Intelligent Customer Service System for a Major State-Owned Bank

| Project Background

The intelligent customer service system supports core credit card operations—including issuance, transactions, and account management—across the bank. However, rapid business growth and evolving regulatory requirements have exposed critical system limitations, highlighting the following pain points:

1.     Inelligent Automation Deficit: Heavy reliance on manual intervention resulted in high operational costs and low efficiency.

2.     Architectural Constraints: A monolithic structure led to tightly coupled modules, requiring full redeployment for new features and prolonging release cycles.

3.     Performance Bottlenecks: During peak traffic, latency increased significantly, with concurrent volumes approaching maximum system capacity.

4.     Inefficient O&M: The absence of automated monitoring increased dependency on manual oversight, slowing troubleshooting efforts and failing to meet RTO and RPO targets.

5.     Security & Compliance Gaps: Urgent need for technological modernization to comply with domestic innovation mandates.

| Solution

Building upon the existing customer service platform, the new intelligent customer service system was designed and deployed with the following innovative features compared to traditional solutions:

Business Layer :

1.     Agent System: Integrated NLP engines and real-time speech recognition (ASR) to analyze customer intent during calls, automatically recommend response templates, and trigger risk control alerts simultaneously.

2.     Ticket Management: Leveraged a in-house adapted distributed workflow engine and rules engine to enable cross-system ticket routing and early warning for ticket handling.

3.     Intelligent Outbound Calls: Utilized TTS synthesis, dialog policy engines, and sentiment analysis to automate outbound scenarios such as marketing and follow-up surveys.

4.     IVR + Intelligent Voice Navigation: Combined ASR and NLU to support high-frequency services (e.g., bill inquiries) through voice and keypad input, allowing natural language processing for straight-through processing, reducing customer wait time, call volume, and manual service pressure.

5.     Intelligent Quality Inspection: Achieved 100% call record transcription via ASR, with a rule engine containing 200+ compliance checkpoints (e.g., “Was the APR disclosed?”, “Was customer identity verified?”) to automatically identify agent non-compliance, service attitude issues, and knowledge gaps.

6.     Knowledge Base: Integrated 100,000+ credit card business rules to construct a domain-specific knowledge graph, linking entities (card types, interest rates, promotions) and rules (risk policies, billing logic), and enabling natural language search.

7.     Smart Assistant: Powered by NLP and conversational AI, it automatically identifies user intents (e.g., repayment failure, credit limit adjustment), delivering 24/7 online support with a first-contact resolution rate exceeding 80%.

Technical layer:

1.     Layered architecture design: Adopts a four-tier "data layer-core layer-business layer-presentation layer" structure. The data layer utilizes ShardingDS+MyBatisPlus ORM framework for data persistence operations; the core layer encapsulates transactions, unified exception handling, data source configuration, and pluggable configurations for other components; the business layer implements service registration/discovery and dynamic configuration via Nacos, constructing 20+ microservice units with 92% business module reuse rate; the presentation layer employs technologies like Vue frontend framework for interface display, predominantly using AJAX asynchronous techniques to enhance system usability and convenience while effectively reducing system load and network bandwidth consumption.

2.     High-Efficiency O&M:Cross-region active-active architecture design, adopting cross-data-center network solutions to achieve system-level disaster recovery with RT0<30 minutes and RPO<30 minutes; automated operations: completing new architecture handover upon deployment, integrating automated deployment, unified monitoring dashboards, ELK log analysis, pre-configured 50+ common fault handling manuals, and supporting automated diagnosis for frontline/second-line issues (e.g., memory leaks, slow queries).

3.     Open Internet-Integrated System:The technological advancement coexists with independent controllability, utilizing mainstream open-source components in the market such as Apache Kafka (message queue), Elasticsearch (search), Redis (cache), and RocketMQ to ensure technological advancement. Additionally, secondary development and reinforcement of open-source software (e.g., the development of a workflow engine, which reduces 80% of the workload in ticket workflows) meet financial-grade security and compliance requirements.

4.     AI Engine for Deep Business Empowerment:Integrated AI models to build intelligent applications, such as smart knowledge bases, intelligent quality inspection, and intelligent outbound calls; implemented intelligent voice navigation services based on speech recognition (ASR) + natural language understanding (NLU) technology; integrated NLP models to achieve automatic responses to high-frequency queries like credit card bill inquiries and repayment plan recommendations, with a manual replacement rate of 60%.

| Outcomes

Following the deployment of the intelligent customer service system, the bank’s credit card business achieved breakthroughs in both technology and operations. Key performance indicators improved significantly:

1. Transaction Processing Performance:System throughput increased by over 5x, with 98% of core transactions completed in under 200ms.

2. Staff Utilization Efficiency:IVR and intelligent voice navigation now handle 40% more self-service requests, reducing manual       agent workload by more than 50%.

3. Accelerated Development and Delivery:Development cycles  shortened by 50%+, with release timelines compressed from four weeks to under two weeks; urgent requests can now be deployed within 72 hours.Model-driven and low-code development reduced coding effort by 40%, allowing developers to focus on high-value innovation.Component reuse rates exceeded 90%, with over 10 common service modules—including message middleware, caching proxies, and transaction brokers—widely adopted across business functions (e.g., ticketing, agent system, backend management), cutting duplicate development work from 50% to below 20%.

4. Significantly Improved System Quality:

● Unit test coverage reached 70%, with defect detection rates rising to 80% through integrated Swagger API testing and SonarQube code scanning.

● Production incident  frequency decreased by 60%, while key transaction success rates (e.g., instalments) remained consistently above 99.99%, supported by canary releases and circuit-breaker mechanisms.

● Enhanced troubleshooting capabilities via comprehensive monitoring and diagnostic tools enabled rapid fault identification and resolution.

5. Enhanced Compliance and  Security:

● Standardization compliance reached 100% across all development activities through DevOps-enforced code and security baselines (e.g., SQL injection prevention, JWT authentication), raising service compliance rates from 65% to 98%.

● Zero security incidents occurred, ensured by end-to-end encryption in transit and at rest, alongside strict access controls.

● Full adoption of domestic technologies (100% localization), including ARM-based Kylin OS and PostgreSQL databases, certified by central bank financial technology standards.