Nubank uses AI risk acquisition modeling to identify customers more accurately, not to lower underwriting standards

Better risk modeling enable Nu to responsibly expand credit access while maintaining the quality of the portfolio.

Nubank launched with a credit card in Brazil, unlike most digital banks that started with payments or savings and added credit later. “We had to manage credit and manage economic cycles. It was either do that or not survive. It becomes much more difficult to make the leap to a credit-led organization later,” said Jeremy Selesner, General Manager of Credit Card Foundations.

„That choice built 17 generations of limit-increase models, 10 generations of acquisition modeling, more than 100 terabytes of behavioral data, and a culture in which more than 1,000 risk-monitoring artifacts are reviewed every week.” – the company said in a blog post.

The most significant recent development is nuFormer, a transformer-based risk model that delivered a 70% reduction in risk for an equivalent population compared to previous generations. Across model iterations, Nubank’s AI improvements now deliver approximately three times the performance gain typically seen in a single model generation.

The critical distinction: Nubank uses these gains to identify customers more accurately, not to lower underwriting standards. „In Q4’25, this translated into a 50-basis-point increase in Nubank’s credit card purchase volume market share in Brazil — the largest absolute gain by any player in the last decade, per management data.” – the company explained.

And the more powerful models are introduced within Nubank’s unchanged, fundamental approach to risk management. “We continue to expect the future to be worse than the past to maintain that bar of resilience. That’s part of Nubank’s core,” said Tyler Horn, Vice President of Credit Risk. In Q4’25, write-offs remained stable at 2.8% to 2.9% of the current book. 

Nubank’s executives still see meaningful room to deepen customer relationships responsibly. Average revenue per active customer stands at US$15 in Q4’25, compared to ~US$40 for Brazilian incumbent banks, reflecting a customer base still in the earlier stages of credit engagement, and a significant opportunity to grow by serving existing customers better. 

The company’s credit book grew approximately 40% from Q4’24 to Q4’25, with US$11 billion in unused credit limits now available to customers who have already been granted them.