Out of the box, large language models know a lot about the world and nothing about your business. There are two main ways to fix that: retrieval-augmented generation (RAG) and fine-tuning.
RAG: give the model the right pages
With RAG, your documents are split into chunks, converted into embeddings and stored in a vector database. When a user asks a question, the system retrieves the most relevant chunks and passes them to the model along with the question. The model answers using that context, and can cite its sources.
Best for: policies, manuals, product catalogues, tickets, anything that changes often or needs citations and access control.
Fine-tuning: change how the model behaves
Fine-tuning trains the model further on examples of the inputs and outputs you want. It is excellent for teaching style, format and specialised tasks, such as classifying tickets into your categories or writing in your brand voice.
Best for: consistent output formats, domain-specific classification, tone, and reducing prompt length at scale.
The practical answer: usually RAG first
| Need | Choose |
|---|---|
| Up-to-date facts with citations | RAG |
| Specific output format or tone | Fine-tuning |
| Both knowledge and behaviour | RAG + light fine-tuning |
For most business use cases, a well-built RAG system with good chunking, hybrid search and re-ranking delivers most of the value. Fine-tuning comes later, once you have real usage data showing where behaviour needs to improve.
Want help choosing? Talk to our AI team.