What is fine-tuning?
Fine-tuning is further training of an existing AI model on a smaller set of examples, so that it performs a particular task or follows a particular style more reliably. It changes the model itself.
Why it matters
A general model knows nothing about how one organisation words its replies or labels its tickets. Fine-tuning can teach it those patterns, which helps with narrow, repeated tasks such as classifying messages or matching a house style. It takes more time and care than editing instructions, so most teams try prompt changes and better source material first and fine-tune only when those are not enough.
How it works
The team collects examples of the input the model will see and the output it should produce, such as past tickets paired with the correct category. The model is trained on these examples, which adjusts its internal settings slightly. The result is a new version of the model. It is tested on examples it has not seen before it replaces the original.
Example
An insurer wants incoming emails sorted into forty claim categories. A general model with written instructions confuses several similar categories. The insurer gathers a few thousand past emails that staff have already labelled and fine-tunes the model on them. On a fresh set of emails, the new version picks the right category more often, and unclear cases still go to a person.
