Fine-tuning can help when a model must produce a very consistent output, such as a fixed classification scheme or a house writing style, and prompting alone is not reliable enough. It can also let a smaller, cheaper model handle a narrow task well.
For example, a team might fine-tune a model on past support tickets labeled by category to improve routing. The common misconception is that fine-tuning is how you teach a model your company's knowledge. For facts that change, such as policies, prices or product details, retrieval-augmented generation is usually easier to maintain, because updating a document is simpler than retraining a model.