TechnologyGuide 1 of 4

What is RAG?

How retrieval-augmented generation connects language models to current, authorised and verifiable enterprise knowledge.

14 min readBeginner to intermediateReviewed July 2026
If you only have one minute

Retrieval-augmented generation gives a model relevant information at the moment a question is asked. Instead of expecting the model to remember company knowledge, the platform retrieves authorised sources and places them in context.

RAG is often the fastest route from a generic model to a useful enterprise assistant. It supports current information, citations and controlled access without retraining the model whenever a document changes.

If we could give you one piece of advice

Design retrieval around permissions and evidence, not just semantic similarity.

There is rarely one universally correct architecture. The best decision is the one that fits the process, risk, people and operating capability of the organisation.

Why it matters

RAG is often the fastest route from a generic model to a useful enterprise assistant. It supports current information, citations and controlled access without retraining the model whenever a document changes.

In practice, value appears when the technology becomes part of a governed process: responsibilities are clear, evidence can be checked and the organisation can observe whether outcomes improve.

What it really means

Retrieval-augmented generation gives a model relevant information at the moment a question is asked. Instead of expecting the model to remember company knowledge, the platform retrieves authorised sources and places them in context.

The useful distinction is between a technical capability and a production service. Enterprise use requires identity, permissions, data handling, evaluation, monitoring and a clear owner for the result.

How it works in practice

The flow can be described in five steps. Each one needs an explicit purpose, an owner and a way to verify that it behaves as expected.

01Documents are ingested while preserving metadata.
02Content is divided into useful chunks.
03Embeddings and search indexes are created.
04Permissions are applied before retrieval.
05The model answers from the selected context and cites sources.

The exact architecture will vary. What matters is keeping the boundaries visible and making failures diagnosable rather than hiding them behind a fluent answer.

Enterprise example

What this looks like in real work

A support team asks about a product warranty. The platform retrieves the current policy for that market and product, excludes superseded documents and presents the answer with links to the source.

The lesson is not that AI produced an answer. It is that the answer appears inside a controlled process, can be verified and is tied to a measurable outcome.

Common mistakes we see

01

Uploading documents without ownership

Content needs a responsible owner, lifecycle and access model.

02

Using fixed chunking everywhere

Contracts, manuals and tables need different segmentation strategies.

03

Evaluating only answer fluency

Retrieval quality and source accuracy matter more than polished language.

04

Ignoring document changes

Deletion, versioning and re-indexing are part of the operational design.

When it makes sense

  • The process has a clear owner and outcome
  • The required information and permissions are understood
  • Results can be checked or measured
  • The organisation can operate the capability responsibly

When it probably does not

  • The problem is still undefined
  • A deterministic rule would be simpler and safer
  • No one owns data quality or exceptions
  • The consequence of failure cannot be controlled

Not using AI can be the right decision. Honest scope is usually more valuable than a technically impressive pilot without a real problem.

How we usually approach it

We work from engineering and operational experience. We do not believe in a universal recipe, but we consistently ask the same questions before building:

Which outcome should improve?
Who is accountable?
Which information is required?
What error level is acceptable?
Which controls are necessary?
How will value be measured?

Once these answers are reasonably clear, choosing models, infrastructure and integration becomes much easier.

If you have made it this far...

Useful enterprise AI is not defined by how impressive the demo looks, but by how reliably it improves a real process.

Technology evolves quickly and we do not claim to have final answers. These guides capture what we have learned while building and operating platforms, shared with the humility that tomorrow may require a better approach.

The question to ask nextWhat evidence would you need before trusting this capability in a real process?
Q
Written by the Quanta team

We are engineers specialising in infrastructure, computing and applied enterprise AI. We share what we have learned from building and operating platforms, grounded in practical experience and in the knowledge that the technology continues to evolve.

Last reviewedJuly 2026

We review our guides to keep them useful, accurate and honest.