What the dashboard numbers mean
Updated 2026-09-19 · owner Puccha team · dashboard metrics reporting governance
The dashboard is the first page of the admin app (/c/<your-org>/dashboard). This page explains each figure on it: what is counted, over which period, whether it is a measurement or an estimate, and who in your organisation can see it. Nothing on the dashboard is sampled or modelled — every number is a count or a ratio over your organisation’s own records, except the two estimates called out below.
Under the counters the page states when the numbers were read (“Read at 14:22”), how old that reading is once it passes a minute, and a refresh link once it is more than ten minutes old — so a tab left open since the morning never reads as current.
The date range
Section titled “The date range”The picker at the top right sets the period for every windowed figure: Today · Last 7 days (default) · Last 30 days · Last 90 days · Month to date · Custom (up to 365 days). Panels that follow the picker say so in their title.
Two things to know about the period:
- Days follow your organisation’s timezone. Set it under Settings → General → Timezone (default: Asia/Bangkok). “Today” starts at midnight in that zone and each point on a chart is one day there — the same clock the “After hours” tile uses. The page names the zone under the counters. For zones with daylight-saving time the offset in force today is applied to the whole range.
- The “vs previous” comparison on the Value panel is against the period of the same length immediately before the one selected.
Figures that are not windowed are inventory counts (documents, sections, members) and the lifetime counters explained under “Counters”.
What Puccha did for you
Section titled “What Puccha did for you”This panel shows outcomes for the selected period. Every tile except the two estimates is a direct count.
| Tile | What it counts | Notes |
|---|---|---|
| Resolved by AI | Conversations that ended with a cited answer, received no thumbs-down and were not handed to a person. A conversation counts once, when it is closed or after 30 minutes of silence. | Counted by a sweep every 15 minutes, so it can lag live chat by up to about 45 minutes — the tile says so. |
| Questions answered | Questions where Puccha gave an answer rather than “I don’t have that information”. The percentage is answered ÷ all questions in the period. | When your organisation uses more than one surface (widget, LINE, team chat) the rate is also shown per surface. “Other” covers questions that arrived through the API. |
| Staff time saved | Estimate. Resolved conversations × minutes a person would have spent on each. | The inputs are shown under the figure and an owner or admin can change them (“Estimate inputs”). Defaults: 6 minutes and ฿250 per staff hour. |
| ≈ ฿ saved | Estimate. Hours saved × cost per staff hour. | Same inputs as above. The panel says so: “estimates from the inputs above, not measurements”. |
| After hours | Share of answered questions that arrived outside 09:00–18:00 in your organisation’s timezone. | The hours are fixed; the zone follows Settings → General. |
| Handed to humans | Conversations passed to a person during the period, as a share of conversations started in the period. | |
| Satisfaction | Thumbs-up as a share of all thumbs given in the period. When customers have rated conversations 1–5, the average CSAT is shown instead. | “Not rated yet” means nobody clicked a thumb in the period. |
| Leads captured | Visitors first seen in the period who left an email address or phone number. | A count only; the dashboard never shows who they are. |
| Unanswered questions | Questions Puccha could not answer that nobody on the team has dismissed yet. | Links to the Knowledge gaps page where they can be answered. |
Test conversations from the Playground are excluded from all of these.
Counters
Section titled “Counters”The two rows of cards below the Value panel.
| Card | What it counts | Period |
|---|---|---|
| Documents | Documents in the knowledge base. The small chart is documents added per day in the period. | Current |
| Queries | Questions asked since the organisation was created, on every surface and through the API. The small chart shows the same questions per day in the period. | Lifetime, within retention (see below) |
| Conversations | Conversations since the organisation was created. Playground tests are not counted. | Lifetime |
| Members | Active members — the same figure the Members page shows as seats used. Pending invitations are not counted. | Current |
| Coverage Score | Share of questions in the period that received an answer. Green ≥ 80 %, amber ≥ 60 %, red below. Shows “—” when there were no questions in the period. | Selected period |
| Feedback | Thumbs-up and thumbs-down since the organisation was created. Playground tests are not counted. | Lifetime |
| Indexed sections | Passages the knowledge base is split into for search. A size signal, not a quality one. | Current |
| Open Tickets | Tickets in the new, open or reopened status — the same number as the “Open” card on the Tickets page, which the card links to. Tickets being worked on appear under “In progress” there. | Current |
Team workload
Section titled “Team workload”Appears when at least one teammate has a ticket that is not yet resolved or closed. One row per active member, showing that person’s unfinished tickets, a red count of tickets that have used 50 % or more of their SLA, and an “OOO” tag when the member has set out-of-office. Clicking a row opens that person’s queue. Owners and admins see every member. A member whose role is limited to one workspace sees only that workspace’s members (plus the organisation’s owners and admins) and only that workspace’s tickets.
Appears when at least one ticket created in the period has an SLA policy attached.
- Tracked tickets — tickets with an SLA policy created in the period.
- First-response p50 / p95 — median and 95th-percentile time from ticket creation to the first agent reply, over tickets that have been replied to. “—” means none has been replied to yet.
- Breach rate — share of tracked tickets that missed either the first-response or the resolution target.
Below five tracked tickets the percentiles are shown in grey with a note; they settle as the sample grows.
Appears when there was traffic in the period. Puccha avoids calling the AI model when it already has the answer (a repeat question within 15 minutes), when the message is only a greeting, or when the question is clearly outside the knowledge base.
- $ saved — estimate. Each avoided model call is valued at US$0.0015; the rate and the list-price date are printed under the figure.
- LLM-skip % — avoided calls as a share of eligible turns, overall and per surface (
/api/answer,/api/chat). - Embedding cache — how often the search step reused a stored embedding instead of computing one.
- Clear cache — owner or admin with a fresh MFA check. Cached answers for your organisation are dropped and regenerated on demand; propagation across regions takes about a minute. The action is recorded in the audit log.
Query volume, Quick actions, Recent documents
Section titled “Query volume, Quick actions, Recent documents”Query volume charts questions per day in the period, from the same records as the “Queries” card. Recent documents lists the five most recently updated documents with their indexing status. Quick actions are links.
AI Copilot
Section titled “AI Copilot”Appears when the team has used Copilot to draft replies in the period. Accept rate is drafts inserted into a reply ÷ drafts generated (target 40 % or better). Blocked by safety counts drafts the safety check refused, with the most common reason. Est. cost is an estimate from the tokens used, at the model’s list price (US$1 per million input tokens, US$5 per million output tokens); the rate and its date are printed under the figure. Figures stay in US dollars because that is the currency the model provider invoices in — converting them would add an exchange rate the page could not vouch for.
Who can see what
Section titled “Who can see what”| Owner · Admin | Editor | Viewer | |
|---|---|---|---|
| All panels above | ✓ | ✓ | ✓ |
| “Your day” strip (your own activity in the period) | ✓ | ✓ | — |
| Edit estimate inputs | ✓ | — | — |
| Clear cache (with MFA) | ✓ | — | — |
Every figure is confined to your organisation. Members with a workspace-scoped role see organisation-wide totals on this page (the counters carry no content and no personal data) but only their own workspace in Team workload.
What is stored, and for how long
Section titled “What is stored, and for how long”- The dashboard reads counts and timestamps. It never displays the text of a customer’s question, a message, or a visitor’s contact details. Question logs store a one-way hash of the question, not the question itself.
- Question logs, audit history and closed tickets are kept for 90 days on Free and Team plans, 365 days on Business, and indefinitely on Enterprise. Lifetime counters therefore cover at most that window — the page prints the window under the counters, and if a chosen date range reaches further back it says from which date the logs have been purged. Documents, conversations, members and feedback are kept until deleted.
- Two actions on this page write data: saving estimate inputs and clearing the cache. Both are recorded in the audit log with who did it and when.
Known limitations
Section titled “Known limitations”- For a timezone with daylight-saving time, a date range that spans the switch is bucketed at today’s offset, so days at the far end can be off by one hour.
- The “$ saved” rate per avoided call is an amortised estimate, not a measured mean.