Ratings and Reports

A visitor sees Did this answer your question? after a conversation is resolved. The default is thumbs up or down. In Settings → Project → Surveys, a lead can choose thumbs or 1–5, English/Spanish/French, delay, lifetime and whether to show widget surveys. Each new resolution keeps its choices; later settings changes do not change an earlier cycle. A rating and optional comment (up to 1,000 characters) belong to that resolution. The server refuses responses before the delay or after expiry. Stop widget surveys is the visitor’s preference for this workspace and persists across cycles; it does not change email preferences. Only the conversation's current owning visitor session can submit a rating. A visitor can change the rating; each change remains in history. The usual widget action rate limit applies. Operators see the current rating and comment in the conversation panel's Ticket → History.

A new visitor message reopens the conversation. Its next resolution has a separate outcome and rating. A rating for an earlier cycle cannot be submitted while the conversation is open or against a later resolution. Contact erasure removes comments from both the current rating and change history.

Outcomes

Each resolution records its evidence at the time it occurs. Confirmed operator-sent AI or paired-local-agent drafts make the cycle assisted, including drafts edited after adoption. Assisted takes precedence when a cycle also has human-written replies. Otherwise confirmed operator-written answers make the cycle human, even when other replies have unknown provenance. AI requires every visible answer to be confirmed AI, a complete cycle history, and no person handoff. Missing evidence produces unknown. Notes, provisional replies, context insights and the workspace's away message are not answers.

Draft provenance travels with the composer through edits, suggestion changes and undo/redo. The server validates the stored completed draft against the workspace, conversation and current operator/device authority before admitting an assisted reply. External integrations that omit provenance are unknown; an operator role alone does not prove human authorship.

The additive migration preserves every original message. A historical paired-local-agent delivery receipt proves message authorship, but cannot identify which resolution cycle owned it. Historical resolution outcomes therefore remain unknown; missing operator draft or handoff records cannot prove human or AI-only resolution. The old current resolved_at does not establish the first historical resolution. Such first-resolution durations remain unavailable. Newly created native conversations record full lifecycle history going forward.

Workspace calendar and cohort

Reports uses Project → Timezone, the workspace-owned IANA timezone, with UTC as the initial default. An operator's profile timezone does not change the report. Last 7, 30 or 90 days include today in that workspace. Custom ranges use an inclusive start date and an exclusive end date, across daylight-saving changes. Choose between 1 and 366 calendar days. Figures are aggregated in the database with bounded output, regardless of cohort size. CSV downloads stream bounded pages through the entire matching cohort.

The cohort consists of conversations created during that range. Both Move in and legacy-import history are excluded, including imported conversations subsequently edited. Daily volume groups this same cohort by its workspace-local creation date.

• First response: time from the first visitor message to the first subsequent visitor-visible answer. Notes, context insights and streamed/provisional text are excluded.
• First resolution: time from conversation creation to the first resolution, only when the complete initial lifecycle is known.
• Median: the middle duration, or the mean of two middle durations. P90: nearest rank, the item at ceil(0.9 × count) in ascending order. Unknown durations are omitted and each sample count is shown.
• AI resolution rate: latest AI outcomes divided by currently resolved conversations in the cohort. Unknown outcomes remain in the resolved denominator.
• Satisfaction: thumbs-up ratings divided by rated, currently resolved conversations in the cohort. Use only the latest rating for the latest resolution. Unrated conversations are omitted from the rated denominator.

Each conversation contributes once. Reopened conversations leave both rate denominators until resolved again. Earlier cycles remain recorded. Latest comments come from this same currently resolved cohort.

Conversations CSV contains the ID, creation timestamp, first-response duration, known first-resolution timestamp and duration, current open/resolved status, latest recorded outcome, rating and comment. An open conversation can retain an earlier cycle's history in the CSV, while remaining excluded from current resolution/rating rates. Summary CSV uses exactly the page's cohort, counts and percentile conventions and includes only its aggregate metrics and daily series. It contains no individual conversation records or comments, and its query count is independent of cohort size. Timestamps in CSV are UTC and durations are seconds. String cells with leading formula characters =, +, - or @ (including after whitespace), or a leading tab or carriage return, receive a leading apostrophe and all cells are quoted.

Team report

Open Reports → Team report to see workload by agent, team, channel, brand, account or tag. Choose a calendar range, team and brand, then apply the filters. Include imported history is off initially. Turning it on includes imported tickets in this panel and its export, with separate native and imported counts. It does not change the original daily-volume report above.

Incoming tickets were created in the selected range and have a message. Workload uses their current assignment and status. Resolved means currently Solved or Closed. A ticket with several tags counts under each tag; unassigned and untagged tickets have their own rows. Merged-away source tickets are excluded.

The current backlog includes unresolved tickets created before the range's end, including tickets older than its start. Its age buckets use the report's generation time. This is the backlog now, rather than a reconstruction of what the queue looked like on a past date.

First-response, next-response and resolution compliance use the stored SLA policy and business-hours calendar for each ticket. They include every recorded cycle. The displayed numerator is commitments met; the denominator is met plus breached. Overdue active commitments count as breached. Pending commitments that have not breached and tickets without SLA records are shown separately. Calendar pauses, holidays and due times come from the SLA records; opening a report does not change them. The panel shows the oldest and newest record times so a lead can judge freshness.

Reopen rate counts distinct tickets that returned to an active status after a recorded resolution, divided by tickets with a known resolution. Repeated reopenings also appear as an event count. Imported history without an observed native resolution boundary is unknown. Import initialization and ordinary field edits do not establish that missing history. Unknown imported history is excluded from both halves of the reopen rate; raw recorded event counts remain inspectable. Handle time adds recorded timer sessions that overlap the range, including work on older matching tickets. Concurrent operators' work counts separately. A manual duration correction is split proportionally across the original session's dates. Group handle time by agent, request form or the visible custom field named Intent. A restricted Intent field remains hidden from a role that cannot read it. Missing form, intent or timer history is shown without estimating work.

Team summary CSV contains every metric, grouping, numerator, denominator and definition from this panel, without sampling. It uses the same filters and formula-safe CSV cells. Report access requires report viewing permission; downloads also require data export permission and a recent security check. Collaborators cannot read reports or timers.

Scheduled CSV

A lead with report and export permission can open Create a report schedule in the team report. Choose daily or weekly delivery, a local hour, filters and workspace teammates. Daily reports cover the previous complete local day; weekly reports cover the previous seven complete local days. The selected workspace timezone controls the boundaries, including daylight-saving changes. A missing local delivery hour runs at the first valid later hour; a repeated hour has one delivery window.

Recipients must be current workspace members with both report and data export permission. Customer addresses, including imported customers, cannot be selected. Each recipient's current role is checked again when the report is prepared and sent. Removing access prevents delivery; removing the creator's access pauses the schedule. Reports use each recipient's field visibility. Pause, resume or delete a schedule from its list.

Each recipient gets at most one send attempt for each schedule window. A rejected or uncertain send is recorded and is never automatically replayed. Local development suppresses external mail. Installation export/import preserves report settings, timers, links and history; restored schedules are paused and pending sends cannot restart. Contact erasure removes the contact's timers and problem preview data.

Workload can also be grouped by Account, using each requester’s current explicitly linked company. Unlinked requesters have a No account bucket. Email domains never assign accounts. Account rows include complete incoming, resolved and backlog totals and appear in manual and scheduled team CSV exports. The team report API (GET /api/dashboard/projects/:projectId/reports/team) and both manual and scheduled team CSVs include complete breakdowns for agent, team, channel, brand, account and tag. Each row publishes response and first-resolution sample counts, median and nearest-rank p90 (milliseconds), latest-cycle CSAT positive/rated counts, all three persisted SLA met/breached/pending counts and denominators, recorded AI closures and failed/uncertain generation attempts. Reopen rows publish their known-resolution denominator, distinct reopened tickets and event count; workload rows also include the current backlog’s oldest age and five age buckets for each dimension. Missing measurements stay null or unknown. AI handed-off counts tickets with a causal receipt committed in the same atomic batch as a successful model-requested human transfer. Historical assignments without this receipt remain unattributed. Assignment handoffs separately count any observed ticket assignment; AI-authored closure does not establish containment or verified customer recovery. Latest-cycle CSAT here includes reopened cycles; the original daily-volume satisfaction rate above remains limited to currently resolved conversations. Use Report metric in the existing team table to inspect workload, backlog age, timing samples, satisfaction, each SLA commitment, reopened requests or recorded AI outcomes. Group by applies each metric to agent, team, channel, brand, account or tag. The displayed samples, numerators, denominators, pending counts and unknowns match the API and team CSV. Wide tables scroll within the card. Procedure-version attribution is pending.

AI containment

Choose Report metric → AI containment in the existing team table. The API and manual/scheduled team CSVs expose the same ai.containment figures for all six groupings.

Participating tickets have an admitted automatic-reply generation, a confirmed AI answer or a causal AI handoff. Draft suggestions and human-only tickets do not establish participation. Contained requires a currently Solved or Closed, AI-only ticket with complete native history and no human/assisted answer, assignment handoff, reopening, unknown answer or noncompleted automatic generation anywhere in its history. A recorded human answer, handoff or reopening makes a participating ticket Not contained. Other participating work stays Unknown, including open, queued, failed, uncertain, erased and incompletely observed history. Imported history cannot establish complete native containment.

The containment rate is contained tickets divided by all participating tickets, including unknowns. Each ticket contributes once per group; zero participation has no rate. This describes observed AI-only service, not technical verification that the customer's issue was fixed. Customer confirmed separately counts contained tickets whose latest resolution has an explicit customer recovery answer. A positive satisfaction score alone never provides that confirmation; a reopen removes prior containment and current confirmation.

Satisfaction results distinguish the signed response link’s issuance surface (email or portal) from the invitation’s email transport. A link remains bound to the same customer and resolution cycle, and email/portal use the same invitation and single-use revision. Widget responses remain identified as widget feedback. Previously unrecorded response surfaces appear as unknown; existing signed links remain valid. The additive 244_csat_response_surface.sql migration preserves existing feedback and adds this metadata without sending invitations.

Reports across separate brands

The native API GET /api/management/projects/:projectId/brands/reports accepts an explicit comma-separated projects selection containing this workspace and up to 19 others, plus the ordinary report period/start/end parameters. Every selected brand requires current membership and report viewing permission. The response includes each brand's metrics and timezone, additive volume/resolution counts and correctly weighted AI/CSAT rates. It excludes customer identities and comments, does not combine medians or percentiles, and explains the selected scope. Revoked or inaccessible brands refuse the entire request rather than silently omitting data. Use the individual brand report door for CSV exports. This API does not grant shared administration or move customers between brands.