Generative artificial intelligence (GenAI) produces human-like text and code and can summarize dense regulatory documents. In finance, it extends data-driven work beyond structured analytics, while requiring guardrails aligned with business strategy and regulation. section 3.0
GenAI can translate outputs from deterministic models into clearer explanations, or help staff retrieve and interpret internal guidance. It supports these workflows; it does not replace the underlying model’s decision or score. section 6.5
Explaining a credit score
A constrained language model receives a SHAP value and customer profile, then explains that balances exceeding 80% of a credit limit negatively affected the score. An underwriting agent can also query applicant PDPs and internal credit guidelines through RAG to suggest conditional approvals. section 6.5
Keep the LLM from making credit decisions or calculating risk scores: deterministic models and their interpretability methods remain the source of truth. Unconstrained reasoning risks fabricated rejection reasons and fair-lending violations. section 6.5
