Prompting, RAG, or fine-tuning: which one do you need?
Teams often jump to fine-tuning when a better prompt or a retrieval step would solve the problem faster and more cheaply. A simple order of operations avoids wasted effort.
Start with prompting
Clear instructions, a few good examples, and a defined output format solve a surprising share of problems. It is the fastest and cheapest approach, and it gives you a baseline to measure everything else against.
Add retrieval when knowledge is the gap
If the model needs your policies, products, or documents, give it that information at the moment of the question with retrieval. Your content stays up to date without retraining, and answers can point to their sources.
Fine-tune for behaviour, not facts
Fine-tuning is best for consistent style, structure, or a specialised task where prompting falls short. It is a poor way to teach changing facts, because the knowledge is frozen at training time.
Measure before you escalate
Build a small test set of real examples and score each approach. Move to the next option only when the numbers show the simpler one is not enough. This keeps systems easier to maintain and to explain.