Industry Applicationscredit underwritingnubanktransformer modelsrisk modeling

Nubank Expands Credit via AI Underwriting

|May 12, 2026

6.8

Relevance Score

Nubank Expands Credit via AI Underwriting — Photo: crowdfundinsider.com · rights & takedowns

CrowdfundInsider reports that Nubank executives described how an AI-driven underwriting framework is widening credit access while keeping portfolio metrics stable. Per CrowdfundInsider, the company detailed a transformer-based model called nuFormer that has produced a 70 percent drop in projected risk for comparable customer segments relative to earlier versions. The report says the lender has iterated its credit models extensively-17 iterations for credit-limit adjustments and 10 iterations for customer acquisition-and that its data repository exceeds 100 terabytes. CrowdfundInsider also reports a dedicated team monitors more than 1,000 monitoring indicators weekly, and that the precision gains contributed to a 50-basis-point rise in the firm's share of credit-card volume in Brazil in Q4 2025. Executives quoted in the report emphasised continued risk discipline: "We had to manage credit and manage economic cycles," Jeremy Selesner said, and Tyler Horn said, "We continue to expect the future to be worse than the past to maintain that bar of resilience," per CrowdfundInsider.

What happened

CrowdfundInsider reports that Nubank senior executives briefed on how the bank's AI underwriting framework is enabling broader credit access while keeping portfolio stability intact. Per CrowdfundInsider, the team named a transformer-based model nuFormer and said it achieved a 70 percent drop in projected risk for comparable customer segments versus earlier model versions. The article reports 17 iterations for credit-limit models, 10 iterations for customer-acquisition models, a data repository exceeding 100 terabytes, and a team monitoring over 1,000 monitoring indicators weekly. CrowdfundInsider attributes a 50-basis-point rise in Nubank's share of credit-card volume in Brazil in Q4 2025 to the improved precision, and carries direct quotes from executives: "We had to manage credit and manage economic cycles," Jeremy Selesner; "We continue to expect the future to be worse than the past to maintain that bar of resilience," Tyler Horn.

Editorial analysis - technical context

Transformer-based scoring models such as nuFormer align with a broader industry shift toward sequence-aware architectures and richer behavioral embeddings for credit risk. Companies adopting similar architectures typically combine large customer-event datasets with frequent model retraining and dense monitoring to capture drift, segmentation shifts, and macroeconomic regime changes.

Industry context

Observed patterns in comparable fintech deployments show a tradeoff between expanded coverage and regulatory scrutiny on fairness and explainability. For practitioners, robust monitoring (hundreds to thousands of indicators) and conservative stress assumptions are common mitigations when rolling models into consumer-facing underwriting.

What to watch

Indicators worth tracking include published delinquency and write-off metrics in subsequent reports, regulatory inquiries or disclosures about model governance, and whether other lenders report similar lift from transformer-based credit models. CrowdfundInsider's piece contains the reported metrics and executive quotes; Nubank has not been quoted beyond those excerpts in the article.

Scoring Rationale

This is a notable fintech application showing material risk reduction from a transformer-based credit model, useful for practitioners evaluating production ML in financial services. It is not a frontier model release, so its impact is significant but not industry-shaping.