Glossary
RAG
Definition
Retrieval-augmented generation (RAG) is a technique where an AI model retrieves relevant passages from your own knowledge base and grounds its answer in them, instead of relying on what the model memorized.
RAG splits answering into two steps: first retrieve the most relevant chunks of your actual content (policies, help articles, product docs), then generate a reply that is constrained to that retrieved evidence. The model answers from your documents, not from its training data.
In customer support this is the difference between "our return window is 30 days" because your policy page says so (with a citation), and a fluent guess. RAG also keeps answers current: update the document and the next answer changes with it, no retraining.
Why it matters
Ungrounded models improvise, and in support an improvised policy is a real liability - customers screenshot promises. Grounding plus citations is what makes AI answers auditable enough for production.
How Resolvo HQ approaches it
Resolvo HQ ingests your help center and policies into a tenant-isolated knowledge base, retrieves against it on every turn, and attaches citations to answers - and when retrieval comes back weak, confidence drops and the conversation routes to a human instead of guessing.
Frequently asked
Does RAG eliminate hallucinations?+
It sharply reduces them but the honest answer is no technique eliminates them - which is why grounding must be paired with confidence scoring, citations a human can check, and escalation when evidence is thin.
What content should an e-commerce store feed into RAG?+
Start with the pages customers already ask about: shipping and return policies, sizing, product FAQs. Live order data comes from the commerce platform directly, not from documents.
See it on your own tickets
Replay real tickets in a sandbox and watch the agent resolve them before you pay.