Grounding

Retrieval-augmented generation grounds answers by retrieving an organisation’s own documents at the moment of answering and passing them to the model as context. It keeps responses current without retraining, and it makes each answer traceable to the specific source the system actually consulted.

A hallucination is a fluent but unsupported answer produced when a model is not grounded in real data. In customer service the risk is specific: an invented policy, price, or delivery date reads as authoritative to the customer and creates a commitment the business never made.

A knowledge base is the maintained source of truth that AI agents and human agents both answer from. Its accuracy sets a ceiling on the quality of every automated answer, which makes knowledge maintenance an operational requirement rather than a documentation project owned by nobody in particular.

Guardrails are the policy, tone, and action limits an AI agent is not permitted to cross, such as refusing to issue refunds above a threshold or to discuss topics outside its remit. Guardrails are enforced in the system rather than merely requested in the prompt.