How do we start?
Select a useful set of questions and approved sources for a bounded pilot, then evaluate answer quality with real users.
Can it update CRM or complete actions?
That needs an explicitly scoped workflow or agent capability with separate permissions and testing.
What if documents conflict?
We agree on source authority and escalation rules so contradictions are not silently treated as certainty.
Can different teams see different information?
Yes, when the identity and source systems provide reliable permissions that can be enforced.
Does RAG eliminate incorrect answers?
No. Evaluation, source controls, refusals and escalation remain necessary.
Can it cite its sources?
Yes. Source references should be part of the answer experience and testing.
Can it use our documents?
Yes, subject to source quality, rights, format, access and an agreed update process.
What does RAG mean?
Retrieval-augmented generation combines a search of approved material with an AI-generated answer. It helps connect the answer to business-specific sources.