Intelligent Customer Service System

A large state-owned telecom operator with hundreds of millions of users faced massive daily inquiries. Traditional call centers had high costs, high attrition, and peak-time access issues, with user experience needing improvement.

We built a next-generation intelligent customer-service platform powered by advanced NLP:

• Multi-turn Dialogue Engine: Based on BERT pretraining and deep reinforcement learning, enables complex context understanding and multi-round interactions for scenarios such as balance inquiry, plan handling, and fault reporting.
• Telecom Knowledge Graph: Constructed a telecom knowledge graph with hundreds of thousands of entities to support precise reasoning and Q&A.
• Human–Machine Collaboration Console: When bots fall short, the assistant recommends standard scripts and solutions to human agents in real time, improving efficiency.
• Omnichannel Access: Unified access via apps, WeChat, website, and SMS to ensure consistent service experience.

The project tackled two major challenges: semantic understanding and integration with business systems:

• Corpus Cleaning & Annotation: Processed hundreds of millions of historical conversation logs to build a high-quality telecom domain corpus.
• Model Training & Iteration: Adopted active learning and continuously used online data to improve model performance.
• System Integration: Connected more than ten core systems including billing, CRM, and network management to enable closed-loop operations.

The intelligent customer service system delivered substantial cost reductions and efficiency gains:

• Serves over 2 million users daily, with deflection rates consistently above 60%.
• Saves tens of millions of RMB annually in agent seat costs.
• Customer satisfaction increased by 15%, with first-contact resolution greatly improved.

With AI-empowered services, the client shifted from a labor-intensive to a technology-intensive service model, setting an industry benchmark.