Research

A Practical Guide to RAG on Business Data

How to ground LLMs in your documents without leaking sensitive data.

8 min read

Retrieval-augmented generation (RAG) lets a language model answer questions using your own documents. Done well, it gives accurate, cited answers. Done badly, it confidently makes things up — or exposes information to the wrong people.

1. Prepare the content

Clean and de-duplicate documents, keep metadata such as owner, date and access level, and split them into meaningful chunks — by section, not by arbitrary character counts.

2. Respect permissions

Filter retrieved chunks by the user’s access rights before they ever reach the model. Security must be enforced in the retrieval layer, not left to the prompt.

3. Measure quality

Build an evaluation set of real questions with expected answers, and track:

  • Retrieval accuracy — did we find the right sources?
  • Answer faithfulness — is every claim supported?
  • Helpfulness as rated by real users.

4. Keep it fresh

Re-index documents automatically when they change, and show users the source and date of each answer so they can judge it themselves.

Back to all insights

Keep Reading

Have a Question About AI?

Our team is happy to discuss how these ideas apply to your organisation.