Building a Multi‑Dimensional Analysis System at Baixin Bank: Practices and Insights

Speaker Luo Hui, BI leader of Baixin Bank’s Data Department, introduces the bank’s fully online business model, its reliance on intelligent technology, and the need for comprehensive data to support decision‑making.

The bank’s data analysis demands are summarized as three principles: accuracy (consistent metric definitions and reliable cross‑validation), completeness (covering core scenarios across the entire value chain), and ease (tailored views for strategic, tactical, and operational users).

To meet these needs, a ladder‑style service design is proposed: first ensuring metric accuracy through cross‑team collaboration; then achieving completeness by building both business and portrait metric families; finally delivering ease via unified dashboards, automated alerts, and self‑service analysis tools.

The indicator system is illustrated with a credit‑analysis example, showing how high‑level business goals are decomposed into measurable metrics, dimensions, and hierarchical groupings (scale, count, ratio) that support both static and dynamic user portraits.

User portraits are divided into static attributes (e.g., demographics) and dynamic behavior profiles, while product portraits capture interest rates, terms, risk, and usage patterns; both are linked through an AARRR lifecycle model (Acquisition, Activation, Retention, Referral, Revenue).

The analysis system progresses through descriptive, diagnostic, predictive, and prescriptive stages, enabling automated monitoring, root‑cause attribution, and scenario‑driven decision support.

Data product construction follows a three‑layer cube model (enterprise‑product‑user) and a four‑block analytical view (overview, change, composition, comparison), facilitating drill‑down from high‑level dashboards to detailed user‑product interactions.

Self‑service capabilities are realized through drag‑and‑drop visualizations, pre‑defined metrics and dimensions, and a governance layer that enforces data quality rules (non‑null, logical consistency, uniqueness) across extraction, modeling, and application stages.

The article concludes with reflections on the importance of combining business understanding with technical skills, staying aware of external factors, delivering practical solutions quickly, and promoting data‑driven value across the organization.