AI & LLMs
What is fine-tuning?
Fine-tuning is further training an existing language model on your own examples so it reliably produces a specific format, style, or behaviour.
What it fixes and what it does not
Fine-tuning teaches behaviour: a house tone of voice, a strict output schema, a classification convention specific to your business. It does not teach facts. If the model needs to know your current inventory or a policy that changes weekly, that is retrieval, not fine-tuning - and putting facts in weights means retraining every time they change.
Try the cheaper things first
In order of cost: a better prompt, then few-shot examples in the prompt, then retrieval over your documents, then fine-tuning. Most problems people bring to fine-tuning are solved two steps earlier. Fine-tuning earns its place when you need consistency at volume that prompting cannot hold, or when you want a smaller cheaper model to match a larger one on a narrow task.
What it costs in practice
The training run is rarely the expensive part. The dataset is - a few hundred to a few thousand high-quality examples that genuinely represent the behaviour you want, built and reviewed by someone who knows the domain. Then an evaluation set you did not train on, or you have no way to know whether the fine-tune helped.
The maintenance cost people forget
A fine-tuned model is pinned to a base version. When the provider ships a better base model, your fine-tune does not inherit it - you retrain. Budget for that cycle before you commit, and keep the training dataset as a first-class asset rather than a one-off script output.
We build automation systems for small teams. Free audit call to map your specific workflows - no pitch, just a plan.