AI and agents
Fine-tuning
Fine-tuning is further training of an existing language model on your own examples so it follows a particular format, style or task more reliably.
How it works in practice
Fine-tuning changes how a model behaves. It is good at teaching a consistent output shape, a house tone or a narrow classification task from hundreds or thousands of worked examples. It is poor at teaching facts that change, because the knowledge is baked in at training time and goes stale.
For most business questions, retrieval (RAG) and a well-written prompt get further for less money than fine-tuning. We reach for fine-tuning when a task runs at high volume and a smaller tuned model can match a larger general one at lower cost, or when prompting alone cannot hold a strict format. It needs a clean, labelled dataset and an evaluation set to prove it helped.
A worked example
A logistics company classifies 5,000 delivery exception notes a day into 12 codes. A small model fine-tuned on 3,000 labelled notes matches the accuracy of a larger model on their test set at a lower cost per note.
Examples in this glossary describe typical builds for companies of 1 to 1,000 people. They are not named client results.















