Generative AI for Fraud Detection

Let’s say we’re working as consultants at Gartner, advising a large fintech client on leveraging generative AI to detect and explain financial fraud in their transaction data. The client has a labeled dataset of 50,000 transactions, each with descriptions and fraud flags.

How would you decide between using retrieval-augmented generation (RAG), prompt engineering, or fine-tuning for this fraud detection use case? What factors would guide your choice, and how might your recommendation change if the dataset size doubled?