JD Tech Data Warehouse Journey, Model Management, and Tag Value Evaluation

JD Tech began accumulating data in 2013, building data warehouses for various business domains. By 2016, inconsistencies in metric definitions prompted a focus on data governance, initially addressing basic data standards. In 2018, the shift moved toward proactive data asset management, including inventory, cataloging, and value assessment.

From 2021, JD Tech entered a phase of quantitative data asset management, using indicators to measure management quality and drive improvements. By 2023, continuous optimization was pursued based on quantified metrics and operational mechanisms.

The core of data warehouse construction is layered architecture. Inspired by Bill Inmon's information factory, JD Tech adopts a multi‑layer model (IDM, SDM, SDP, DIM, DEV, TMP, STG, ODM) to ensure clear data structures, lineage tracking, reduced duplication, and isolation of upstream changes.

Model design combines normative standards, data lineage mapping, and a mixed modeling strategy: relational (normative) modeling for IDM and DIM layers, dimensional modeling for SDM, and wide‑table models for the market layer. Early data profiling identifies source issues such as missing values, outliers, and duplicates, guiding appropriate model design.

Key model management practices include establishing data and design standards, documenting field‑level lineage mappings, and conducting model reviews with a scoring card that evaluates compliance, cost, reusability, and data‑ops considerations.

During model usage, JD Tech defines a comprehensive metadata schema (24 attributes) covering production, real‑time, and model metadata, with automated quality checks and monitoring dashboards to ensure data quality and governance.

Model governance enforces single ownership, asset handover procedures, health assessments (storage strategy, metadata quality, security), and user feedback mechanisms to maintain model reliability and usability.

Tag development aligns with the warehouse layers, transforming detailed business data into user‑level tags and supporting downstream applications such as customer data platforms and profiling.

Tag value is evaluated using a weighted formula: BVI = 5% × Effectiveness + 5% × Stability + 5% × Frequency + 22% × Coverage + 15% × Scarcity + 15% × Breadth + 8% × Depth + 25% × Economic Impact. Factors include tag quality, update frequency, coverage, rarity, heat, and economic relevance.

Results from the evaluation guide targeted improvements—e.g., enhancing coverage or stability—and have helped retain high‑value tags, increasing the proportion of valuable assets in JD Tech's data asset management platform.

Looking forward, JD Tech plans to relax strict data flow controls, adopt data virtualization for cross‑layer reuse, and integrate financial benefit assessments for tags, while continuing to evolve data warehouse architecture and model management practices.