Privacy Computing Case Study: Multi‑Party Secure Computation for Financial Risk Control by Jiangsu Bank and Ningbo Bank

In the context of industry‑wide digital upgrade and accelerated marketization of data elements, the secure and trustworthy circulation of data elements has become urgent. Privacy computing technology can ensure a balance between data value circulation and privacy protection, achieving “usable but invisible, computable but unidentifiable, controllable and measurable,” and has been maturely applied in inclusive finance, joint risk control, anti‑fraud, and other financial scenarios.

To actively respond to the central bank’s “Financial Data Comprehensive Application Pilot” call, Jiangsu Bank and Ningbo Bank, following the “Multi‑Party Secure Computation Financial Application Technical Specification,” used Ant’s Windhole multi‑party secure computation platform and the privacy‑computing framework “YinYu” to conduct multi‑party secure joint modeling.

China expands inclusive finance to micro‑enterprises, farmers, low‑income urban residents, the poor, disabled, elderly, etc., who have high demand for financial services but lack effective credit data. Financial institutions must first consider risk prevention; risk assessment relies on massive data, and the development of inclusive finance also depends on data circulation.

Privacy computing ensures secure data flow. Jiangsu Bank and Ningbo Bank, based on the “YinYu MPC Financial Risk Control Full‑Chain Solution,” raised security levels across all stages: both parties share common customers and complementary feature dimensions, using joint models to improve approval rates, reduce new‑customer risk, and strengthen mid‑loan risk control.

Compared with traditional single‑institution risk decisions with limited data samples, the new privacy‑computing solution strengthens data value sharing while protecting user privacy, enhancing independent risk‑control capabilities.

The solution, built on YinYu’s layered design and ready‑to‑use privacy‑preserving data analysis and machine learning functions, reduces development and integration workload for risk‑control platforms and lowers the barrier for algorithm teams, enabling deeper application of privacy computing in large‑scale business practice.

The Ant “Windhole Multi‑Party Secure Computation Platform” offers security, efficiency, and industry customization, meeting three high requirements of financial‑grade full‑chain risk control, large‑scale production, and precise decision‑making, and can be easily integrated. It follows the central bank’s standards, combines MPC SQL and other leading technologies to achieve full‑chain secure data joint computation, breaking the data security analysis dilemma for small institutions.

Technical Advantages

2.1 Supports Diverse Scenarios – By abstracting different privacy‑computing technologies through ciphertext computation devices, a single framework can support all mainstream privacy‑computing techniques and be flexibly assembled to meet varied scenario needs.

2.2 Full‑Chain Protection – Provides a complete technical solution for the entire lifecycle of financial risk‑control data, from data discovery, analysis, joint modeling, to joint prediction, supporting MPC SQL operators and facilitating broader application from AI to data analysis.

2.3 Scale Production – Supports multi‑task concurrency, elastic scaling, gray‑release, quick rollback, helping businesses transition from PoC to large‑scale production.

Case Application

The open‑source privacy‑computing framework “YinYu” builds on MPC research, leveraging Ant Group’s extensive financial industry scenarios, meeting the high‑security demands of the full lifecycle of financial risk‑control data, delivering end‑to‑end secure data processing capabilities, and providing a full‑chain technical solution for multi‑party data collaboration in financial risk control.

Jiangsu Bank and Ningbo Bank, using Ant’s Windhole platform and YinYu’s full‑chain capabilities, developed a joint risk‑control model while protecting privacy and data security, enabling safe data value sharing between institutions and offering higher‑quality inclusive finance to micro‑customers.

In this cooperation, the two banks modeled pre‑loan A‑card and mid‑loan B‑card; compared with single‑institution models, the joint model’s KS increased by over 10%. The case has been deployed in more than 20 financial institutions, improving approval rates, reducing new‑customer risk, and strengthening mid‑loan risk control.

Financial risk control involves many stages, each facing data security challenges; providing high‑security MPC solutions for the full chain is difficult, and research on end‑to‑end intelligent risk control is still exploratory. Data analysis and model serving in privacy computing are as challenging as machine learning, with MPC data analysis even more difficult.

The full‑chain MPC solution not only supports financial risk control but will also open models such as NN and DeepFM for marketing recommendation, gradually extending to medical, marketing, and other scenarios.