Model fine-tuning is further training of a model on your own examples so that its outputs match a required style, format or specific task. In business practice it is the least necessary step, even though it is the one most often requested. In most cases, a better prompt or giving the model access to company documents when answering is enough, and this is cheaper, faster and easier to update. Fine-tuning makes sense when a fixed output format, specific specialist language, or a task that cannot be explained in a prompt is needed. It requires a sufficient number of good-quality examples, training costs, and repetition with every model change. Fine-tuning does not add up-to-date information – document access serves that purpose.
See also: RAG over a company's own content, Prompt writing as a skill, Training data management.
