Intelligent Customer Service System for a Large State-Owned Bank

| Project Background

1. Intelligence Enhancement: Most business processes require manual intervention, resulting in high operational costs and low efficiency.
2. System Architecture Optimization and Upgrading: The monolithic architecture leads to tight coupling of functional modules. New functions require full-version releases, resulting in long version iteration cycles.
3. Performance Improvement: Response delays occur during peak hours, and the actual business concurrency is close to the system’s maximum load capacity.
4. O&M Efficiency Improvement: Lack of automated monitoring leads to heavy reliance on O&M personnel; production problem troubleshooting is slow, failing to meet Recovery Time Objective (RTO) and Recovery Point Objective (RPO) requirements.
5. Security and Compliance Enhancement: Urgent need for information technology innovation (ITI) transformation and upgrading.

| Solution

? Business Level:

1. Agent System: Integrates a Natural Language Processing (NLP) engine and real-time Automatic Speech Recognition (ASR) to realize real-time customer intent analysis during calls, automatic recommendation of script templates, and simultaneous triggering of risk control alerts.
2. Work Order Management: Implements cross-system work order circulation and work order processing alert functions based on a self-developed and modified distributed workflow engine and rule engine.
3. Intelligent Outbound Calling: Enables outbound scenarios such as marketing and follow-ups based on Text-to-Speech (TTS) synthesis, conversation strategy engines, and sentiment analysis.
4. IVR + Intelligent Voice Navigation: Based on ASR and Natural Language Understanding (NLU), it supports high-frequency services (e.g., bill inquiry) through voice and key-press operations, allowing direct processing via natural language. This saves customer time, diverts call volume, and reduces the workload of manual agents.
5. Intelligent Quality Inspection: Converts 100% of call records into text using ASR technology. A rule engine with over 200 preset compliance checkpoints (e.g., "whether the annual interest rate is mentioned" and "whether customer identity is verified") automatically identifies agent compliance risks (e.g., improper commitments), service attitude issues, and business knowledge gaps.
6. Knowledge Base: Integrates over 100,000 credit card business rules to build a credit card domain knowledge graph. It associates entities (card types, interest rates, promotions) with rules (risk control strategies, account logic) and supports natural language retrieval.
7. Intelligent Assistant: Based on an NLP engine and conversational robots, it automatically identifies user intents (e.g., repayment failure, credit limit adjustment) and provides 24/7 online responses with a resolution rate of over 80%.

? Technical Level:

1. Layered Architecture Design: Adopts a four-level "Data Layer - Core Layer - Business Layer - Presentation Layer" architecture:
   • The Data Layer uses the Sharding-JDBC + MyBatis-Plus ORM framework for data persistence.
   • The Core Layer encapsulates transactions, unified exception handling, data source configuration, and pluggable configuration for other components.
   • The Business Layer implements service registration, discovery, and dynamic configuration based on Nacos, building more than 20 microservice units with a business module reusability rate of 92%.
   • The Presentation Layer uses front-end technologies such as the Vue framework for page display. Most of the system adopts AJAX asynchronous technology to improve system friendliness and convenience, effectively reducing system load and network bandwidth usage.
2. Efficient O&M:
   • Designs an active-active architecture for cross-regional applications and adopts a cross-data center network solution to achieve system-level disaster recovery, with RTO < 30 minutes and RPO < 30 minutes.
   • O&M Automation: Completes the transition to the new architecture upon deployment, integrating automated deployment, a unified monitoring dashboard, and ELK (Elasticsearch, Logstash, Kibana) log analysis. It preconfigures over 50 manuals for common fault handling and supports automated diagnosis of first- and second-line issues (e.g., memory leaks, slow queries).
3. Open Internet Integration System: Balances technological advancement and independent controllability:
   • Uses mainstream open-source components such as Apache Kafka (message queue), Elasticsearch (search), Redis (caching), and RocketMQ to ensure technological advancement.
   • Conducts secondary development and reinforcement of open-source software (e.g., the development of a workflow engine reduces the workload of work order workflows by 80%) to meet financial-level security and compliance requirements.
4. Intelligent Engine for In-depth Business Empowerment:
   • Integrates AI models to build intelligent applications such as an intelligent knowledge base, intelligent quality inspection, and intelligent outbound calling.
   • Realizes intelligent voice navigation services using ASR + NLU technology.
   • Integrates NLP models to automatically answer high-frequency questions such as credit card bill inquiries and repayment plan recommendations, achieving a manual substitution rate of 60%.

| Application Effects

1. Transaction Processing Performance: Compared before and after the transformation, the system throughput has increased by more than 5 times, and over 98% of core transactions can be completed within 200ms.
2. Personnel Utilization: The self-service proportion of IVR + intelligent voice navigation has increased by nearly 40 percentage points, reducing the workload of manual agents by over 50%.
3. Doubled Development and Delivery Efficiency, Industry-Leading Response Speed:
   • Development Cycle Reduced by 50%+: The average deployment cycle has been shortened from 4 weeks to less than 2 weeks, and urgent requirements can be launched within 72 hours at the fastest.
   • Significant Empowerment from Models and Low-Code: 80% of public logic is encapsulated in the framework for plug-and-play use, reducing coding workload by 40% and allowing developers to focus on high-value innovation and business requirement implementation.
   • Component Reusability Rate Exceeds 90%: More than 10 general service components encapsulated in the new architecture (e.g., message middle platform, cache proxy, transaction proxy) are frequently used by various business modules (e.g., work orders, agents, back-end management), reducing repeated development from 50% to less than 20%.
4. Improved System Quality, 70% Reduction in Defect Rate:
   • Unit Test Coverage Reaches 70%: Integrates Swagger-level API testing tools and SonarQube code scanning for unit testing, increasing the defect interception rate to 80%.
   • Production Failure Rate Reduced by 60%: System stability is significantly enhanced through mechanisms such as gray release, circuit breaking, and rate limiting, with the success rate of key transactions (e.g., installment plans) remaining stable above 99.99%.
   • Efficient Production Fault Troubleshooting: Quickly locates problems through various monitoring tools.
5. Dual Improvement in Standardization and Security:
   • 100% Development Standardization Rate: The DevOps platform enforces code specifications and security baselines (e.g., SQL injection prevention, JWT token authentication), increasing the compliance rate of business services from 65% to 98%.
   • Zero Security Incidents: Capabilities such as data transmission encryption, storage encryption, and permission control ensure zero transaction leakage.
   • 100% Localization Rate: Core hardware and software have been fully replaced with ITI products (e.g., Kylin OS with ARM architecture, PostgreSQL database) and have passed the People’s Bank of China’s fintech certification.