Qontak | Chatbot & AI | Unified Agent Quality Scorecard — Phase 3: Unified Analytics Report

Template: PHASE PRD v1.2 · Companion to PRD Section Reference v1.5 + Hierarchy v1.0 Note: Phase 3 of the Unified Agent Quality Scorecard initiative. Hybrid build: aggregate report = embedded Metabase (reusing the existing Service Quality Score Report infra); per-conversation drill-down = the Phase 2 in-room panel.

---

HEADER BLOCK

Field	Value
PM	Dimas Fauzi Hidayat
PRD Version	1.3
Status	DRAFT
PRD Type	PHASE
Epic	QC-XXXXX — add once Epic is created
Squad	BOT — Bot, AI & Automation (with Data/Analytics)
RFC Link	Pending — RFC to follow via rfc-starter
Figma Master	Pending — report host page (Stitch prompt in Appendix A); Metabase dashboard design owned by Data/Analytics
Anchor	Qontak | Chatbot & AI | Unified Agent Quality Scorecard — ANCHOR
Labels	epic:qontak-chatbot-ai | module:chatbot-ai | feature:unified-agent-scorecard
Last Updated	2026-06-19

---

Table of Contents

• HEADER BLOCK
• 2. CONDITIONAL BLOCK: PHASE CONTEXT
• 3. One-liner + Problem
• 4. What Happens If We Don't Ship This Phase
• 5. Target Users + Persona Context
• 6. Non-Goals
• 7. Constraints
  • 7.1 Data Lifecycle
• 8. New Features
• 9. API & Webhook Behavior
• 10. System Flow + User Stories + ACs
  • 10.1 System Flow
  • 10.2 User Stories
• 11. Rollout
• 12. Observability
  • 12.1 Post-Launch Monitoring Cadence
• 13. Success Metrics
• 14. Launch Plan & Stage Gates
• 15. Dependencies
• 16. Key Decisions + Alternatives Rejected
• 17. Open Questions
• Appendix A — Stitch UI Prompt
• PRD CHANGELOG

---

2. CONDITIONAL BLOCK: PHASE CONTEXT

Field	Detail
Anchor PRD	Qontak | Chatbot & AI | Unified Agent Quality Scorecard — ANCHOR
Phase	Phase 3 of 8
Phase Goal	Ship the AI + human Agent Scorecard Report — KPI cards, per-agent trends, conversation drill-down, PDF/CSV export.
Prior phases	Phase 1 (shipped): scorecard settings + rubric config. Phase 2 (shipped): per-conversation AI auto-scoring (per actor/segment, with veto) + the in-room Scorecard panel. This phase aggregates the Phase 2 scores and reuses the Phase 2 panel as its drill-down.
Deferred to later phases	Validation/testing harness → Phase 4; go-live gate → Phase 5.
Cross-phase dependencies	(1) Phase 2 scores are the report's data source. (2) The Phase 2 in-room panel is the drill-down target. (3) The existing embedded-Metabase infra (get_metabase_report / get_iframe_url) is reused. (4) Data/Analytics builds the dashboard + warehouse pipeline.

---

3. One-liner + Problem

One-liner: Give QA leads one report — AI and human agents on the same quality lens — with trends, drill-down to any conversation's score, and PDF/CSV export.

Problem: Phase 2 now scores every AI conversation and shows it in the in-room panel, but there is no aggregate view — a QA lead still cannot see team-level trends, compare AI against human, or hand leadership a report (the Enterprise NPS ask was verbatim: "team-level trends over time, SLA compliance dashboards, PDF export"). Today's reporting is embedded Metabase (the Service Quality Score Report), but it covers only the legacy human scorecard and carries no AI quality. Without this phase, AI scores stay locked to one-conversation-at-a-time review and the paid Scorecard still cannot prove ROI to a buyer.

---

4. What Happens If We Don't Ship This Phase

• The Enterprise reporting ask stays unanswered — a named NPS request since early 2026; every renewal cycle without it, those accounts keep assembling reports by hand and the dissatisfaction compounds.
• AI quality can't be proven at the account level — blocks the Q3 2026 ROI story for renewals/upsell; the paid Scorecard differentiator stays unprovable to buyers indefinitely.
• Phase 2's scores are underused — the per-conversation data accrues daily but stays invisible in aggregate, so the value of P1+P2 is capped until this ships.

---

5. Target Users + Persona Context

Primary Persona: QA Lead / Supervisor

Field	Detail
Role	QA Lead or Supervisor accountable for conversation quality across human and AI agents
Goal	See team-level quality trends, compare AI vs human, and drill into any conversation's score — in one place
Pain	Phase 2 gives per-conversation scores but no aggregate; no trend, no comparison, no export
Workaround	Opening conversations one at a time in the inbox; manual spreadsheets for any trend

Secondary Persona: CS Manager / Team Lead

Field	Detail
Role	The manager who reports quality up to leadership
Goal	A leadership-ready report (trends + PDF) for weekly reviews — the exact NPS ask
Pain	"My management keeps asking for reports I can't generate" — no team-level trend or PDF export today
Workaround	Manually compiling numbers into slides each week

---

6. Non-Goals

1. Not the per-conversation scoring — Phase 2 produces the scores; this phase only aggregates them.
2. Not the in-room panel — Phase 2 owns it; this phase reuses it as the drill-down target.
3. Not the go-live gate — Phase 5.
4. Not the validation/testing harness — Phase 4.
5. Not a new BI/reporting engine — reuse the existing embedded-Metabase infra (Service Quality Score Report pattern).
6. Not real-time streaming — the report reflects the warehouse refresh cadence (see Open Q#3), not live.
7. No mobile — web only.
8. Not a custom report-builder / ad-hoc query UI — a fixed unified dashboard this phase.

---

7. Constraints

Field	Value
Platform	Web only — Qontak omnichannel web app
Performance	Embed-URL mint ≤ 500ms P95. Dashboard render is Metabase-side (not Qontak-controlled). Drill-down panel ≤ 2s P95 (Phase 2).
Data limits	Report reflects the 13-month aggregate retention (Phase 2 data lifecycle). Embed token: 10-minute expiry (existing pattern).
Plan scope	Professional + Enterprise only. Not Starter/Free.
Feature flag	ai_qa_unified_scorecard (report behind the same flag) + per-org dashboard config via system_preference / org settings (reusing the rollout_hologres_metabase_report pattern).
Read/write	Read: QA Lead/Supervisor (team scope), Bot/AI Admin (own agents), CS Manager (team). Export: same. No write — the report is read-only.

7.1 Data Lifecycle

Artifact Type	Retention Period	Cleanup Trigger	User-Visible Effect
Metabase embed token (signed JWT)	10 minutes	exp claim TTL	Link expires; re-minted on next load
Export file (PDF/CSV)	Ephemeral — generated on demand by Metabase	Not persisted by Qontak	None
Aggregate score data (in warehouse)	13 months (mirrors Phase 2)	Upstream TTL	Trend shows up to 13 months

---

8. New Features

Feature: Agent Scorecard Report (host page)

Field	Detail
URL	/chatbot/agent-scorecard/report
Access	QA Lead/Supervisor (team scope), Bot/AI Admin (own agents), CS Manager (team)

Component Tree:

Component	Parent	Purpose
AgentScorecardReportPage	—	Native host page for the embedded report
ReportHeader	AgentScorecardReportPage	Title, plan badge, "data as of" timestamp (warehouse freshness)
MetabaseEmbedFrame	AgentScorecardReportPage	Signed iframe — the aggregate dashboard: KPI cards (avg score, pass rate, containment, hallucination rate), per-agent table (AI/human badge + trend), filters (agent type, date, channel)
DrilldownLauncher	AgentScorecardReportPage	Handles a Metabase cell click-behavior → opens the native Phase 2 Scorecard drill-down for the selected conversation/actor

The KPI cards, per-agent table, filters, and export all live inside the Metabase dashboard (owned by Data/Analytics). The native host provides the page chrome, the signed-embed plumbing, and the drill-down hand-off.

UI States:

State	Description
Empty	Dashboard shows "No scored conversations yet" (Metabase); host shows a hint + link to Scorecard settings.
Loading	Skeleton while minting the embed URL and the iframe loads.
Error	"Couldn't load the report. Try again." + Retry (embed mint failed). Log: report_embed_failed.
Success	Embedded dashboard rendered; drill-down available.

Figma: Pending host chrome (Stitch prompt in Appendix A). Metabase dashboard layout is owned by Data/Analytics.

📊 UI State Diagram — Agent Scorecard Report (host page)

stateDiagram-v2
    [*] --> Loading: Open report page (mint embed URL)
    Loading --> Success: Dashboard rendered
    Loading --> Empty: Org has no scored conversations
    Loading --> Error: Embed mint / iframe load fails
    Error --> Loading: Retry
    Empty --> [*]: Link to Scorecard settings
    Success --> [*]: Drill into conversation / export

---

9. API & Webhook Behavior

Behavior 1: Get the unified scorecard report embed URL

Field	Detail
Entity affected	Signed Metabase embed token/URL for the unified scorecard dashboard
Triggered by	User opens the Agent Scorecard Report page
Information passed	Org id; the dashboard config (group_code/code or dashboard_id) resolved via system_preference / org settings / env — reusing get_metabase_report / get_iframe_url
Expected behavior	Mint a signed JWT embed URL scoped to the org (10-minute exp) for the unified dashboard; return url + expires_in
Failure behavior	• Missing dashboard config → 422 (existing behavior). • Mint fails → error surfaced as the report Error state, report_embed_failed logged. • Expired token → re-minted on reload.

Behavior 2: Deep-link from a dashboard cell into the native drill-down

Field	Detail
Entity affected	Navigation from a Metabase dashboard cell → the native Phase 2 Scorecard drill-down
Triggered by	User clicks a conversation/agent row in the embedded dashboard (Metabase click-behavior → URL)
Information passed	room_id / score_record_id (+ actor) carried in the target Qontak URL
Expected behavior	The Qontak app opens the native Phase 2 in-room Scorecard panel (or drill-down route) for that conversation/actor
Failure behavior	• Unknown/expired record → "not found / expired" state. • Cross-frame navigation blocked → fallback to opening the conversation in the inbox.

Claude resolves during RFC: HTTP method, path, request/response JSON schema, the Metabase click-behavior URL contract.

---

10. System Flow + User Stories + ACs

10.1 System Flow

Flow: Open the Unified Report and Drill Into a Conversation Type: User Journey + API Sequence

1. A QA Lead opens /chatbot/agent-scorecard/report.
2. FE requests the embed URL; BE resolves the org's unified-scorecard dashboard config (system_preference / org settings / env) and mints a signed JWT scoped to the org (10-min exp).
3. FE embeds the Metabase dashboard iframe → KPI cards + per-agent table + filters render.
4. The QA Lead filters (agent type AI/human, date, channel) inside the dashboard.
5. The QA Lead clicks a conversation/agent row → Metabase click-behavior fires a URL into the Qontak app (carrying room_id/score_record_id).
6. The app opens the native Phase 2 Scorecard drill-down (9-metric + reasons + sources + veto + transcript) for that conversation/actor.
7. The QA Lead exports the dashboard (Metabase PDF/CSV) for leadership.
8. Failure branch — if the embed mint fails (missing config / error), the host shows "Couldn't load the report" + Retry and logs report_embed_failed.

📊 System Flow — Report + Drill-Down

sequenceDiagram
    participant QA as QA Lead
    participant FE as Qontak FE
    participant BE as Qontak BE
    participant MB as Metabase
    QA->>FE: Open report page
    FE->>BE: Request embed URL
    BE->>BE: Resolve org dashboard config + mint signed JWT (10-min exp)
    BE-->>FE: url + expires_in
    FE->>MB: Embed iframe (org-scoped dashboard)
    MB-->>QA: KPI cards + per-agent table + filters
    Note over BE,FE: Mint fails → "Couldn't load" + Retry, log report_embed_failed
    QA->>MB: Click a conversation row
    MB->>FE: Click-behavior URL (room_id / score_record_id)
    FE->>QA: Open native Phase 2 drill-down (metrics + reasons + sources)
    QA->>MB: Export (PDF / CSV)

10.2 User Stories

[P3-S01] — View the unified AI + human scorecard report

User Story	As a QA Lead, I want one report showing AI and human agents on the same quality lens with trends, so that I can see team performance and compare at a glance.
Before State	Only the legacy human Service Quality Score Report (embedded Metabase) exists — no AI quality; Phase 2 per-conversation scores have no aggregate view.
After Delta	A new report page embeds the unified Metabase dashboard (org-scoped signed iframe) with KPI cards, a per-agent AI/human table, trends, and filters.
Importance	Must Have
Mockup / Technical Notes	Figma: Pending — Appendix A Stitch prompt Data Fields: • organization_id (string, required) — Auth session • dashboard_config (object, required) — from system_preference / org settings / env • user_role (string, required) — Auth session Technical Notes: Reuses get_metabase_report / get_iframe_url; signed JWT, 10-min exp, org-scoped.
Acceptance Criteria	— Happy Path — • AC-1: Given a QA Lead with the flag ON opens the report page, when it loads, then BE mints a signed org-scoped embed URL and the dashboard renders in the host. • AC-2: Given the dashboard, when the QA Lead filters by agent type = AI, then the dashboard scopes to AI agents (filter handled in Metabase). — Edge — • AC-3: Given an org with no scored conversations yet, when the report loads, then the dashboard shows an empty state and the host shows a link to Scorecard settings. — Error / Unhappy Path — • ERR-1: Given the embed mint fails (missing dashboard config or error), when the page loads, then "Couldn't load the report. Try again." + Retry is shown and report_embed_failed is logged. • ERR-2: Given the embed token expired (10-min), when the user returns, then the FE re-mints transparently on reload. — Permission Model — • CAN: QA Lead/Supervisor (team), Bot/AI Admin (own), CS Manager (team). • CANNOT: end CS agents. • Unauthorized: route not in nav; direct access returns the standard not-authorized view. — UI States — • Loading: skeleton while minting + iframe loads. • Empty: dashboard empty state + settings link. • Error: as ERR-1. • Success: embedded dashboard rendered. — Negative Scenarios — • NEG-1: Given a Starter/Free org, when a user navigates to the report URL, then it is not available (plan-gated; not rendered).

Dependencies: Phase 2 scores; Data/Analytics dashboard — see S15.

---

[P3-S02] — Drill from the report into a conversation's score

User Story	As a QA Lead, I want to click a row in the report and see that conversation's full score, so that I can investigate why an agent scored low.
Before State	No path from the aggregate to a per-conversation score; the drill-down only exists in the inbox panel (Phase 2).
After Delta	A Metabase cell click-behavior deep-links into the native Phase 2 Scorecard drill-down for the selected conversation/actor.
Importance	Must Have
Mockup / Technical Notes	Figma: Reuses the Phase 2 panel Data Fields: • room_id (string, required) — clicked row • score_record_id (uuid, required) — clicked row • actor_id (string, required) — clicked row Technical Notes: Metabase click-behavior → Qontak URL (see Open Q#1 for the cross-frame deep-link mechanism).
Acceptance Criteria	— Happy Path — • AC-1: Given the report is open, when the QA Lead clicks a conversation row, then the native Phase 2 Scorecard drill-down opens for that conversation/actor with the 9-metric breakdown + reasons + sources + veto. • AC-2: Given a multi-actor room, when the QA Lead drills in, then the actor whose row was clicked is preselected. — Edge — • AC-3: Given the transcript snapshot is past 90-day retention, when drilled in, then scores render but the transcript reads "expired". — Error / Unhappy Path — • ERR-1: Given the target record is unknown/expired, when drilling in, then a "not found / expired" state is shown. • ERR-2: Given cross-frame navigation is blocked, when the cell is clicked, then the app falls back to opening the conversation in the inbox. — Permission Model — • CAN: QA Lead/Supervisor (team), Bot/AI Admin (own). • CANNOT: end CS agents (cannot open others' scores). • Unauthorized: link does not resolve. — UI States — • Loading: opening drill-down. • Empty: N/A. • Error: "not found / expired". • Success: drill-down shown. — Negative Scenarios — • NEG-1: Given an end CS agent, when a drill-down URL is opened, then it does not resolve to another agent's score.

Dependencies: P3-S01; Phase 2 in-room panel.

---

[P3-S03] — Export the report (PDF / CSV)

User Story	As a CS Manager, I want to export the report to PDF or CSV, so that I can share AI + human quality with leadership.
Before State	No export of AI quality; numbers are compiled into slides by hand.
After Delta	Metabase's native export (PDF/CSV) of the filtered dashboard.
Importance	Should Have
Mockup / Technical Notes	Figma: Metabase-native export Data Fields: • filter_scope (object) — current agent type/date/channel • format (enum pdf|csv) — user choice
Acceptance Criteria	— Happy Path — • AC-1: Given the report is open with filters applied, when the CS Manager exports as PDF, then a PDF of the filtered view is generated. • AC-2: Given CSV is chosen, when exporting, then a CSV of the underlying aggregate is generated. — Edge — • AC-3: Given no data in the filtered range, when exporting, then an empty-report notice is shown instead of a broken file. — Error / Unhappy Path — • ERR-1: Given the export fails (too large / Metabase error), when exporting, then an error + Retry is shown and report_export_failed is logged. — Permission Model — • CAN: QA Lead/Supervisor, Bot/AI Admin, CS Manager. • CANNOT: end CS agents. • Unauthorized: export control not rendered. — UI States — • Loading: "generating…". • Empty: N/A (covered by AC-3). • Error: as ERR-1. • Success: file ready. — Negative Scenarios — • NEG-1: Given a Starter/Free org, when the report is unavailable, then no export control exists.

Dependencies: P3-S01.

---

11. Rollout

Field	Value
Feature flag	ai_qa_unified_scorecard — report behind the same flag; dashboard resolved per-org via system_preference / org settings (the rollout_hologres_metabase_report pattern)
Stage 1	Internal QA: 3–5 internal accounts — validate dashboard data vs the in-room panel + drill-down deep-link
Stage 2	Closed beta: TransGo, Talenta LMS + 3 partners
Stage 3	All Professional + Enterprise on request
GA	All Professional + Enterprise (flag on)
Backward compat	Yes — the legacy Service Quality Score Report (human-only) is unaffected; this is a separate unified dashboard
Migration	None to existing records. New: a unified Metabase dashboard + the warehouse pipeline feeding it (Data/Analytics).

---

12. Observability

Key Events:

Event Name	Trigger	Properties
report_viewed	Report page opened	org_id, user_role, dashboard_id
report_embed_failed	Embed URL mint / iframe load failed	org_id, reason
report_drilldown_opened	A conversation drill-down opened from the report	org_id, room_id
report_export_generated	Export produced	org_id, format
report_export_failed	Export failed	org_id, reason

Field	Detail
Dashboard owner	Bot, AI & Automation (squad: BOT) + Data/Analytics (warehouse + Metabase)
Alert 1	report_embed_failed rate > 5% of views in 1h → Slack: #bot-ai-oncall
Alert 2	Warehouse data freshness lag > the agreed threshold (Open Q#3) → Slack: #data-oncall

12.1 Post-Launch Monitoring Cadence

Field	Detail
Review cadence	Weekly for the first 4 weeks post-GA, then monthly
Owner	Dimas Fauzi Hidayat (PM) + BOT + Data/Analytics
Review scope	report_viewed, report_embed_failed, report_drilldown_opened, report_export_generated, data freshness lag
Trigger threshold 1	report_embed_failed > 5% week-over-week → investigate embed config / Metabase
Trigger threshold 2	Report adoption < 20% of enrolled accounts after 4 weeks → revisit entry-point discoverability
Rollback consideration	If embed failures persist > 48h, PM disables the report flag for affected orgs pending fix.

---

13. Success Metrics

Adoption & Usage:

Metric	Definition	Baseline	Target
⭐ Report adoption	% of enrolled Pro+Ent accounts viewing the report monthly	N/A — new surface	≥50% within 60 days of GA
Drill-down usage	% of report sessions that open a conversation drill-down	N/A	≥30% within 60 days of GA

Quality & Accuracy:

Metric	Definition	Baseline	Target
Report–panel consistency	Sampled checks where the dashboard aggregate matches the in-room panel scores	N/A	100% match on the alpha validation sample

Efficiency & Impact:

Metric	Definition	Baseline	Target
Leadership exports	# report exports per month (the NPS ask)	0 — manual spreadsheets	Tracked; ≥1 export/month per active CS Manager within 90 days of GA

---

14. Launch Plan & Stage Gates

Stage	Audience	Duration	Success Gate to Advance	Owner
Internal Alpha	3–5 internal QA accounts	2 weeks	Dashboard aggregate matches the in-room panel on the validation sample; drill-down deep-link works; report_embed_failed ≤2%	PM + QA + Data
Closed Beta	TransGo, Talenta LMS + 3 partners	2 weeks	≥1 export by a CS Manager; no P0; data freshness within agreed lag	PM + BOT
Open Beta	All Pro+Ent on request	2 weeks	Report adoption ≥30% of enrolled; drill-down used; no P0 for 2 weeks	Eng Lead
GA	All Pro+Ent	Ongoing	All Open Beta gates sustained 2 weeks; PMM launch approved	PM + PMM

---

15. Dependencies

Dependency	Owning Team	Deliverable Needed	Blocking?
Unified Metabase dashboard + warehouse pipeline	Data / Analytics	A dashboard (KPIs, per-agent AI/human table, trends, filters) + the Hologres ETL feeding it from the agent_scorecard AI+human scores	YES
Phase 2 scoring (data source)	BOT (Phase 2)	Per-conversation AI scores in production	YES
Phase 2 in-room panel (drill-down target)	BOT (Phase 2)	The native drill-down the report links into	YES
Existing Metabase embed infra	BOT (existing)	BE get_metabase_report / get_iframe_url and the live FE pattern hub-chat features/report/quality_scoring/QualityScoringPage.vue calling GET /api/gpt/v1/reports?group_code=metabase_report&code=sqs — cloned with a new code	NO — already shipped
Metabase click-behavior → deep-link	Data/Analytics + FE	Dashboard cell click → Qontak URL carrying room_id/score_record_id	YES
Design / UX	Design squad	Report host-page chrome + nav entry	YES

---

16. Key Decisions + Alternatives Rejected

8a — Decisions Made

Date	Decision	Rationale
2026-06-19	Hybrid build: aggregate report = embedded Metabase; per-conversation drill-down = the native Phase 2 panel	Reuses the shipped reporting infra (consistent with the Service Quality Score Report) and avoids reinventing the drill-down; Metabase can't render per-conversation reasons/source-links/veto, but the Phase 2 panel already does
2026-06-19	Reuse the existing embed-iframe pattern (signed JWT, org-scoped, per-org dashboard via system_preference)	No new reporting engine; lowest build cost; matches current behavior
2026-06-19	Drill-down via a Metabase cell click-behavior → Qontak deep link → native panel	Bridges the embedded dashboard to the rich in-app drill-down without rebuilding it in Metabase
2026-06-19	Clone the live QualityScoringPage.vue pattern (Metabase iframe via GET /api/gpt/v1/reports?code=…) with a new report code	Verified in cloned hub-chat — the embed pattern is already shipped for the Service Quality Score report; lowest build cost and proven

8b — Alternatives Rejected

Alternative	Why Rejected	Date
Native in-product report (build the whole screen + API)	More build cost; duplicates the existing reporting infra for no added value over Metabase aggregates	2026-06-19
Metabase-only (including the drill-down)	Metabase can't elegantly render the 9-metric reasons, cited source links, veto flags, or transcript — poor drill-down UX	2026-06-19
Adopt a new BI tool	Unjustified — Metabase is the established Qontak reporting standard	2026-06-19

---

17. Open Questions

#	Type	Question	Owner	Deadline
1	Risk	How does the Metabase dashboard cell deep-link into the native drill-down across the iframe boundary (click-behavior → URL → Qontak route)? Mitigation: prototype the click-behavior→URL in Internal Alpha; fallback = open the conversation in the inbox (P3-S02/ERR-2).	Data/Analytics + FE	2026-07-15
2	Open Question	The live report fetches code=sqs (Service Quality Score = the human scorecard). The unified AI+human report = a new report code (e.g. ai_sqs) → a new Metabase dashboard. Confirm the dashboard id/code with Data/Analytics.	Data/Analytics	2026-07-15
3	Risk	Warehouse ETL cadence (real-time vs nightly) determines whether the report lags the in-room panel. Mitigation: define an acceptable lag (proposed ≤1h) and show a "data as of" timestamp; confirm pipeline cadence with Data.	Data/Analytics	2026-07-15
4	Assumption	The existing get_metabase_report / system_preference config can host a second (unified) dashboard without code changes.	BOT	2026-07-01

---

Appendix A — Stitch UI Prompt

Generated for the native host chrome only — the dashboard interior is owned by Data/Analytics in Metabase. Use in Stitch; hand to Design.

=== SHARED PREAMBLE ===
Product: Mekari Qontak — Omnichannel
Users: QA Lead / Supervisor, CS Manager, Bot/AI Admin
Design tone: Enterprise B2B SaaS — clean white surfaces, purple accent; match the existing Qontak reports shell
Persistent UI: left icon rail + top bar
=== END PREAMBLE ===

#	Screen	Stitch Prompt (paste in full after the preamble)
1	Agent Scorecard Report — host page	Screen: the Agent Scorecard Report page that hosts an embedded Metabase dashboard. Purpose: QA Lead/CS Manager views unified AI + human quality and drills into a conversation. Components: a report header (title "Agent Scorecard Report", plan badge, a small "data as of [timestamp]" label); a large embedded-dashboard area (show a representative Metabase-style dashboard: 4 KPI cards across the top — avg score, pass rate, containment, hallucination rate — then a per-agent table with an AI/human badge column and a trend sparkline, with filter chips for agent type / date / channel); the rows are clickable to drill down. Generate states: Loading (skeleton header + skeleton dashboard block); Empty ("No scored conversations yet" inside the dashboard + a "Configure in Scorecard settings" link in the host); Error ("Couldn't load the report. Try again." + Retry); Success (full dashboard). Do NOT include: an in-app chart builder, the per-conversation drill-down panel (that's the Phase 2 panel, opened on row click), mobile layout.

---

PRD CHANGELOG

Version	Date	By	Section	Type	Summary
1.0	2026-06-19	Claude	All	CREATED	Phase 3 PRD (Unified Analytics Report) — hybrid: embedded Metabase aggregate (reusing the Service Quality Score Report infra) + Phase 2 native drill-down.
1.1	2026-06-19	Claude	S6, S1b	MODIFIED	Post-score polish: added the report host-page UI-state diagram and time horizons to "What Happens If We Don't Ship".
1.2	2026-06-19	Claude	CB	MODIFIED	Phase count 7→8 (added the parked Phase 8: multi-agent + selectable scorecard to the ANCHOR).
1.3	2026-06-19	Claude	S15, S16, S17	MODIFIED	hub-chat grounding: cited the live QualityScoringPage.vue + code=sqs embed pattern (cloned with a new code); answered Open Q#2 with the new-report-code mechanism + a decision.