Data‑Driven Precise Marketing: Architecture and Case Studies at Meituan‑Dianping

Precise marketing uses data‑driven methods to quickly acquire users and boost conversion in fragmented markets. With mobile internet booming, data volume grows exponentially, making precise marketing in mobile and big‑data scenarios a major challenge and research focus.

The article shares Meituan‑Dianping’s data application team’s architecture and practice, concentrating on O2O marketing, especially in‑site user‑targeted campaigns.

An in‑site activity follows six steps: define goals, select audience, design plan, configure & launch, conduct real‑time optimization, and monitor & evaluate effects.

To solve manual target selection and uncontrolled budgets, a layered pyramid architecture was built: a data warehouse/model layer (profiles, operations, marketing, traffic), a data‑service layer, and front‑end products.

Profile data includes >180 tags across five categories (basic info, device, consumption, browsing, demographic). The team evolved from statistical models (RFM) to machine‑learning techniques for user discovery and preference analysis.

Redis powers real‑time profile services, while Elasticsearch handles analytical workloads, chosen over Kylin and Druid for incremental cube support and lower storage overhead.

Three tools are offered: Hoek for audience analysis, Cord for intelligent coupon issuance, and Cloud/Star maps for multi‑dimensional real‑time KPI queries.

Case studies: (1) potential user discovery for food‑delivery using clustering, classification, and association‑rule algorithms; (2) a WeChat red‑packet coupon engine that treats coupon selection as a recommendation problem, featuring modular flow, recall, filtering, and ranking components.

The team distills design principles: build accurate, reusable data models from requirements; adopt layered SOA to decouple services and select suitable technologies per scenario. Future work focuses on faster model development and expanding data‑driven marketing to new scenarios.