If the task is well within a capable general model's ability and the main problem is inconsistent output format or occasional misunderstanding, more specific instructions, examples, and structured output usually close that gap faster and cheaper than fine-tuning.
Fine-tuning earns its cost when you need a smaller, cheaper model to match a larger model's quality on a narrow, well-defined task at high volume — the economics flip once per-call cost at scale outweighs the fine-tuning investment.
If the underlying issue is actually an ambiguous or poorly-specified prompt, fine-tuning on examples that encode that same ambiguity just bakes the problem in more permanently — worth ruling out a prompting fix first, every time.