What is fine-tuning?
Fine-tuning is additional training that changes a model's weights using your examples. It is not the same as uploading a file for the model to read.
In plain English
Uploading your SOP so the assistant can search it is RAG. Fine-tuning bakes patterns into the model. It costs more, it is harder to undo, and a stale example teaches the stale habit. Most shops are not at that step. Ask what you will do when the policy changes next month. A tuned model keeps yesterday's answers until someone pays to tune it again. A file you can edit tonight does not. Collect the corrections anyway. They are useful as examples in the prompt long before they are useful as a training set. Keep the before-and-after emails in a folder. They are the evidence if someone later asks why you skipped the custom model.
How an owner would use it
Before you pay anyone to fine-tune, write the prompt, clean the SOP, and ground answers in current files. Fine-tune only if a specialist can show a repeated failure that files and instructions do not fix.
Common mistakes
Fine-tuning on old emails teaches old habits, including the discount a former manager used to give away. Names like LoRA are techniques inside that work, not a product an owner needs to shop. Do not fine-tune to change a price or a policy. Change the document. Do not fine-tune to get a friendlier voice if a system prompt and ten examples would do it. Fine-tuning is hard to undo.
A worked example
A bridal shop
The return policy was the real fix
This is an illustration, not a customer result. A vendor offered to fine-tune a model on old alteration emails. The owner instead saved eight real replies and pointed the assistant at the current alteration policy. The drafts matched the shop. The fine-tune quote stayed in the drawer.
Related terms
Prompts and guides
Questions owners ask
What is fine-tuning?
Fine-tuning trains a model further on your examples so its defaults change. It is different from giving the model documents to look up.
How do I use fine-tuning in a small business?
Usually you do not, yet. Fix the prompt, the SOP, and the documents the model retrieves. Consider fine-tuning only when those still fail the same way.
Is uploading a PDF fine-tuning?
No. Uploading a file for the assistant to search is retrieval. Fine-tuning changes the model itself with a training set. If you can delete the file and the answer should change, you wanted retrieval.
Product guides are on AI apps for business.