How Kuaishou E‑commerce Leverages OLAP and a Unified Data Architecture to Solve Business Data Challenges

Introduction In the data‑driven era, Kuaishou e‑commerce faces unprecedented data challenges across user behavior, product management, market trends, and content delivery, requiring innovative OLAP techniques and metric management to support real‑time and offline analysis.

Business Overview The e‑commerce model consists of two supply sides—large brands and self‑owned merchants—distributed across various sales channels such as live‑streaming, short videos, shelf pages, promotional events, and external traffic sources, with a recommendation system driving user conversion.

Challenges Rapid business growth led to exploding data requests, inconsistent dashboards, duplicated metric definitions, and the need to rebuild data pipelines for real‑time scenarios, highlighting gaps in data accuracy, metric consistency, and a unified dimensional matrix.

Solution Overview The team established a three‑layer data warehouse (DIM, DWD, DWS) to provide consistent dimensions and facts, standardized metric definitions, and a unified OLAP service. They performed thorough business and requirement analysis, designed models with clear naming and rollback capabilities, and built both cold (Hive) and hot (ClickHouse) storage for scalability.

Data Architecture The architecture includes a data ingestion layer, consistent fact and dimension layers, and an upper layer of scenario‑specific assets, all managed through a central metric platform that feeds dashboards, self‑service queries, and AB‑testing tools.

Construction Results The platform enabled multi‑dimensional analysis, self‑service querying, and real‑time battle screens for live hosts, achieving a growth from ~1,000 to over 750,000 queries per month, expanding user coverage from 20 to over 900, and significantly improving operational efficiency.

Future Plans The roadmap envisions AI assistants powered by large language models that can answer natural‑language queries using the standardized metric metadata, further integrating AI with the OLAP system for seamless data interaction.

Q&A Highlights Topics covered include metric standardization across teams, the necessity of a unified metric layer for large models, the rationale behind cold storage in Hive and hot storage in ClickHouse, and performance challenges with joins in ClickHouse.