Sigma Logic AI Lead with AI. Thrive with Innovation.
Architecture

Fine-tuning, retrieval or a better prompt

The three levers fix different problems. Why fine-tuning does not reliably teach facts, what it is genuinely good at, and the order to try them in.

On this page 11 sections
  1. Key takeaways
  2. Who this applies to
  3. The question each lever answers
  4. Why fine-tuning does not teach facts reliably
  5. What fine-tuning is genuinely for
  6. The order, and when to stop
  7. What each one costs to own
  8. What we actually do
  9. When this is not worth deciding carefully
  10. Frequently asked questions
  11. Next step

Prompting fixes instruction and format problems. Retrieval fixes knowledge problems. Fine-tuning fixes behaviour problems, and it does not reliably teach facts. Most teams asking about fine-tuning describe a retrieval problem, so the expensive lever gets pulled for something the cheap one solves better. Try them in that order, with a test that says when to stop.

“Should we fine-tune?” is almost never the real question. The real question is what kind of wrong the system currently is.

Key takeaways

  • The three levers answer three different questions. Match the lever to the symptom, not to the budget.
  • Fine-tuning teaches form and behaviour. Facts learned this way cannot be updated without retraining, and cannot be cited.
  • Retrieval is the right answer for anything that changes, anything that needs a source, and anything with permissions attached.
  • Prompting is not the amateur option. It is the fastest to change and the easiest to get wrong at length.
  • Cost of ownership matters more than build cost: who is allowed to change this, and what happens when the base model is deprecated.

Who this applies to

You have a working system that is wrong in some specific way, and someone has proposed fine-tuning as the fix. Or you are scoping a build and deciding how much of the behaviour to bake in.

If you have not yet measured what kind of wrong it is, that comes first. Measuring whether an AI system works is the prerequisite, because all three levers are indistinguishable from each other when the only evidence is an impression.

The question each lever answers

SymptomThe leverWhy
It does not know our products, prices, policiesRetrievalThe facts change, need a source, and are often permissioned
It ignores an instruction, or follows it inconsistentlyPromptingThe instruction is present but competing with others
Output shape is unreliableStructured output, then promptingA schema constraint is cheaper and stricter than either alternative
Tone is wrong, or the format is house-specific and hard to describeFine-tuningStyle is easier to demonstrate with examples than to specify in words
A narrow classification the base model keeps getting wrongFine-tuningSmall, closed task with plenty of labelled examples
It is correct but too slow or too expensiveFine-tune a smaller modelThe one case where fine-tuning is a cost lever, not a quality lever

The first row is the one that gets misdiagnosed. “The model does not know our returns policy” sounds like a knowledge gap in the model, so training on company documents sounds like the fix. It is a knowledge gap in the request, and the fix is putting the policy in front of the model at the moment it answers.

Why fine-tuning does not teach facts reliably

Fine-tuning adjusts the model’s weights on examples. It is very good at shifting the distribution of how the model answers, and poor at installing a specific fact you can then depend on.

Three consequences follow, and each one is a reason to prefer retrieval for knowledge:

You cannot update a weight the way you edit a document. When the returns window changes from 30 days to 14, a retrieval system needs one paragraph rewritten. A fine-tuned model needs another training run, and until it happens the system is confidently wrong.

There is no source to show. A fine-tuned answer arrives with no passage attached, so it cannot be cited, cannot be checked by the customer, and cannot be audited later. This is exactly the property that makes unsupported answers the measurable metric rather than wrong ones.

Teaching new facts can make hallucination worse. Training a model on statements it has no basis for teaches the form of confident specificity along with the content. The model learns that answering this kind of question with a precise-sounding detail is what it does here, and it generalises that behaviour to questions where it has no detail.

Parameter-efficient methods have made fine-tuning far cheaper to run. They did not change what it is good at.

Knowledge in the weights compared with knowledge in the context Two places an answer can come from. Baked into the weights, meaning fine-tuned on your facts: changing a price needs another training run, and the answer carries no source and no permissions. Supplied in the context, meaning retrieved at the moment of asking: changing a price means editing one paragraph, and the answer can be cited and filtered per user. Until a retrain lands the trained model is confidently wrong, and nothing in the answer itself tells a reader which of the two it came from. Where the answer comes from, and what that costs you Baked into the weights Fine-tuned on your facts Change a price: another training run NO SOURCE, NO PERMISSIONS Supplied in the context Retrieved when asked Change a price: edit one paragraph CITED, AND FILTERED PER USER Until the retrain lands, the trained model is confidently wrong. Nothing in the answer tells the reader which of the two it came from.
The difference is not accuracy on the day it ships. It is what happens on every day after that, when the fact changes and the model does not.

What fine-tuning is genuinely for

It is not a last resort, and dismissing it is as lazy as reaching for it first. Four cases where it is the right tool:

A house format you cannot describe. If your output has to match a document style with a hundred small conventions, twenty examples teach it better than two pages of prose instructions, and the prompt gets shorter rather than longer.

Tone that survives. Prompted tone drifts as conversations lengthen and as prompts accumulate. Trained tone does not.

A narrow classification with real labelled data. Routing, triage, extraction into a fixed taxonomy. Closed task, thousands of examples, no facts required.

Cutting cost and latency. Fine-tune a small model on the outputs of a large one for your specific task. This is the one place fine-tuning routinely pays, and it is a cost decision rather than a quality one - which means it needs a before number or you will not be able to tell what you gave up.

The order, and when to stop

Work up the list. Stop at the first lever that closes the gap, and require a measurement before moving to the next.

1. Fix the prompt, then shorten it. Most under-performing prompts are not too short. They have accumulated a year of instructions, several in tension, and the model is resolving the conflict in a way you did not intend. Removing an instruction is a change worth testing, which is what an evaluation set makes safe - see regression testing prompts like code.

2. Fix retrieval. If the answer exists in your documents, the failure is usually that the right passage was never retrieved, not that the model could not use it. Measure retrieval hit rate separately from answer quality, because end-to-end scoring hides which half is broken. The six causes and the test for each are in why retrieval misses the answer that is there.

3. Constrain the output. If the complaint is shape rather than substance, structured output is stricter and cheaper than training.

4. Then consider fine-tuning, with a held-out set you have not touched, and a specific number you expect it to move.

The exit test at each step is the same: what number would tell me this worked, and do I have it from before the change? If the answer is no, the next lever will not help either, because you will not be able to tell whether it did.

The order to try the three levers in Four steps. First fix the prompt and then shorten it, which takes minutes to change. Second fix retrieval, measuring hit rate on its own. Third constrain the output, which is stricter than training and cheaper. Fourth, and only then, consider fine-tuning, with a held-out set and a named number it is expected to move. The exit test is the same at every step: what number says this worked, and do I have it from before the change? Without that number the next lever will not help either, for the same reason. Stop at the first lever that closes the gap 1 Fix the prompt, then shorten it minutes to change 2 Fix retrieval measure hit rate on its own 3 Constrain the output stricter than training, and cheap 4 Then consider fine-tuning held-out set, and a named number The exit test is the same at every step. What number says this worked, and do I have it from before? Without that number the next lever will not help either.
Each step is cheaper to reverse than the one below it. Fine-tuning is the only one that is hard to undo, which is why it is last rather than because it is ineffective.

What each one costs to own

Build cost is the number that gets discussed. Ownership cost is the one that decides how the system ages.

PromptingRetrievalFine-tuning
Time to first versionHoursDays to weeksDays, plus data preparation
Cost to changeMinutesEdit a documentA training run
Who can change itAnyone with review accessWhoever owns the contentAn engineer, with data
Facts stay currentOnly if in the promptYes, by editing the sourceNo, until retrained
Answer can cite a sourceNoYesNo
Permissions per userNoYesNo
Survives a base model changeUsually, with regression testingYesNo, retrain against the new base

That last row is the one that surprises people. A fine-tune is tied to a base model version, and base versions get deprecated on a published schedule. When that happens you are not choosing whether to retrain, you are scheduling it - the same forced-migration problem described in your provider changed the model, with more work attached.

The permissions row rules fine-tuning out entirely for a class of systems. If two employees must see different answers from the same question, the difference has to live in retrieval, because a model’s weights cannot be filtered per user. That is the subject of RAG permissions.

Six symptoms and the lever each one actually needs Six symptoms with the lever each needs. It does not know our policies needs retrieval. It ignores an instruction needs prompting. The JSON is unreliable needs structured output. The tone is wrong needs fine-tuning. A narrow class always being wrong needs fine-tuning. Correct but slow and expensive needs fine-tuning a smaller model. Three of the six are commonly misdiagnosed as a training problem, because the request arrives as should we fine-tune whatever the symptom is. What people ask for, and what the symptom actually is It does not know our policies Retrieval It ignores an instruction Prompting The JSON is unreliable Structured output The tone is wrong Fine-tuning This narrow class is always wrong Fine-tuning Correct, but slow and expensive Fine-tune a small model Three of the six are misdiagnosed as a training problem. The request arrives as "should we fine-tune" either way.
The bottom three are real fine-tuning cases and they are the minority. The question asked is almost never the question that needs answering.

What we actually do

We have never opened an engagement with fine-tuning, and we have said so to clients who had already budgeted for it. It has cost us work at least twice, because “you do not need the expensive thing” reads as a smaller engagement, and sometimes the buyer has already sold it internally as the plan.

The reasoning is the same every time. Fine-tuning is the only one of the three levers that is hard to reverse, hard to explain to the person who owns the content, and dependent on a base model somebody else controls. Everything it does for knowledge, retrieval does better and cheaper. What is left - format, tone, narrow classification, distilling for cost - is real, and it is a smaller share of the requests we get than the requests suggest.

Our default sequence on a support or knowledge system is retrieval first with hit rate measured separately, a deliberately short prompt, structured output where the shape matters, and fine-tuning considered only once there is a held-out set and a number it is expected to move. On roughly nine builds in ten, that number never gets a chance to justify the work, because the gap closes earlier.

When this is not worth deciding carefully

If the system handles low volume and low stakes - an internal drafting helper, a prototype in front of five people - use the best model with a good prompt and move on. The comparison in this article costs more to run than the difference is worth at that scale.

The same applies if you are still deciding what to build. Architecture decisions taken before the workflow is settled tend to be re-taken, and the cost of the analysis is the loss.

And if your volume is genuinely small, a fine-tune to save cost is arithmetic that does not work. The training run, the data preparation and the eventual forced migration are fixed costs; you need the traffic to amortise them.

Frequently asked questions

How many examples does fine-tuning need?

For style and format, low hundreds of good examples usually beats thousands of mediocre ones. For classification, more is genuinely better. The binding constraint is almost always label quality rather than count, and preparing that data is the part teams underestimate.

Can we fine-tune and use retrieval together?

Yes, and for a specialist domain it is often the right combination: fine-tune for the form of the answer, retrieve for its content. The mistake is using fine-tuning to supply content that retrieval should be supplying.

Is fine-tuning cheaper now that adapters exist?

The training run is much cheaper. The data preparation, the evaluation and the forced migration when the base model is deprecated are not, and they are the larger share of the cost over a year.

Our vendor says their fine-tuned model is more accurate. How do we check?

Ask what it was measured against, whether the cases were held out, and what the base model scored on the same set. Without those three, “more accurate” is a claim about a demo. The same test applies to any accuracy number in a proposal - see how to evaluate an AI agency proposal.

What if the answers are wrong in a way retrieval clearly cannot fix?

Then check whether the task is genuinely ambiguous before reaching for training. If two of your own experts would answer differently, no amount of fine-tuning produces a right answer, and you have found the ceiling instead - the measurement is in testing against the human baseline.

Next step

If someone has proposed fine-tuning and nobody can say which number it is expected to move, that is worth an hour before it is worth a budget. The AI consulting and strategy engagement covers this choice as a scoping question, and the answer is frequently the cheaper lever.

Related: Context windows and long documents · Structured output: reliable JSON · RAG permissions · Regression testing prompts like code · AI consulting and strategy

Let's talk

Got a workflow this applies to?

Describe it in a couple of sentences. We will tell you whether it is worth automating, what we would build, and roughly what it takes.