A large language model (LLM) is an artificial intelligence system trained on massive text corpora using deep learning techniques, enabling it to generate, summarize, classify, translate, and reason about human language. GPT-4, Claude, and Gemini are well-known examples.
In customer support, LLMs power several high-value capabilities:
- Response drafting: Suggesting or auto-composing replies based on conversation context.
- Summarization: Condensing long ticket threads into a short case summary for an agent.
- Knowledge extraction: Pulling relevant answers from documentation without requiring exact keyword matches.
- Sentiment and intent analysis: Inferring customer mood and purpose beyond simple rule-based classifiers.
LLMs are typically accessed via an API (application programming interface) and embedded into existing support platforms rather than deployed standalone. Key evaluation metrics for CX use include answer accuracy, hallucination rate (how often the model fabricates information), and latency. Because LLMs can produce plausible but incorrect responses, human review and retrieval-augmented generation (RAG) guardrails are essential in support contexts.