Qontak | AI Agent | Knowledge — ANCHOR
ANCHOR PRD — the initiative master index. It orchestrates all knowledge-source phases beneath it and carries no acceptance criteria of its own (ACs live in each phase PRD). Reconciled against the actual codebase (
chatbot,chatbot-fe,qontak-designer).Scope: AI Agent Knowledge = the sources the Qontak AI Agent learns from. Static sources (PDF, Website/URL) already ship; this umbrella adds dynamic, proprietary sources that let clients activate a high-quality knowledge base from data they already own. The differentiator is turning living conversation data into a low-maintenance, self-updating knowledge asset.
HEADER BLOCK
| Field | Value |
|---|---|
| PM | Dimas Fauzi Hidayat |
| PRD Version | 1.0 |
| Status | ACTIVE |
| PRD Type | ANCHOR |
| Labels | epic:qontak-chatbot | module:ai-agent | feature:ai-agent-knowledge |
| Last Updated | 2026-06-18 |
Source Index
| Source | Goal | PRD Link | Epic | Status |
|---|---|---|---|---|
| Conversation History | Learn from selected expert agents' resolved conversations (dynamic, PII-masked, daily sync) | prds/conversation-history.md | TBD | 🔄 In Progress (Discovery) |
| PDF / Website (static) | Learn from uploaded documents and crawled URLs | — (pre-existing) | — | ✅ Shipped |
| Data-driven agent selection | Suggest expert agents by CSAT / AHT instead of manual selection | TBD | TBD | ⏳ Not started |
Status options: 📝 Draft · 🔄 In Progress · ✅ Shipped · ⏸ Paused · ❌ Cancelled Static PDF/URL sources predate this ANCHOR and are listed for completeness. Future phases (data-driven selection, multi-media ingestion, real-time sync, cross-division pooling) are seeded from the Conversation History "Future Considerations" — placeholders until their phase PRDs exist.
2. One-liner + Problem
One-liner: Let the Qontak AI Agent learn from proprietary, living sources — starting with high-performing human agents' conversation history — so clients get accurate, "human-like" answers without manually building a knowledge base.
Problem: Admins struggle to keep AI knowledge bases updated because official documentation (PDFs/websites) doesn't cover the edge cases or brand tone that human agents use daily. "Tribal knowledge" — how top agents solve complex problems — is locked in chat logs and lost to the AI. For Qontak360 clients with large agent teams, the bottleneck to AI adoption isn't lack of data, it's the manual effort to collect and format training sources.
3. Target Users + Persona Context
| Persona | Role | Goal | Pain | Workaround |
|---|---|---|---|---|
| Primary — Agent | Front-line agent handling daily inquiries | Get accurate, "proven" answers immediately via Airene Copilot, without leaving the Inbox | Faces misinformation / knowledge gaps for specific inquiries → slower resolution | Searches old chats / Slack, or waits for an SPV to be free |
| Primary — SPV / Admin | Supervisor / Admin who sets up AI knowledge | Activate a high-quality knowledge base without weeks of manual documentation | Manual onboarding + constant micro-coaching of new agents; high effort to compile sources | Manually compiles PDFs/FAQs; coaches agents 1:1 |
Target profile: Qontak360 clients using the AI Agent — active companies with large agent teams (>100) or high transaction volumes (e.g., tax/PPN) that demand high accuracy.
4. Success Metrics (initiative-level)
⭐ Primary KPI: Conversation-History feature adoption
- Definition: % of active Qontak360 companies that have configured ≥1 dynamic (Conversation History) knowledge source
- Baseline: N/A — new capability
- Target: ≥ 30% of AI-Agent-active Qontak360 companies within 90 days of GA
Quality: AI resolution rate uplift
- Definition: Reduction in "Unanswered Question" report count for clients who move from static-only (PDF/URL) to combined (static + Conversation History) sources
- Baseline: N/A — measured per-client pre/post
- Target: Measurable reduction within 60 days of a client enabling the source
Efficiency: New-agent onboarding speed
- Definition: Reduction in time for a new agent to reach full productivity (avg handle time) in divisions where conversation history is active
- Baseline: N/A — new capability
- Target: Established during beta
5. Key Decisions + Alternatives Rejected
5a — Decisions Made
| Date | Decision | Rationale |
|---|---|---|
| 2026-04-13 | Agent-based conversation-history ingestion (Admin "points" the AI at selected expert agents) | Filters for quality at the source while keeping an automated "set and forget" workflow |
| 2026-04-13 | Qontak360-exclusive (Service Suite / Pro-above) | Drives tier value + stickiness; client's indexed brand voice raises switching cost |
| 2026-04-13 | Reuse the existing "Training Source" UI pattern | Reduces dev time and keeps consistency with current knowledge management |
5b — Alternatives Rejected
| Alternative | Why Rejected | Date |
|---|---|---|
| Full history ingestion (every resolved conversation, company-wide) | High noise & risk — includes low-performers, junk chats, outdated workflows → poor accuracy | 2026-04-13 |
| Manual QA export (Admins export "gold standard" chats and re-upload as PDF/Excel) | High friction — defeats the automation value prop for low-maintenance Qontak360 clients | 2026-04-13 |
| Division-wide auto-inclusion (all agents in a division, no individual selection) | Lack of control — can't prioritize senior experts over new joiners | 2026-04-13 |
6. Open Questions
| # | Type | Question | Owner | Deadline |
|---|---|---|---|---|
| 1 | Risk | PII masking must redact all sensitive entities before indexing. Mitigation: mandatory NLP masking engine replacing entities with generic tags ([EMAIL], [PHONE_NUMBER]); QA gate on standard formats before GA. | Data & AI team | 2026-07-15 |
| 2 | Open Question | Jira epic + delivery timeline not yet created (stories tracked as TEMP placeholders). | Dimas (PM) | 2026-07-01 |
| 3 | Open Question | Copilot quota model for this source (e.g., 15 quota/seat Service Suite) — confirm metering. | Dimas (PM) | 2026-07-15 |
PRD CHANGELOG
| Version | Date | By | Section | Type | Summary |
|---|---|---|---|---|---|
| 1.0 | 2026-06-18 | Claude | All | CREATED | ANCHOR created from the "AI Agent Knowledge: Conversation History" Confluence draft. Conversation History set as Phase 1; static PDF/URL sources noted as pre-existing; data-driven selection and other Future Considerations seeded as future phases. |