How China Postal Savings Bank Built an Enterprise‑Level AI‑Powered Anti‑Fraud Platform

On September 5, 2023, the Ministry of Industry and Information Technology and the Beijing Communications Administration co‑hosted the 2023 China International Service Trade Fair’s “Enterprise Digital Transformation Forum” in Beijing, focusing on “Digital Leadership for High‑Quality Development” and gathering officials, researchers, operators, and digital‑transformation enterprises.

The forum announced the results of the 2023 “Digital Influence” case‑collection activity, which attracted nearly one hundred enterprises or projects, with rigorous expert review selecting 10 leading enterprises, 5 leading experts, 5 outstanding experts, and 58 innovative cases that reflect the latest trends in China’s corporate digital transformation.

Among the highlighted cases, China Postal Savings Bank’s Enterprise‑Level Intelligent Anti‑Fraud Platform was selected as an innovative case, demonstrating a comprehensive digital risk‑control system.

The platform leverages digital technologies such as stream computing, graph computing, decision engines, machine learning, and graph databases to create a layered defense: pre‑admission customer screening, real‑time transaction warning and interception, and post‑event tracing and analysis. Core capabilities include an enterprise‑level blacklist, feature library, device fingerprinting, rule engine, decision engine, real‑time stream processing engine, query engine, and a configurable risk‑control management console, addressing known low‑loss fraud, unknown medium‑loss fraud, and organized high‑loss fraud.

Innovation case highlights

1. Integrated “one‑stop” defense network – The platform combines credit‑loan and transaction data to build a full‑bank customer risk profile and a unified fraud‑prevention model, enhancing detection and interdiction across the entire transaction lifecycle.

2. Intelligent modeling for higher accuracy – Using image recognition, visual analysis, and graph neural networks, the system automatically verifies faces, IDs, cards, and video feeds, deploying over 300 smart models to evaluate risks for billions of transactions.

3. Stream‑batch unified technology for real‑time processing – A self‑developed metric processing engine merges real‑time, near‑real‑time, and offline data, delivering millisecond‑level metric computation, high precision, and flexible cross‑window calculations, handling about 2 billion transactions daily.

4. Deep customer value mining for risk‑driven credit products – Comprehensive data collection enables precise fraud risk assessment, supporting credit products for rural customers and boosting high‑quality development of consumer, business, and credit loan services.

5. Post‑loan fraud identification for a closed‑loop anti‑fraud chain – Continuous monitoring of overdue collections identifies fraudulent borrowers, reports them to authorities, and feeds the insights back into models to improve accuracy.