Call center agents are transforming from 'information transmitters' to 'problem-solving experts,' and LLM-driven agent assistance systems are the core engine behind this shift. According to the CCW Digital 2025 survey, enterprises that have deployed LLM-based agent assistance tools saw a 32% increase in average daily handle volume per agent, while customer complaint rates dropped by 18%.
The latest technology trend focuses on 'real-time knowledge retrieval and generation.' Unlike traditional knowledge bases that rely on fuzzy matching, systems based on Retrieval-Augmented Generation (RAG) can extract the most relevant fragments in real time from enterprise documents, historical tickets, and external knowledge sources, and generate context-appropriate response suggestions. The latest versions of Salesforce and Zendesk have both integrated this feature, saving agents an average of 45 seconds per interaction by eliminating the need for manual searches during calls.
Industry case: A leading U.S. health insurance company deployed an agent assistance system fine-tuned on LLaMA 3. The system not only provides policy explanations and claims process suggestions but also analyzes real-time call sentiment to prompt agents when to express empathy or escalate. After six months of deployment, the first-call resolution rate improved by 26%, and employee turnover decreased by 14%.
GlobalConnect has launched its 'Intelligent Agent Companion' service in this field, featuring a 'multi-model collaboration' architecture: a lightweight model monitors conversations in real time and extracts key information, while a larger model performs deep inference in the cloud. This architecture keeps latency under 200 milliseconds while reducing inference costs by 55%. The service has already helped a European e-commerce company shorten its agent training cycle from 4 weeks to 10 days.
Looking ahead, LLMs will empower agents with 'predictive suggestion' capabilities—before customers even raise a request, the system will preemptively anticipate the next action based on historical patterns and current dialogue flow, proactively prompting the agent. It is expected that by 2026, over 50% of high-performing call centers will adopt such proactive collaboration systems.