The full answer
The two solve different problems. RAG addresses knowledge: what the model knows about your products, policies and records. Fine-tuning addresses behaviour: how the model writes, which format it follows and how it handles domain-specific patterns. Many teams ask for fine-tuning when their real need is knowledge, which RAG handles better and more cheaply.
RAG is usually the first step because it needs no training data beyond your existing documents, updates immediately when content changes, can cite its sources and respects access permissions at retrieval time. It also works with any model, so provider choice stays open.
Fine-tuning makes sense when there is a large set of high-quality examples, the task is narrow and repeatable, and a smaller fine-tuned model would cut cost or latency compared with prompting a large one. Creuto does this with open-weight models hosted in your environment, or through provider fine-tuning services where the data terms are acceptable.
The decision is made with evidence. Creuto builds an evaluation set from your real cases, measures a prompted baseline, then RAG, then fine-tuning if justified, and shows you the difference in accuracy and cost before you commit.
In some products the answer is both: retrieval for knowledge and a fine-tuned model for tone and format. Discovery is where that gets worked out.