LG H&H Marketing Innovation PoC
Published 2026-05-14Updated 2026-07-026 min read
"A marketing innovation PoC that ties together first-party member, SKU, campaign, and omnichannel transaction data with external trend, weather, economic, and competitor signals via ontology + Agentic AI."
A working demonstration that semantically connects scattered data so that 5 departments can use it from their own perspectives. Using only public information, it presents a blueprint for solving the data silo problem at LG H&H's Marketing Innovation Division.
Interactive Demo
👉 Open Demo Page (new window)
Switch between the 8 scenarios in the sidebar to directly experience core UI patterns such as the persona switcher, charts, KG graphs, SSE chatbot traces, and Swimlane timelines.
1. PoC One-Line Summary
| Axis | Content |
|---|---|
| For whom | LG H&H Marketing Innovation Division (Brand Marketing, Insights, D&A, CRM, MD — 5 departments) |
| What | 3 BU (Beauty + HDB + Refreshment) first-party data + 4 types of external signals integrated |
| How | Ontology (Neptune KG, 25 classes) + Agentic AI (Bedrock Sonnet + AgentCore) |
| How is it shown | 8-scenario working demonstration + 5-department persona switcher |
| Differentiator | Not a concept but a working demonstration (real data N=500~5K + synthetic 50K + live external) |
2. 3 BU Integration (based on public information)
| BU | Key Brands | Core Channels |
|---|---|---|
| Beauty | The History of Whoo / Su:m37 / OHUI / belif / CNP / VDL / THE FACE SHOP | Duty-free · Department store · Owned mall · Olive Young · Coupang |
| HDB | Elastine / Perioe / Tech Plus / Bamboo Salt / Dr.Groot / Saffron / Safran | Hypermarkets · Owned mall · Coupang · Convenience stores |
| Refreshment | Coca-Cola / Fanta / Sprite / Minute Maid / Toreta / Powerade / Seagram / Galbae Cider | Convenience stores · Hypermarkets · QSR · Vending machines |
3-BU integration is the core value: A single LG Members member is observed via one KG to see how they move across BUs and channels.
3. 5-Department Personas (Switcher)
| Code | Department | KPI |
|---|---|---|
| P1 | Brand Marketer | Campaign ROAS · New product adoption rate |
| P2 | Consumer & Trend Insights | LTV · Persona share · SOV |
| P3 | D&A · MarTech | Model accuracy · Drift · Pipeline SLA |
| P4 | CRM · LG Members | Tier conversion · Active members · NPS |
| P5 | MD · Channel Sales | Channel GMV · Inventory turn · New product adoption |
Same data, different perspectives. The sidebar ordering, card emphasis, and chatbot tone all change.
4. 8 Scenarios (One-Line Summary)
| # | Scenario | Data Mix (first-party + external) |
|---|---|---|
| S1 | Natural-language semantic search | SKU · Member · First-party reviews + Olive Young · X reviews |
| S2 | 5-department persona chatbot | Autonomous invocation of all tools |
| S3 | Category · BU insight cards | First-party GMV + search trends + weather + competitors |
| S4 | Lifestyle persona matching + clustering | Member RFM + category affinity + social personas |
| S5 | Omnichannel campaign ROAS simulation | Campaign · sell-through + search · SNS responses |
| S6 | External signal fusion (trend · weather · economy · competitor) | First-party GMV × all 4 external types |
| S7 | Omnichannel member journey | Owned mall → SNS ads → Olive Young → Mart → Repurchase |
| S8 | Marketing consent · PII · minor guardrails | Bedrock Guardrails + cosmetics guard |
Detailed mapping: 03. Scenarios
5. 4 External Data Types (based on public information)
| Type | Example Sources |
|---|---|
| Social trends | Naver · Google Trends / X / Instagram / Shorts · Tistory / Olive Young reviews |
| Weather · Environment | KMA (Korea Meteorological Administration) / Air quality index |
| Economy · Consumption | KOSIS · Bank of Korea exchange rates · CPI · employment |
| Competitor signals | Public appearances of Amorepacific · Aimo · Yuhan-Kimberly new product trends |
6. AWS Stack (Single Plan)
| Layer | Component |
|---|---|
| Edge / Auth | CloudFront + Lambda@Edge + Cognito (RS256 JWT) |
| Frontend | Next.js 14 (App Router) |
| Backend | FastAPI (Python 3.12) on ECS Fargate ARM64 |
| AI | Bedrock Sonnet 4.6 + Cohere embed-v4 / rerank-v3.5 |
| Agent | AgentCore (Memory + Code Interpreter) |
| Graph | Amazon Neptune (openCypher, ~500K edges) |
| Search | OpenSearch Serverless (Nori + KNN + RRF) |
| Data | S3 + DynamoDB (sessions) + Glue (external data ETL) |
| Governance | Bedrock Guardrails (4 topics + consent + minor) |
Details: 05. Architecture
7. Top-5 Demo Flow (30 minutes)
| Order | Scenario | Demo Highlight | Time |
|---|---|---|---|
| 1 | S1 Semantic Search | Korean natural language → SKU + Member + external reviews simultaneously | 4 min |
| 2 | S2 Persona Chatbot | Compare answers across 5 departments via switcher | 5 min |
| 3 | S3 Insight Cards | Auto-report combining 4 external types | 4 min |
| 4 | S5 Campaign ROAS | Bayesian attribution + channel mix | 4 min |
| 5 | S7 Member Journey | Omnichannel single timeline | 3 min |
Details: 99. Demo Storytelling
8. Deliverables Index
- 01. Personas — 5-department persona definitions
- 02. Ontology — KG 25 classes + Mermaid
- 03. Scenarios — 8-scenario mapping + prioritization
- 04. Data Sources — First-party + 4 external types specification
- 05. Architecture — AWS single architecture
- 06. Design Specs — Design specs for 8 scenarios
- 99. Demo Storytelling — 30-minute demo script