E7.1Included
Is it needed?
We test instructions and document search first, and only go further if they fall short.
Detail E7 — Fine-tuning
A model trained on your own examples.
Fine-tuning means training an existing AI model further on examples of your work, so it learns your format, your style or your way of sorting things. It's powerful for the right job and unnecessary for most. We first check whether better instructions or your own documents would do, and fine-tune only when they won't.
- Price
- On request
Fig. E7.1 — Do you need fine-tuning?Not to scale
Better instructionsTry this first
Your documentsAdd your knowledge
Fine-tuningP1When the first two aren't enough
Notes — When it's worth it
Jobs fine-tuning is good at.
- 01Producing a very specific format or style, every time, like your reports, product listings or replies.
- 02Sorting or labelling large volumes consistently, like tickets, documents or transactions.
- 03Getting a smaller, cheaper or private model to do one job as well as a large one.
- 04Using specialist language that general models handle badly.
Specification — When it isn't
When it's the wrong tool.
- Facts that changePrices, stock or policies belong in documents the model looks up, not in its training.
- Too few examplesFine-tuning needs a good set of reviewed examples. Without them, results disappoint.
- A quick fixBetter instructions often solve the problem in a day. We try that first.
Table E7 — Scope
What a fine-tuning project includes.
E7.2Included
Training examples
Your examples gathered, cleaned and checked, because their quality decides the result.
E7.3Included
Model choice
A commercial model with fine-tuning options, or an open-weight model you can run privately.
E7.4Included
Training and testing
The model trained, then compared with the original on examples it hasn't seen.
E7.5Included
Putting it to work
Connected to your workflow or assistant, with monitoring for drift and mistakes.
General notes — FAQ
Questions about fine-tuning.
How many examples do we need?
It depends on the job. Simple formatting can work with a modest set; complex tasks need more. The quality of the examples matters more than the number, and we'll tell you after looking at what you have.
Will it learn our prices and products?
Not reliably, and it shouldn't try. Facts that change belong in documents the model looks up. Fine-tuning is for how it writes and decides, not what it knows.
Can a fine-tuned model stay private?
Yes, if we fine-tune an open-weight model and run it on infrastructure you control. See private and open-weight AI.
Fig. E7.2 — Next step
Wondering if fine-tuning would help?
Tell us the job you have in mind and what examples you have. We usually reply the same working day.