Fine-tuning
Continuing a model's training on your own examples, so it takes on a form or a task without being told about it in every prompt.
It is good at form and bad at facts. A model can be taught to answer in your house style, to return your exact JSON shape, or to handle a narrow classification; it cannot be taught yesterday's price list, because facts move and weights do not.
Reach for it after the cheaper moves fail: a better prompt, examples in the prompt, and retrieval from a source. Each of those changes in a minute, while a fine-tune has to be trained, evaluated, and repeated on every base-model upgrade.