■ Project Overview
AI and data integration platform to enhance sales and customer experience
7 members × 5 months
PM / BA / Developers / QC
Requirements definition / UI-UX design / Development / Testing / Deployment / Maintenance & Operations
■ Technology Stack
AI / Machine Learning: LightFM / LightGBM / Prophet / GPT-4o
AI Platform: LangChain / Qdrant
Data Platform: dbt / BigQuery
Backend: Python (FastAPI) / Redis / PostgreSQL
External Integration: LINE Messaging API
Frontend: React 18 / TypeScript
Cloud / MLOps: GCP (GKE / Cloud Run / Vertex AI) / Terraform / GitHub Actions
■ System Features
Data integration platform centered on a CDP combined with AI-driven architecture
Hybrid structure combining real-time and batch processing
Continuous model improvement through MLOps
Integration of LINE data
Centralized management of 1.2 million customers
Customer unification through identity resolution
High-accuracy forecasting using LightGBM + Prophet
Predictions by SKU × store × weekly basis
Supports Japan-specific seasonality (holidays, fashion trends)
Weekly model retraining
RAG architecture (Qdrant + GPT-4o)
Supports product search / order inquiries / reminders
High-quality responses in Japanese (including polite expressions)
Natural language-based segment generation
AI-driven content generation
A/B testing and real-time performance measurement
■ Project Outcomes
CTR +47% / CVR +31% improvement
Integrated 1.2 million customer records within 2 weeks
Demand forecasting accuracy: MAPE 11.3% (34% improvement)
Reduced product search time by 40%
Improved bounce rate by 22%
Chatbot satisfaction score: 4.7 / 5
Supported store inventory optimization
Reduced excess inventory by 28%
Marketing setup time reduced from 2 days to 3 hours
Reduced planning workload (3 days per person per season)
E-commerce sales +18% (exceeded target)
Sales via LINE +35%
ROI achieved within 8 months
Annual cost reduction: approx. ¥120 million
Recognized as the most successful project among DX initiatives
Highly evaluated for addressing Japan-specific requirements
Decision made to increase AI investment by 40% for the following year
