XAI in Practice: Architecting Compliant "Black Box" Models for Credit Decisioning

Traditional financial models are trained on data generated by human actors, yet behavioral finance has proven this data is saturated with cognitive biases like loss aversion, herding, and confirmation bias. When complex "black box" models (e.g., XGBoost, neural networks) are trained on this data, they don't just learn market alpha; they risk becoming high-speed "bias amplifiers," automating and scaling our worst irrational instincts. This presentation bridges behavioral finance and engineering, proposing an architectural pattern where Explainable AI (XAI) is deployed as a real-time "rationality audit" service. We will demonstrate how to use XAI techniques like SHAP to: Detect and mitigate hidden biases in credit lending models that traditional accuracy metrics miss (e.g., zip code as a proxy for affinity bias). The session will also cover the "last mile" problem: using these rational insights to communicate effectively with inherently irrational customers, framing denials as opportunities.
Who is this presentation for?
Fintech Implementation Leaders
Prerequisite knowledge:
AI, Finance
What you'll learn?