Choose retrieval for changing reference knowledge and fine-tuning for consistent task behaviour—not as interchangeable shortcuts.
- retrieval augmented generation
- RAG
- LLM fine tuning
- generative AI
Use retrieval for grounded, changing information
Retrieval-augmented generation supplies selected source material at answer time. It can make knowledge easier to update and lets a product cite relevant records when the source pipeline is designed well.
AI & AutomationAI & Automation
Thoughtful decisions compound over time.
Practical product work brings technical choices back to the people and workflows they are meant to serve.
Fine-tune for repeatable patterns
Fine-tuning may help a model follow a specialised format or task style, but it is not a dependable way to keep fast-changing business facts current. Evaluate data quality, privacy, and maintenance effort.
Compare with a representative evaluation set
Test realistic questions, missing information, and incorrect-source cases using the same success criteria. A small proof of concept can reveal whether retrieval, fine-tuning, or a simpler search experience is appropriate.
Practical application
Try retrieval when answers must reflect changing policy documents: index approved sources, test citation accuracy, and verify permissions. Consider fine-tuning only when a stable task pattern remains inconsistent after prompt and workflow improvements, and measure against the same evaluation set.
