Tearsheet Podcast: Exploring Financial Services Together How Upstart's AI is mastering growth, credit performance, and profitability
Oct 22, 2025
In this insightful conversation, Paul Gu, Co-founder and CTO of Upstart, shares his journey from a Yale dropout to leading AI-driven innovation in consumer lending. He discusses how Upstart's advanced AI techniques have redefined credit risk assessment, enabling growth, credit performance, and profitability simultaneously. Paul also outlines the company’s ambitious 2025 roadmap, focusing on achieving a 10x AI advantage and expanding services to cover the entire credit lifecycle, all while navigating the complexities of the evolving financial landscape.
AI Snips
Chapters
Transcript
Episode notes
Thiel Fellowship Sparked Upstart Pivot
- Paul Gu dropped out of college to join the inaugural Thiel Fellowship and pursue entrepreneurship.
- He and co-founder Dave pivoted from income-share agreements to clear-fixed repayment loans after learning customer preferences.
Three Pillars Of Model Improvement
- Upstart's model progress depends on more rows, more columns, and better model architectures.
- Growing training data from zero to ~91 million unlocked tree and neural network ensembles beyond linear models.
Quantifying Model Separation
- Upstart measures differentiation via a model separation accuracy metric versus textbook credit models.
- Their model is 171% better at separation, yet 87.5% of prediction error still remains to be solved.