Search the portfolio

↑ ↓ to select · Enter to open · Esc to close

Back to the work

AI GROWTH.

Prototype / Integration

An agentic growth and revenue system.

AI AGENTSAUTOMATIONGROWTH
01 / CONTEXT

The qualification gap.

Social engagement and inbound questions often contain sales intent, but manual qualification is inconsistent and difficult to scale.

02 / SYSTEM

The operational chain.

Facebook engagement → Wira FB → sales-intent detection → Pemburu → qualification → scoring → Duta handoff

03 / AGENT ROLES

Specialized responsibilities.

WIRA FB

Facebook organic/community layer.

Detects relevant commercial intent from engagement.

PEMBURU

AI SDR and lead-intelligence layer.

Extracts structured context, qualifies, scores and recommends next action.

DUTA

Human-facing follow-up.

Receives qualified handoffs for relationship management.

Note: Agent roles are distinct from the underlying automation infrastructure.

04 / QUALIFICATION MODEL

Objective scoring dimensions.

HOT / WARM / COLD / SPAM

The system evaluates intent, fit, urgency, and authority signals. It uses BANT-style qualification where evidence exists in the conversation.

Unknown information remains UNKNOWN. The system must not invent budget or authority if it wasn't provided.

05 / EVIDENCE

Verified prototype behavior.

TEST POOL

15 synthetic leads evaluated.

Validates HOT / WARM / COLD / SPAM classification.

STRUCTURED EXTRACTION

Consistent schema generation.

The system reliably maps conversations to a structured lead schema, executing BANT-style qualification and product/service matching.

HANDOFF TICKETS

Deterministic completion.

Produces qualified-lead handoff tickets, demonstrating zero-silent-drop recovery behavior via DLQ.

ARCHITECTURE

Centralized inference gateway.

Validates the core routing layer that manages AI model execution.

RELIABILITY

Engineered for failure recovery.

AI INFERENCE GATEWAY

Centralized model routing abstracts the LLM provider from the core logic.

TELEMETRY

Latency, model choices, and usage tracking are built into the pipeline.

FALLBACK

A deterministic local engine steps in if external AI is unavailable.

DLQ & SELF-HEALING

Failed lead events are preserved for recovery in the DLQ. Automated retry, incident creation, and verification ensure resilience.

CURRENT LIMITATION

Honest prototype status.

The Meta Page access token had expired during live integration testing.

The latest prototype cannot yet claim a continuously verified end-to-end production flow.

Next validation target: real Facebook comment → detection → qualification → scoring → Duta handoff → audit record.

FUTURE KPIs

Measurable targets.

  1. 01Intent-to-qualified rate
  2. 02Qualified-to-handoff rate
  3. 03Response latency
  4. 04False-positive rate
  5. 05Dropped-lead rate
MYT / UTC+8