In recent years, Sichuan Postal Branch has independently developed an AI identification algorithm model for surveillance and a corresponding AI monitoring system targeting 15 scenarios with operational risks including large cash management and unauthorized agency operation at postal outlets. The pilot of “AI recognition plus manual review” has significantly improved the efficiency, coverage and precision of risk monitoring in “daily surveillance review” of agency financial services, with the model leading the industry in its number of application scenarios, algorithm complexity and recognition accuracy.

Sichuan Postal Branch has achieved data fusion between its business system and the video surveillance system to endow video data with business attributes, and generated an evidence chain of the complete transaction process complete with transaction information, AI analysis results and multi-angle video segments, thereby comprehensively enhancing the capabilities of precisely analyzing and identifying complex scenarios. At the same time, Sichuan Postal Branch has modularized its business processes and established an atomic algorithm model. This model enables rapid, flexible assembly of complex business scenarios to meet the needs of risk identification in different business scenarios, eliminating scenario scalability constraints. Currently, it can cover 90 percent of business scenarios in postal outlets.

Leveraging its new technology architecture, Sichuan Postal Branch has established algorithm-driven scenarios by mapping the entire business process. Focusing primarily on key risk points in agency financial services, it can conduct a second review of the complex or error-prone scenarios by applying advanced large models such as DeepSeek and Tongyi Qianwen, to improve the overall accuracy of AI identification. The adoption of the new technology architecture and domestically manufactured equipment has reduced the promotion and operation costs of the AI early-warning management system by over 60 percent.