Best AI for AdvisorsWhat is fine-tuning?
Short answer
Further training an existing AI model on a specific set of examples so it performs better at a narrow task.
Definition
Fine-tuning takes a general model and trains it a little more on a focused set of examples, such as compliant marketing copy or a particular document layout. The result is a model that's better at that task, without building one from scratch.
It's different from retrieval (RAG), which gives a model documents to read at the moment you ask, without changing the model itself.
Why it matters for advisors
Fine-tuning raises a key data question: whose data was the model trained on? If a vendor fine-tunes on customer data, your clients' information could end up shaping a model other firms use. That's one reason due diligence asks whether a vendor trains on your data.
For most advisory firms, fine-tuning a model yourself isn't necessary. Good prompts and retrieval cover most needs.
It also has data implications. Fine-tuning on client notes or emails puts that information into the model itself, where it's hard to remove. Most advisory firms that want a tool to know their practice use retrieval over approved documents instead.
Example
A compliance-review vendor says its model was fine-tuned on marketing reviews. The CCO asks whether any of the training examples came from customers, and whether the firm's own submissions will be used for future training.
Watch out for
"We fine-tuned on financial data" can mean public data or customer data. Ask which.
Related terms
- Model training on customer dataWhen a vendor uses the data you put into its tool to improve or train its AI models, which may expose client information.
- Foundation modelA large, general-purpose AI model that other products are built on top of, such as the models behind major chat assistants.
- Retrieval-augmented generationAn AI setup that first looks up relevant documents, then writes an answer based on them, so output is grounded in your sources.
Learn more
Last updated .