2025 Jan - 2025 Jul · Industry
Marketing Research Agentic AI
Agentic AI pipeline spanning product, market, audience, and competitive intelligence—with structured outputs for strategy and communications.
An agentic AI system concept that automates end-to-end marketing research: from product and industry scans to audience archetypes, insights, brief-ready communication tasks, and competitor synthesis.
Career chronology (About) This project was developed while I worked at RED Comm Indonesia (2024 — 2025). Portfolio dates: 2025 Jan - 2025 Jul.
Tech Stack
AI/LLM
- CrewAI
- OpenAI API
- Claude API
- Google Gemini API
Data
- MySQL
- Weaviate
Backend
- Python
Problem
Marketing research is fragmented across spreadsheets, slide decks, and ad hoc searches. Teams lose time reconciling sources, and downstream creative and media briefs rarely inherit traceable evidence from upstream analysis.
Approach
Modeled the workflow as cooperating agents with shared memory and citation-backed outputs: retrieval and normalization of sources, structured extraction per research dimension, cross-checks between agents, and a final synthesis pass that produces both strategic narrative and brief-ready artifacts.
Marketing analysis coverage
The pipeline follows familiar marketing research steps so outputs line up with planning, strategy, and creative—summarized here at a high level for confidentiality.
Product Research
Pull positioning, features, price signals, and claims into one place so the story is consistent. Agents highlight where decks, sites, or sheets disagree so marketers can reconcile before the work goes downstream.
Market Research: Industry
Size the opportunity: growth, regulation, seasonality, and how the category really moves through channels. Confidence levels and dates stay attached so every headline in a plan or presentation can be defended.
Target Audience & Audience Archetype
Start from who buyers are, what they do, and the jobs the product is hired for. Agents group evidence into segment hypotheses, flag gaps, then shape archetypes—motivations, friction, trusted touchpoints—grounded in that research instead of generic persona slides.
Key Insight
Distill the “so what” when product, audience, and competition are read together—e.g. whitespace where demand is strong but competitors are thin—each point traceable to the sources behind it.
Challenge & communication
Once product, audience, and competition line up, output a brief-ready line on the strategic challenge, the behavior you want to shift, the proof to use, and the guardrails (tone, channels, compliance) creative and media need.
Competitor Analysis
Compare how rivals position, message, price, and show capability side by side with your product read. Map overlap, gaps, and whitespace to sharpen differentiation and counter-messaging.
System Architecture
- Ingestion agents gather product docs, public web, syndicated data, and internal notes into a normalized evidence store.
- Specialist agents fill the standard marketing workstreams—product, industry, audience, competition—with fields and citations a planner would expect.
- Synthesis agents turn cross-checks into insights and draft the communication challenge for brand or media partners.
- A reporting agent packages views for stakeholders—from tables to narrative—without losing the link back to sources.
Methodology
- Orchestration: clear handoffs between steps (e.g. industry context informs audience framing; competitor claims checked against the product fact base).
- Quality: structured outputs per workstream, source lists by section, and human review before anything goes external.
- Delivery: parallel work where topics are independent; sequenced passes where creative or strategy needs locked inputs first.
Outcome
Defined a repeatable research stack that accelerates time-to-insight while keeping claims grounded in attributable sources, ready for human review before campaign or product decisions.
Current Status
- Deployed internally: the application is in active use as a supplementary project alongside core work to support the company's business.
- Development continues under a new team after my departure; I resigned to pursue a Master's degree (ongoing).
- Portfolio descriptions remain high-level for confidentiality.
Key Contributions
- Mirrors how marketing teams actually work—product, market, audience, competition, insight, brief—so agent outputs map to familiar deliverables.
- Keeps claims source-backed and structured so plans and decks hold up in review.
- Closes the loop with a communication challenge and brief inputs, not a pile of disconnected notes.