3 Working with generative AI in finance
This chapter covers
- Understanding the paradigm shift from model building to AI engineering
- Architecting a secure and compliant generative AI system for finance
- Making pragmatic model selections beyond generic benchmarks
- Mastering core techniques like RAG, PEFT, and guarded AI agents
- Developing deployment strategies for managing cost, risk, and governance
Finance has always gravitated toward data-driven innovation, yet the debut of generative AI marks a decisive turning point. Until recently, most machine learning in banks and asset management firms focused on structured data analytics, like credit scoring. Then came the proliferation of Large Language Models (LLMs)—tools capable of generating human-like text, auto-writing code, or summarizing dense regulatory documents in seconds. This leap has triggered something of a gold rush in financial services, with institutions like Morgan Stanley embedding chat-driven experiences into their advisory platforms.
The implications go far beyond automating a few back-office forms. It is precisely these conversation-centric, generative capabilities that prompt both excitement and profound caution. Navigating these opportunities and perils demands a new lens on AI—one that addresses not just the technology itself, but the organizational guardrails necessary to keep it aligned with business strategies and regulatory mandates.