Analytics: what every number means

Where analytics lives 📊

Zuzana's analytics are dashboards built from your conversations, in three places:

• Dashboard, the first item in the sidebar (Dashboard): volume, resolution, response times, CSAT and tags for the whole team.
• More → Insights: every dashboard you can open. Click a dashboard's title to jump to another one.
• Home in the helpdesk: each agent's own numbers under Your performance.

Insights always lists the Dashboard and the AI dashboards Drafts, Policies, Workflows and Chat Revenue (which needs a connected Shopify store or synced orders). Depending on your setup you may also see Backlog, CSAT, Agent Performance, Explorer and custom reports.

Note: dashboards are for Admins. Agents with the User role are sent back to the helpdesk if they open the Dashboard or Insights (Backlog included). Their own figures are on the helpdesk Home.

Filters and switches on the Dashboard

• Date range 📅 The last 30 days up to today by default. You cannot pick dates before your team went live.
• Markets 🌍 One or several. Empty means all.
• Channel 💬 E-mail (contact forms included) or Chat, plus Phone if phone is switched on.

They sit behind the Filters button, show as chips above the numbers, and each dashboard remembers its own while you stay signed in. The Messages / Conversations switch changes what you count, the chart switches between Daily, Weekly and Monthly, and the time cards between Avg and Median.

Click a number to see the conversations behind it 🔍

Almost every number on the Dashboard is clickable. It opens the exact conversations behind it, with the same dates and filters, each linking to the ticket. When a number surprises you, reading ten of them beats any definition. The Auto-closed row opens the Auto Closer's run log instead.

Intake, Noise, Resolved, Engaged and Untouched

The Dashboard puts every conversation created in the date range into exactly one bucket, so Noise + Resolved + Engaged + Untouched = Intake.

Noise is split into Closed on arrival (reached Zuzana already closed, usually from a connected helpdesk), Auto-closed (spam, out of office, marketing, auto-reply, courtesy, blacklist, bounce, abandoned chat), Duplicates, Closed without response (closed with no reply at all, including a missed call nobody returned) and Other (for example forwarded to an outside address). Note that a courtesy "thank you" close is noise here even though it still gets a CSAT survey, and a ticket closed by hand without a reply is noise whoever closed it.

Who resolved it: AI, AI-assisted or Human 🤖

At the close, Zuzana looks at the replies sent before it:

• AI auto-resolved: only the AI replied.
• AI-assisted: the AI and a person both replied, or only a person replied but with AI help: a reply at least 50% the same as an AI draft, or an AI-proposed action a person approved.
• Human: only people replied, with no AI draft used and no AI action approved.

Internal notes, automatic messages and anything sent after the close (like the CSAT survey) never count. The AI rate ring is AI auto-resolved plus AI-assisted, out of all resolved conversations. Resolved by channel shows the same split per channel.

First response time and resolution time ⏱

• Avg. First Response: from the customer's first message to the first real reply. AI replies count; notes, automatic acknowledgements and failed sends do not. The bars split by who replied first, so on chat, where the AI answers first, the Human bar is small or empty. That is expected.
• Avg. Resolution Time: from creation to the last reply sent before the close, not to when someone pressed Done. The bars split by who resolved it.

Both are calendar time (nights and weekends count), leave noise out, and are recorded when a conversation is resolved, so open tickets are not in them yet. For a call, first response is the time until the customer was reached, and resolution time is the wait plus the talk time.

The Messages view

Switch to Messages and every customer message is one unit of work, counted when it arrived and labelled by the first reply after it: AI, AI-assisted (a person replied using at least 50% of an AI draft), Human or Awaiting. Response time runs from each message to the next reply. Use it when long back-and-forth threads make conversation counts misleading.

Which day a number lands on 🗓

• Conversations count on the day they were created. A ticket created on the 3rd and closed on the 10th is a resolved ticket of the 3rd, so last month can still change as old tickets close or reopen.
• Days are UTC, not your team's time zone. A ticket created just after midnight in the Nordics lands on the previous day.
• Exceptions: the AI lines on the Support over time chart use the close date, the Messages view uses each message's time, and CSAT uses the day the survey was submitted.
• Never counted: history imported from your previous helpdesk, test cases, playground and demo conversations, requests sent with Write a colleague, and deleted conversations.

How fresh the numbers are

Analytics refresh about every 15 minutes and are rebuilt every night at 03:00 UTC. A ticket you just closed moves from Engaged to Resolved after its resolution is recorded (a little later if the AI is still working on it) and the next refresh runs. The helpdesk board itself is live.

Your own numbers on the helpdesk Home

Each agent's Home shows the last 30 days by default, with a period picker: Conversations (ones you replied in, noted or closed), Closed (you did the final close), First response time (your own first reply, not the AI's), Interactions (replies you sent), CSAT, and for phone teams Calls handled and Avg. talk time. Click a tile to open the conversations behind it, including the ones behind your CSAT.

Why numbers differ between pages

• AI rate vs AI Resolution Rate in Alerts. The ring counts AI auto-resolved plus AI-assisted. The alert metric under Settings → Alerts counts AI auto-resolved only.
• Headline vs chart lines. The headline follows the day conversations were created; the AI-only % and AI-assisted % lines follow the day they were resolved, older backlog included.
• CSAT card vs the tag table. The Dashboard's CSAT card and the CSAT page are the same report: the share of 4 and 5 ratings out of all ratings, by submission date. A 3 is not positive, so 2 twos, 1 three and 7 fives is 70%. The CSAT column in By Conversation Tag is the average score as a percentage of 5 (4.2 shows as 84%), by conversation date.
• Agent CSAT vs team CSAT. An agent's CSAT only covers human-resolved conversations they replied to first. AI-assisted ratings stay in the team's Assisted CSAT.
• Tag volumes add up to more than Intake. A conversation with three tags counts once under each. Noise is left out.
• Policies and Workflows count AI runs, not conversations. A conversation can have several runs, and a run using two policies shows under both.
• Dashboard vs SLAs. An SLA can pause outside business hours. Dashboard times never pause.
• Noise vs billing. Separate rules. Billing always leaves out tests, imports, spam, blacklist and bounce closes, calls that never connected and (by default) tickets that arrived closed. Out of office, marketing, duplicates and closed without response are noise here but can still be billable. See Settings → Billing.

Getting data out of Zuzana

There is no CSV, Excel or PDF export of the dashboards. Instead:

• Ask Claude through Zuzana's MCP. It queries the same metrics, lists the conversations behind them, and searches conversations for questions no dashboard answers. It also has metrics the Dashboard does not show, like first-time resolution: the share of resolved conversations where the customer only wrote once.
• Use the metrics API for a BI tool or wallboard, with an API key you can ask Zuzana support for.

Good to know

• Drafts, Policies, Workflows report on the AI itself; see Policies and Workflows. Chat Revenue keeps Sales driven by chat, Sales assisted by chat and Shoppers who chatted first apart and never adds them up.
• Side conversations count. A message to a carrier is a conversation, so it is in Intake. A request to another department is not.
• Social, SMS and connected-helpdesk conversations are not in the Channel filter, but they are in the totals and in Resolved by channel.
• Board counts are not analytics. A helpdesk column badge like "50+" holds more tickets than it shows; hover it for the exact count, up to 500.
• Agent numbers count involvement. A conversation two agents replied in counts once for each.
• CSAT appears once it is on under Settings → CSAT; the survey itself is explained in CSAT.