Real-time Data Warehouse Empowering Fine-grained Intelligent Operations in Finance – A Practical Case Study

Shi Xingtian, Senior Director of Data at Zhongan Insurance, introduced the session titled “Real-time Data Warehouse Empowering Fine-grained Intelligent Operations in Finance,” outlining five key topics: an overview of Zhongan Insurance, the company’s data middle‑platform, the JiZhi platform and case studies, the construction of a real-time data warehouse with StarRocks, and technology‑driven solutions.

Zhongan Insurance Overview – In 2021 the insurer generated RMB 200 billion in premiums, ranking among the top ten property‑insurance firms in China. Its product portfolio includes high‑coverage health plans, return‑shipping insurance, flight‑delay coverage, and credit‑guarantee insurance. Approximately 50% of its staff are engineers, creating substantial data volumes and processing demands.

Digital Framework 4633 – Zhongan’s “4633” framework consists of four layers: Application, Algorithm, Platform, and Assurance. The Application layer focuses on business value (increase, reduce, optimize, expand). The Algorithm layer provides six capabilities: recognition, prediction, clustering, optimization, cognition, and risk control. The Platform layer integrates data processing, business intelligence, and machine learning, while the Assurance layer emphasizes talent development and management‑driven data teams.

Data Middle‑Platform – Built on the 4633 architecture, the platform combines a core JiZhi platform (handling data processing, machine learning, and governance) with front‑end systems for public‑domain acquisition and private‑domain operations, supporting diverse user groups from data engineers to product managers.

JiZhi Platform & Use Cases – The platform enables multi‑scenario fine‑grained analysis, integrating external data (e.g., Douyin, Kuaishou) with internal business, operational, and financial data. It supports real‑time visualization, attribution analysis, and predictive modeling to answer “what happened,” “why it happened,” and “what will happen.”

Real-time Data Warehouse Evolution – Initially, the JiZhi platform used ClickHouse for OLAP queries, but growing concurrency and multi‑table joins caused severe performance degradation and operational challenges (ZooKeeper dependency, lack of automatic re‑sharding, limited DML support). To address these issues, Zhongan migrated to StarRocks, which excels in multi‑table joins, concurrent queries, and simplifies operational tasks.

Performance Comparison – Benchmarks showed StarRocks matching ClickHouse on single‑table, single‑thread workloads, while outperforming it significantly in multi‑table, high‑concurrency scenarios. Although StarRocks’ bulk‑write performance lags behind ClickHouse, the overall gains in query latency and operational simplicity justified the switch.

Advertising Scenario & Real-time Processing – In advertising, data pipelines ingest external video platform metrics and internal user/transaction data. By adopting a Flink‑on‑StarRocks architecture, Zhongan achieved a 3–5× speedup in real‑time reporting, reducing query response times from over 10 seconds to a few seconds, thereby enabling timely business decisions.

Conclusion – The presentation demonstrated how Zhongan’s data team leveraged a fine‑grained, high‑performance data platform and StarRocks to solve real‑time analytics challenges, improve operational efficiency, and drive digital transformation across finance and insurance services.