New AI System Bypasses Limitations of Retrieval-Based Assistants in Customer Service

Contact center agents face mounting pressure to handle customer inquiries quickly, often spending significant time manually searching knowledge bases. While retrieval-augmented generation (RAG) systems powered by large language models have become commonplace in industry, they come with drawbacks: they can misunderstand customer intent and waste cycles repeatedly pulling the same FAQ content.

A Smarter Hybrid Approach#

Researchers have developed a decision support system that strategically combines two retrieval methods. The system first attempts to identify whether a customer's question matches a known FAQ. If it does, the answer is fetched directly from the FAQ database. For novel or complex queries that don't map to existing FAQs, the system generates a response using traditional RAG techniques. This selective routing sidesteps redundant retrieval while maintaining the flexibility to handle new problems.

Real-World Implementation#

The team deployed their solution as part of an AI-assisted agent platform at Minerva CQ. In practice, the system delivers responses to human agents within 2 seconds—fast enough to support live conversation flow. The deployment has reduced average handling time (AHT) per call and lowered the overall operational burden on contact centers.

Automating FAQ Discovery#

The researchers also introduced an automated, LLM-powered workflow to surface frequently asked questions from historical conversation transcripts when no predefined FAQ collection exists. This agentic approach helps organizations jumpstart their FAQ databases without manual curation, unlocking the efficiency gains even for teams without mature knowledge repositories.

The work demonstrates how thoughtful system design—rather than wholesale replacement of legacy approaches—can improve AI assistance in high-volume, time-sensitive environments.