Skip to main content

Qontak | AI Agent | Meta Business Agent Integration — ANCHOR

ANCHOR PRD — the initiative master index. It orchestrates all phases beneath it and carries no single-delivery execution detail (Non-Goals, Constraints, Observability, and Dependencies live in the Phase PRDs).

HEADER BLOCK

FieldValue
PMDimas Fauzi Hidayat
PRD Version1.6
StatusDRAFT
PRD TypeANCHOR
AnchorYes — this IS the Anchor
Labelsepic:qontak-chatbot | module:ai-agent | feature:meta-business-agent
Last Updated2026-07-21

Table of Contents


1. PHASE INDEX

This table is the source of truth for all phases. Goal column must exactly match each child Phase PRD's "Phase Goal" field.

PhaseGoalPRD LinkEpic KeyStatusShipped
Phase 1: MBA-Aware FoundationMake the Qontak inbox fully aware of MBA-run conversations — persist standby copies, track who controls each room, and let agents take over and hand back.phase-1-mba-aware-foundation.mdTBD📝 Draft
Phase 2: MBA Management & Config SyncLet org admins onboard, configure, and operate MBA from Qontak — eligibility → onboarding → allowlist → settings on the channel page, with admin-selected knowledge/persona synced from the existing AI agent on publish.phase-2-mba-management-config-sync.mdTBD📝 Draft (gated: MBA GA + pricing in Indonesia)
Phase 3: Connector Gateway + Test/EvalLet MBA execute Qontak-hosted actions through an authenticated connector gateway, and run existing AI agent test cases against Meta's agent test/eval APIs.phase-3-connector-gateway-test-eval.mdTBD📝 Draft (sequenced after Phase 2)

2. One-liner + Problem

One-liner: Enable Qontak organizations to run, observe, and control Meta Business Agent — Meta's hosted AI responder on WhatsApp — entirely from Qontak.

Problem: Meta Business Agent (docs updated 2 Jul 2026) becomes the primary responder on a WhatsApp number when enabled, and the business app is demoted to a standby observer — but Qontak today does not process the standby or messaging_handovers webhook fields, has no Thread Control client, and exposes no MBA management surface. Any Qontak customer with an eligible WABA can enable MBA directly in WhatsApp Manager, at which point their inbox goes partially blind (agent replies invisible, control state unknown) and a human who takes over can never hand the conversation back. Left unaddressed, this displaces Qontak's core chatbot value proposition on its highest-volume channel and disintermediates Qontak into a passive inbox — while from 1 Oct 2026 Meta also meters every service message, changing the economics of the channel itself.


3. What Happens If We Don't Build This

  • Blind inbox on first customer adoption — the moment one Indonesian customer enables MBA in WhatsApp Manager (Meta pilot posture suggests months, not years), their Qontak transcript silently loses MBA's replies and control state; support tickets follow.
  • Disintermediation of the AI layer — customers who adopt MBA outside Qontak rationalize "Meta's agent, Meta's channel, one bill"; Qontak loses the chatbot seat and the data, keeping only human-handling scraps.
  • No honest competitive data — without standby persistence Qontak cannot measure MBA containment vs its own AI agent side-by-side, so pricing/positioning decisions against MBA stay guesses.

4. Target Users + Persona Context

Primary Persona: CS Supervisor / Org Admin (WhatsApp-first support org)

FieldDetail
RoleSupervisor or org admin running WhatsApp customer service on Qontak, evaluating or enabling Meta Business Agent on their number
GoalAdopt Meta's AI responder without losing the Qontak inbox, transcript history, human escalation flow, and reporting they run the operation on
PainEnabling MBA today makes Qontak partially blind; there is no way to configure MBA from Qontak or return a conversation to it after a human steps in
WorkaroundNone — either don't enable MBA, or enable it in WhatsApp Manager and accept a broken inbox

Secondary Persona: CS Agent (inbox user)

FieldDetail
RoleAgent handling WhatsApp conversations in the Qontak inbox
GoalSee the full conversation (including MBA's replies), know who is driving, and take over cleanly when a customer needs a human
PainMBA-run conversations would appear with missing messages and no control indicator; after replying, the agent owns the room forever
WorkaroundNone — asks the customer to repeat context; room never returns to the AI

5. Success Metrics (Initiative-level)

North-star metrics for the full initiative — per-phase metrics live in each Phase PRD.

Adoption & Usage:

MetricDefinitionBaselineTarget
⭐ MBA-through-Qontak attach rate% of MBA-enabled customer WABA numbers where MBA is onboarded/managed via Qontak (vs. directly in WhatsApp Manager)N/A — MBA not yet GA≥70% of MBA-enabled Qontak customer numbers within 2 quarters of Phase 2 GA
MBA-aware orgs# of organizations with mba_aware_inbox active and ≥1 MBA conversation observedN/A — new capability100% of orgs with MBA-enabled numbers (Phase 1 is the safety net, not opt-in value)

Quality & Accuracy:

MetricDefinitionBaselineTarget
Transcript completeness on MBA roomsMBA-room messages persisted in Qontak ÷ messages delivered on standby/messages webhooksN/A — standby not processed today (0%)≥99% within 30 days of Phase 1 GA
Control-state accuracyRooms whose displayed controller matches the last messaging_handovers eventN/A100% (measured in beta and continuously)

Efficiency & Impact:

MetricDefinitionBaselineTarget
Side-by-side containment comparabilityAbility to report MBA containment vs Qontak AI containment per org (rooms resolved without human takeover)Not measurable todayReportable for 100% of MBA rooms from Phase 1 data (reporting itself ships via the AI ROI report initiative)
Time-to-take-overMedian time from customer message needing a human to agent's first reply on MBA roomsN/A≤ existing human-first-response median on non-MBA WhatsApp rooms

6. Key Decisions + Alternatives Rejected

6a — Decisions Made

DateDecisionRationale
2026-07-07Adopt MBA defensively inside Qontak (Track 1 of the three-track strategic response) rather than ignore or block itIf customers will get MBA anyway via WhatsApp Manager, running it through Qontak keeps the seat, the data, and the actions; code grounding shows ~90% of the config surface already exists
2026-07-07Phase 1 (MBA-aware foundation) ships ungated; Phases 2–3 are gated on MBA GA + pricing in IndonesiaThe blind-inbox risk exists the moment any customer enables MBA — awareness cannot wait for our management build; the management build shouldn't precede a sellable product
2026-07-07The existing AI agent (Profile / Capabilities / Routing) is the single source of truth for MBA config; sync fires on the existing publish eventNo second agent builder; mirrors how SyncToAiService already pushes config to Qontak's AI Service on publish — grounded against chatbot repo
2026-07-07All MBA UI lands on existing pages (channel Account details section, Resources/Actions columns, inbox elements) — no new pages or routesIA-alignment rule: existing pages already carry per-number Meta capability sections (call/catalog/commerce anatomy); wireframed field-accurate 2026-07-07
2026-07-07MBA connector tools will execute through a Qontak-hosted gateway (Phase 3), never customer-hosted endpointsKeeps credentials and action execution in Qontak ("we keep the hands") — the strategic lock-in that survives even if Meta owns the conversation
2026-07-18MBA and the Qontak AI agent live side-by-side at org level off one shared linked config, with scoped divergence on existing surfaces: per-source Meta-sync toggle (P2 S06), per-action expose toggle (P3 S01), per-field Profile Meta overrides (P2 S08), MBA-only operational settings on the channel section (P2 S07)One authoring surface + opt-out/override beats a forked agent entity — no drift, no double authoring, no new pages; the one-AI-per-number guard is untouched (coexistence is across numbers, never two responders on one)
2026-07-19Zero Mekari AI Services (DSAI) dependency across all MBA phases — the integration is wholly Qontak-owned: chatbot/hub-core talk to Meta directly, Meta does its own ingestion/crawling/embedding; AI Services remains only in the existing Qontak-agent pathPM requirement per the standing cross-team-dependency rule; technically free because Meta consumes raw bytes/URLs, not embeddings — code-verified that raw knowledge files already live in Qontak's OSS
2026-07-18PRDs re-grounded against Meta's published MBA Platform API docs v2.0.0 (API surface & build maps: §7 of each Phase PRD)Corrects Thread Control to release (P1 v1.4), adds settings/audience/follow-up + Business info + skills-collection reality (P2 v1.4), resolves the connector auth mechanism + adds tool smoke-test and connector-logs endpoints (P3 v1.4)
2026-07-21Omnichannel-AI direction (vision — feeds the Automation Flow Builder initiative, does not change Phase 1–3 scope): (a) the AI provider is chosen per channel/number on the existing AI agents page — one "Runs on" column + one provider field in the existing deploy-to-channel drawer, no new page; (b) MBA additionally becomes a hand-off node in the generalized flow: a capability-gated terminal node (executes Thread Control release) with a fallback output port for channels/numbers where MBA is unavailable, guarded by NodeRegistry supported_channels/requires metadata + a save-time lint. No MBA-specific flow type — one multi-channel flow serves all providers. Coexistence accordingly evolves from per-number mutual exclusion (OQ-4) to composition: per number, either MBA-primary (flow dormant; Qontak observes on standby) or flow-primary (flow orchestrates; MBA reachable as a hand-off node)Delegated providers can only be orchestrated at hand-off boundaries (release / take-over / observe — never per-turn; after release the flow receives only standby copies until a take-over), so the node is terminal by protocol. A dedicated MBA flow would fork the one-flow-many-channels model the flow builder exists to unify. The node lands with the Automation Flow Builder engine; the legacy tree gets it only if needed earlier

6b — Alternatives Rejected

AlternativeWhy RejectedDate
Do nothing / advise customers not to enable MBAUnenforceable — MBA is a tab in WhatsApp Manager on the customer's own WABA; Qontak would still go blind on standby traffic2026-07-07
Build a separate MBA console (new pages/module)Inflates scope and drifts from the IA; the channel detail page already stacks per-number Meta capability sections that MBA fits exactly2026-07-07
Duplicate persona/handoff config in the MBA sectionCreates two sources of truth; persona, tone, language, and handoff already live in the AI agent builder and map 1:1 to MBA Skills/settings2026-07-07
Wait for MBA GA before building anything (including awareness)Leaves the blind-inbox window open for the entire pilot period; Phase 1 components (webhook fields, thread control) are small and reusable regardless2026-07-07
Full build now (all 3 phases before GA)Meta's allowlist/pilot posture and unknown pricing make the management surface premature; risk of building against a moving API2026-07-07

7. Open Questions

#TypeQuestionOwnerDeadline
1Open QuestionWhen does MBA reach GA in Indonesia, and at what price? (Gates Phase 2 commit)Dimas F — via Meta partner manager2026-08-31
2Open QuestionBSP / Tech Provider terms for MBA: does managing MBA via BISU tokens carry revenue-share or ToS implications for Qontak — and how does Qontak obtain the bizai_wa_enterprise_api_3p_access tech-provider capability the MBA Platform APIs accept as the alternative to per-customer permissions (docs v2.0.0)?Dimas F + Partnership2026-08-31
3Open QuestionPackaging: is MBA management a paid add-on (WhatsApp-commerce upsell pattern) or included (WhatsApp-call pattern)?Dimas F + Pricing2026-09-15
4AssumptionOne AI responder per number: enabling MBA requires disabling the Qontak bot flow / AI agent on that number, and migration UX is a Phase 2 concern — long-term direction: this guard evolves to composition via the Automation Flow Builder hand-off node (§6a, 2026-07-21); until that engine lands, this assumption governs Phase 2Dimas F2026-08-15
5RiskMeta bundles MBA free/cheap with WABA, undercutting Qontak's AI pricing — Mitigation: Track 3 system-of-record principle (transcript, connector gateway, scorecard stay in Qontak regardless of who answers) + Track 2 offensive plays ship firstDimas F2026-10-01
6RiskBuilding Phase 2 against a pilot-stage API that changes at GA — Mitigation: Phase gating (Open Question 1) + Phase 1 limited to webhook/thread-control surfaces already shared with the stable Handover ProtocolDimas F2026-08-31

Risk-type rows carry mitigation paths — required before this ANCHOR can move to ACTIVE.


PRD CHANGELOG

VersionDateBySectionTypeSummary
1.02026-07-08ClaudeAllCREATEDInitial ANCHOR PRD created from the MBA strategic-response grounding (code + design/IA grounding of 2026-07-07).
1.12026-07-08ClaudePhase IndexMODIFIEDPhase 2 and Phase 3 PRDs drafted (ahead of their gates, per PM) — links added, statuses moved to Draft.
1.22026-07-18ClaudePhase IndexMODIFIEDPhase 2 goal now reads "admin-selected knowledge/persona" — per-item MBA-vs-Qontak scope choice added across Phase 2 (per-source Meta sync toggle, MBA2-S06) and Phase 3 (expose-toggle scope semantics, MBA3-S01/AC-3); goal wording kept verbatim-matched with the Phase 2 PRD (v1.3).
1.32026-07-18Claude6, 7MODIFIEDAPI-grounding sweep vs Meta MBA Platform docs v2.0.0 (API surface & build maps in each Phase PRD §7): side-by-side + scoped-divergence model recorded as a decision; child PRDs bumped (P1 v1.4 release fix, P2 v1.4 S07–S09, P3 v1.4 auth/smoke-test/logs); OQ-2 extended with the bizai_wa_enterprise_api_3p_access tech-provider capability.
1.42026-07-18Claude6MODIFIEDStandalone documentations/mba-api-development-inventory.md removed per PM instruction — each Phase PRD's §7 now carries its own "Meta API surface & build map" slice (P1/P2/P3 → v1.5); decision + changelog references updated accordingly.
1.52026-07-19Claude6MODIFIEDZero-Mekari-AI-Services decision recorded (PM requirement): MBA path is wholly Qontak-owned, AI Services stays Qontak-agent-only (P2 v1.6 §5 constraint + P3 v1.6 §5 mirror).
1.62026-07-21Claude6, 7MODIFIEDOmnichannel-AI direction recorded (vision): provider choice on the existing AI agents page + MBA as a capability-gated hand-off node (fallback port) in the Automation Flow Builder; coexistence evolves from per-number mutual exclusion to composition (OQ-4 pointer added). Lifecycle wireframe set (Onboard/Configure/Operate + HTML index) added under wireframes/. Phase 1–3 scope unchanged.