Hallucination

A hallucination is when an AI model gives a confident answer that is wrong or invented. In analytics it usually does not look like nonsense. It looks like a clean number that happens to be computed the wrong way.

This is the dangerous failure mode for an analytics agent. A blank answer gets ignored. A wrong number that looks right gets pasted into a deck and acted on. The agent did not refuse, it guessed a definition you never gave it.

How to reduce it

Hallucinations in analytics drop sharply when the agent has real context to lean on, the metric definitions, the rules, the schema. That is the job of context engineering. Testing the agent with evaluation catches the ones that slip through.