a ticket is resolved. Until you add a rubric, the built-in criteria apply. with Open the Inbox. Resolve a ticket or two and the first scores follow.
- With scoring off, it reads Scoring is turned off, so no conversations are scored. If you can manage quality, Turn scoring on opens Configuration, Quality. Otherwise it adds A manager can turn it back on.
Deterministic checks that always apply, and an AI review of the conversation itself.
- Checks that run without AI: rules the app checks itself, such as whether the customer got a reply inside your target and whether the conversation was actually resolved. Free and unlimited on every plan.
- Scored by AI: an AI review reads the conversation and grades it against your active rubric (or the built-in criteria). Included at no cost. It never uses your AI credits.
The Lanes section of Configuration, Quality holds a switch for each lane, and only a service administrator can change them. Turning one lane off never changes the other lane's setting. But nothing is scored while Scored by AI is off, so the checks lane pauses too, and its row says Nothing is scored while Scored by AI is off, so this lane is paused too. With both lanes on, every resolved conversation carries both the deterministic checks and the AI review.
Define what a good conversation means for your team, and test it before it goes live.
Until you author a rubric, conversations are scored on the built-in criteria: empathy, sentiment, comprehension, and resolution. A rubric replaces that default with your own bar: a set of criteria, each with a weight, and optionally marked Critical (a conversation that misses a critical criterion fails outright, whatever its other marks). With no rubric yet, the Rubrics section reads No rubrics yet and Create a rubric, test it, then activate it. Creating and activating rubrics needs permission to assign tickets.
Create the rubric
In Configuration, Quality, press New rubric. Give it a Rubric name and a Pass mark, out of 100, then Add criterion for each thing you judge: what good looks like, its Weight, percent, whether it is Critical, and its options and their scores. New rubrics always start inactive and untested: nothing is graded on your draft while you shape it.
Test it against real conversations
Run test saves the rubric and dry-runs it against your recent resolved conversations without writing a single score: The dry run reads real resolved conversations without calling a model, so it costs nothing and changes no score. Under Worth knowing before you activate, the report warns when one criterion soaks up most of the signal (several criteria asking the same question) or never fires at all (too narrow to matter). Run test needs Scored by AI to be on.
Activate it
Activation is refused until the rubric has been tested, because going live changes how everyone's work is graded from that moment on. One rubric is active at a time; activating a new one stands the previous one down. Stand down on the active rubric returns scoring to the built-in criteria.
A score that grades a person's work has to be contestable by that person.
The agent a score is about can press Dispute this score, say Why this score is wrong, and Send dispute. Until a manager looks, the score reads Waiting for a manager to review it. A manager then rules on the dispute, with an optional Note for the record: Uphold leaves the score standing, and Overturn records that the team decided the score was wrong. The score then reads Upheld by a manager or Overturned by a manager. Nothing is deleted or rewritten at any step: the original score, the objection, and the ruling all stay on the record, which is what makes the number defensible when it comes up in a coaching conversation.
Neither side has to go looking. Filing a dispute sends a Conversation quality dispute filed notification to everyone who can decide it, and ruling on one sends a Conversation quality dispute decided notification to the agent the score is about. Both arrive in the app, in your notifications, and neither is emailed on its own. They are there so nobody has to poll a tab: the objection reaches a decider and the outcome reaches the person it is about, without either of them refreshing anything.