Founders often ask whether they should fine-tune a model on their data. Usually the better first question is: what do you want the model to change — what it knows, or how it behaves?
Use RAG when the model needs to know things
Retrieval-augmented generation looks up relevant documents at question time and gives them to the model as context. The model's knowledge stays current because you update the documents, not the model.
Use fine-tuning when the model needs to behave differently
Fine-tuning adjusts the model's weights with examples. It is good at teaching a format, a tone, or a narrow task, and at making a small model perform like a larger one on that task. It is a poor way to add facts: the model can still blend them incorrectly, and there is no source to show the user.
Often, the answer is both
A common production pattern is a fine-tuned small model that follows your output format exactly, fed by a retrieval layer that supplies the facts. Start with retrieval and good prompts, measure with real questions, and fine-tune only when you can name the behaviour you need to change.
