Think Build Implement Repeat
AI & Machine Learning

RAG or Fine-Tuning? A Decision Guide for Business Owners

Last updated:

The question to ask first

When an AI feature gives a disappointing answer, there are only two underlying causes. Either it did not know something it needed to know, or it knew and expressed it wrongly.

Retrieval fixes the first. Fine-tuning fixes the second. Teams routinely reach for fine-tuning to solve knowledge problems, spend a month on it, and end up with a model that is confidently wrong in a more polished way.

What retrieval actually does

Retrieval-augmented generation looks up relevant material at question time and puts it in the prompt. Your documents stay where they are, updates are immediate, and the answer can cite its source.

  • New or changed documents are available instantly — no retraining
  • Answers can be traced back to a source, which matters for trust and audit
  • Access control can be enforced at retrieval time, per user
  • Cost is per query rather than a large one-off

The trade-off is that every query carries retrieval cost and latency, and quality depends heavily on how documents are chunked and indexed. That work is where the skill sits.

What fine-tuning actually does

Fine-tuning adjusts the model's weights on examples of the behaviour you want. It is very good at teaching a consistent output format, a house tone, or a classification scheme with your specific labels.

It is poor at teaching facts. The knowledge gets blurred into the weights, it cannot be cited, it cannot be updated without retraining, and the model will happily generalise beyond what you taught it.

A straight comparison

RetrievalFine-tuning
FixesMissing knowledgeWrong behaviour or format
Setup effortDays to weeksWeeks, plus dataset preparation
UpdatingAdd the documentRetrain
CitationsYesNo
Typical cost£8,000–£30,000 build, then per query£15,000–£60,000 including data work
Access controlStraightforward per userNot possible — it is in the weights

When fine-tuning genuinely wins

Three cases in our experience. First, a rigid output format that prompting keeps drifting away from at volume. Second, a specialist classification scheme with thousands of labelled examples and vocabulary the model has not seen. Third, cost at very high volume, where a small fine-tuned model replaces a large general one for a narrow task.

Notice what these have in common: a large number of examples of the desired behaviour already exists. If you would have to create the training data from scratch, the case is much weaker.

The order to do things in

  1. Prompt properly first, with clear instructions and a few examples. This solves more than people expect.
  2. Add retrieval if the failures are knowledge failures.
  3. Improve chunking, indexing and re-ranking before doing anything more exotic.
  4. Only then consider fine-tuning, and only for behaviour that survived all of the above.

Most projects stop at step three, correctly. The teams that jump to step four first are usually solving a problem they have not yet diagnosed.

Frequently asked questions

Can we do both?

Yes, and for mature systems it is common: a fine-tuned model for consistent behaviour, with retrieval supplying current facts. Sequence matters though — get retrieval right first, because it will change what behaviour you actually need.

How much data do we need to fine-tune?

For format and tone, a few hundred good examples can be enough. For classification with many labels, low thousands. Quality and consistency of the examples matters far more than quantity, and inconsistent labels actively teach the model to be inconsistent.

Does retrieval keep our documents private?

It keeps them in your control: the documents live in your store and only relevant fragments are sent at query time. Whether those fragments leave your infrastructure depends on which model provider you use and under what terms, which is a decision worth making explicitly.

How long does a retrieval system take to build?

A working first version over a defined document set is typically four to eight weeks including evaluation. Getting from adequate to genuinely good is mostly iteration on chunking, retrieval and re-ranking rather than new components.

Keep reading

Not sure which problem you have?

Send us three answers your current system got wrong. We can usually tell from those whether you have a knowledge problem or a behaviour problem.

Book a free 30-minute call Get a project estimate WhatsApp us

Related services

What we build for problems like this one

AI AgentsMachine LearningAI Integration