A company wants an LLM-based chatbot to provide customer-service agents with contextual, policy-based answers in real time. The knowledge source is the company's policy documents. What is the most cost-effective solution?

RAG is the most cost-effective solution because it dynamically retrieves relevant policy documents and injects them into the LLM's prompt at runtime. This allows the LLM to generate answers based on current, accurate information without requiring expensive and time-consuming retraining or fine-tuning of the entire model. Retraining or fine-tuning the entire LLM is significantly more expensive and computationally intensive, especially for large models, and would require frequent updates as policies change. Additional pre-training and data augmentation are also costly and overkill for this specific use case, as the goal is to leverage existing policy documents, not to fundamentally alter the LLM's core knowledge.