Customer Service Automation
CITIC Financial Holdings needed to make its wealth-management platform more intelligent and scenario-driven so customer-service and wealth-adviser interactions could scale across a very large client base.
Reported impact
CITIC improved the quality and scalability of digital wealth-service conversations and created a knowledge-operations loop that can keep expanding as new scenarios are added.
CITIC Financial Holdings and Baidu AI Cloud upgraded the group's wealth-management collaboration platform with a large-model customer-service and adviser layer. The implementation uses Baidu AI Cloud Keyue as the conversational surface and organizes its improvement work into three linked knowledge loops: knowledge expansion, knowledge tuning and the knowledge flywheel. Knowledge expansion uses FAQ mining and task-oriented dialogue techniques to broaden the wealth-adviser knowledge base and cover more service scenarios. Knowledge tuning uses expanded-question generation and semantic-confusion detection to improve intent recognition and raise answer precision. The flywheel layer then closes the loop by feeding knowledge expansion and optimization back into continuing operations. Rather than treating the platform as static FAQ automation, CITIC turned it into an iteratively improved digital-wealth-assistant stack that sits on top of an existing group-wide wealth-management platform and helps advisers and customers interact more effectively.
Customer and adviser questions enter the Keyue conversational layer, where expanded and tuned knowledge assets are used to classify intent, retrieve wealth-service answers and feed improvements back into the knowledge flywheel.
Baidu AI Cloud Keyue and CITIC built the overall solution around knowledge expansion, tuning and a knowledge flywheel (original: 百度智能云客悦与中信金控携手……通过知识扩充、知识调优、知识飞轮三大关键方案).
End-to-end Q&A accuracy exceeded 85% (original: 85%+端到端问答准确率).
Knowledge expansion uses FAQ mining and task-oriented dialogue to broaden wealth-adviser scenarios (original: 基于FAQ挖掘、任务型对话引擎拓展财顾场景应用并提升知识扩充效率).
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Baidu AI Cloud Sample Jun 30, 2025 Industrial Bank Co., Ltd. (兴业银行股份有限公司) Industrial Bank wanted its digital-service channels to deliver better customer experience, broader self-service coverage, and more precise operations as customers moved online, but legacy service models could not scale… Average daily handled volume reached nearly 100,000 calls. Baidu AI Cloud Intelligent Customer Service · Intelligent Speech · Smart Outbound Calling Baidu AI Cloud Sample May 16, 2025 Shanghai Pudong Development Bank Shanghai Pudong Development Bank needed a scalable, contactless way to deliver personalized wealth-management guidance and everyday banking service to a very large retail customer base without forcing customers to wait… If Xiaopu serves 10,000 customers per day for 20 minutes each, it saves 3,333 labor hours per day. AI Middle Platform Solution · High-Precision User Profiling · Knowledge Middle Platform Baidu AI Cloud Sample May 14, 2025 China Life Property & Casualty Insurance Co., Ltd. (中国人寿财险) China Life P&C needed a scalable way to collect, analyze, and act on large volumes of customer NPS survey data across policy, claims, and value-added service touchpoints, especially the unstructured feedback that… Customer sentiment recognition accuracy exceeded 95%. BML Full-Featured AI Development Platform · PaddlePaddle · Sugar BI 