Quality report

The Quality report turns what shoppers and your team say about the assistant, and what the AI judge finds in its conversations, into a quality signal for the whole site. Open Reports in the sidebar and choose Quality, or go to /dashboard/sites/{siteId}/reports/quality. It's the fifth item in the Reports section, after Overview, Conversations, Sales, and Engagement.

The five tabsCopy Anchor Link

Quality is one sidebar entry with a row of five tabs under its title.

The Conversations, Needs review and Knowledge tabs carry a Beta badge: they count conversation labels, which stay Beta. The page you are on now is the Satisfaction tab. Feedback was already a page of its own; it is now a tab.

SatisfactionCopy Anchor Link

This is the aggregate view of the ratings you can also read one at a time: the tab's own ratings list below the charts, or the Feedback tab's full table.

If no rating was left in the selected range, the whole report shows a single empty-state message instead of zeroed charts. A range reaching further back than 12 months (transcripts' retention window) shows a notice naming the cutoff date instead; see Filters. The same notice appears on the full Feedback table below.

Feedback qualityCopy Anchor Link

Everything in this section is live, computed from the ratings you've already collected. It's scoped to the current site and the selected date range.

KPIsCopy Anchor Link

• Ratings received: the total number of thumbs (up and down) left in the range.
• Satisfaction rate: positive ratings as a share of all ratings (positive ÷ (positive + negative)).
• Negative ratings: the count of thumbs-down, the complaints to act on.
• Conversations with a negative: the share of rated conversations that got at least one thumbs-down. A conversation counts once here no matter how many of its messages were rated down.

When comparison is on, a delta badge appears on Ratings received and Satisfaction rate only. The two "lower is better" tiles (Negative ratings and Conversations with a negative) show no delta on purpose: a rising number there isn't good news, and the badge only ever highlights an increase as positive.

Charts and listsCopy Anchor Link

• Feedback over time: one point per day, positive and negative ratings stacked (green for positive, red for negative), so you can see both the volume and the mix change day to day. On a range of two calendar days or fewer it has one point per hour instead, labelled in your time zone, with hours that haven't started yet without a value rather than at zero. See Charts per hour on a short period.
• Negatives by category: a bar chart of thumbs-down grouped by issue category, most frequent first: where the assistant disappoints most. Positive ratings (which have no category) and negatives left without a category aren't counted here. If no categorized negatives fall in the range, it shows an empty state.
• Satisfaction by agent: each agent's satisfaction rate as a percentage, highest first. Ratings are grouped by the agent as it stands today. With an Agent versions selection, only ratings on the selected versions' conversations count, for every figure on the report.
• Recent negative feedback: the five most recent thumbs-down in the range, each showing its category and comment. Click one to open that conversation's transcript scrolled straight to the rated message.

The charts and lists always reflect the current period only. The comparison toggle affects the KPI deltas, not these.

All ratingsCopy Anchor Link

Below the charts, a Ratings list shows every rating left in the selected period and filter selection: both positive and negative, unlike the five-item "Recent negative feedback" card above. A See all link opens the full table at /dashboard/sites/{siteId}/reports/quality/feedback, carrying the report's whole resolved filter bar along.

Unlike the report above it, the full Feedback table's own filter bar offers all six filters, including Customer feedback, since this list isn't computing a ratio the filter could distort. You can narrow it further from what you arrived with, or add/remove filters independently of the report. Its Subject or keyword filter also reaches further than everywhere else in the Analytics section: a term matches the conversation's subject or the feedback comment, so you can search what shoppers actually wrote, not just the conversation title.

Above the table, a result count shows how many ratings match the current period and filters. Empty for the current filters shows one of two messages: if the period genuinely has no ratings at all, a plain "No feedback yet" message; if your filters are the reason the list is empty (a Panel language or Agents selection too narrow, say), a distinct "No feedback matches your filters" message suggests widening the period or clearing a filter. The message carries no button of its own: use the filter bar's own Reset control (see Filters) to clear back to the defaults.

Each row, here or on the full table, shows:

• Type: a green Positive badge or a red Negative badge.
• Category: the issue category on a negative rating (see the list under Rate a message); a dash for positive ratings and for negatives left without one.
• Comment: the free-text comment, trimmed to 80 characters (hover to read it in full); a dash when there's none.
• Conversation: the title of the conversation the message belongs to, trimmed to 40 characters (hover for the full title); "New conversation" when it has no title yet.
• Created: when the rating was given (the feedback's own timestamp, not the message's).

Click any row to open that conversation's transcript scrolled straight to the rated message's feedback block.

Choosing a range and comparingCopy Anchor Link

The Quality report uses the same header as the other reports: the same date range picker and compare-to-previous-period checkbox. The range filters on when each rating was given, not when the underlying message was sent.

Needs reviewCopy Anchor Link

The Needs review tab is the AI judge's review queue: the conversations whose current label has a review reason, so the judge thinks a person should look at them. It's marked Beta. The queue used to be a tab on the Conversations list; it lives here now.

The rows and the count follow the date range and filters in the page's filter bar. This page offers the conversations list's filters, including Resolution and Issues, which the other Quality tabs don't. The count shows as a line above the table, for example "12 conversations".

A trust reason means one of the trust criteria failed. For Stuck to known facts, that's a hard fact the assistant stated with no source behind it: a price, a stock level, a link, a delivery, return, payment or warranty term, a delay or an amount committed in your name, a figure about the product, or an ingredient or allergen said to be present or absent. Since taxonomy v1.2, a product description, a usage tip, a benefit or a comparison between your products that the sources don't back no longer puts a conversation here.

The queue has its own order, and the column headers don't sort it. Trust reasons come first, then service, then fallback. Within a reason, conversations with purchase intent come first, then the newest.

The page is read-only. You can't correct a label, assign a conversation or mark it done from here. Conversations with no review reason aren't on it, and neither is the lower-priority list described under Review priority.

Opening a row shows that conversation's transcript. Previous and Next arrows aren't available from this page yet, so a conversation opened here has no place in the queue to step through. The back arrow returns you to the Conversations list, not to this page. The list opens with the period and filters you last set there, which this page's filter bar shares.

The freshness lineCopy Anchor Link

Above the table, a line says whether labeling is keeping up. It has two figures:

• When the hourly schedule last ran. Clicking Label now on a conversation doesn't count, so a stopped schedule can't look alive.
• How long the oldest waiting conversation has waited, counted from the end of its 30-minute quiet period. It covers every site of your organization.

It reads in plain text when the last scheduled pass is under 2 hours old and no conversation has waited more than 3 hours, for example "Labeling last ran 25 minutes ago. Nothing is waiting." Otherwise it's a warning that names the figure at fault:

• "Labeling has not run for 3 hours. New conversations are not being labeled." The last scheduled pass is over 2 hours old.
• "Labeling has not run yet. Conversations are not being labeled." No scheduled pass has been recorded.
• "The oldest conversation has waited 4 hours. Labeling is behind." The schedule ran recently, but a conversation has waited over 3 hours.

The line ignores the filter bar. If it can't be computed, the line is hidden and the page still works.