Fintech Takes x FairPlay Presents Model Citizens Ep 2: Credit Risk Analytics in the Age of AI

Jul 29, 2025

Guest

Andrada Pacheco

Guest

Kevin Moss

Guest

Kareem Saleh

Kareem Saleh, founder of FairPlay, teams up with Kevin Moss, a former Chief Risk Officer, and Andrada Pacheco, Chief Data Scientist at VantageScore. They delve into the evolving landscape of credit risk analytics powered by AI. The trio discusses the balance between enhancing predictive accuracy and maintaining transparency in credit models. They highlight the importance of using traditional and AI methods, while stressing that existing language models aren't ready for credit decisions yet. Fairness remains pivotal as regulatory scrutiny grows.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

ANECDOTE

Kevin's AI Credit Modeling Journey

Kevin Moss shared how AI techniques like CART and gradient boosting have been used in credit risk modeling for over 30 years.
He gave examples of using transaction and cash flow data to improve credit risk models and achieve significant predictive lift.

INSIGHT

Explainability Challenges in Credit AI

Explainability remains a major challenge with sophisticated machine learning models in credit scoring.
Andrada Pacheco emphasized leveraging AI to generate interpretable features, blending machine learning and explainable frameworks for regulatory compliance.

INSIGHT

LLMs Are More Than Predictors

Large language models (LLMs) differ from classical machine learning in that they can generate new, human-like text content instead of just predicting outcomes.
They offer storytelling and explanation capabilities but pose challenges in credit decisions due to lack of transparency and complexity.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

Welcome back to Model Citizens: AI Compliance for Banks and Fintech Lenders, a six-part miniseries from the Fintech Takes podcast in partnership with FairPlay.

With FairPlay’s Kareem Saleh (Founder & CEO) at my side, we unpack: how can banks and fintechs build fair, compliant lending systems in a time of regulatory uncertainty?

In Episode 2, we’re joined by Kevin Moss (Senior Advisor at Baselayer, former CRO) and Andrada Pacheco (EVP & Chief Data Scientist at VantageScore) to explore how AI is reshaping credit modeling, and where caution is still very much required.

We trace the evolution from decision trees and logistic regression to gradient boosting, cash flow data, and the emergence of AI. Along the way, we tackle the core dilemma: how to boost predictive power without losing explainability or fairness.

Highlights include:

Why better performance often comes at the cost of transparency (and how to bridge the gap with hybrid models)

AI isn’t new in credit risk modeling; tree-based methods like CART, CHAID, and gradient boosting have been around for decades (what’s changing now is the scale, the data, and the complexity)

LLMs are great for fraud, operations, and consumer education, but they’re not safe for credit decisions just yet)

Fairness is expanding: Disparate impact enforcement may be fading federally, but state attorney generals and plaintiffs’ attorneys are picking up the slack.

We close with a reminder: credit modeling doesn’t just need to be effective. It needs to be explainable, equitable, and defensible; especially as AI raises the stakes.

Don’t forget to subscribe and catch more insights on Model Citizens in upcoming episodes!

This miniseries is brought to you by FairPlay.

FairPlay is an AI enablement company for financial services. They help companies build, test, optimize, validate and govern AI models. Learn more at Fairplay.ai

Sign up for Alex’s Fintech Takes newsletter for the latest insightful analysis on fintech trends, along with a heaping pile of pop culture references and copious footnotes. Every Monday and Thursday: https://workweek.com/brand/fintech-takes/

And for more exclusive insider content, don’t forget to check out my YouTube page.