BrazilFinance and insurance2024 – ongoingNubank
Predictive foundation models trained on transaction sequences
Nubank deployed transaction foundation models in several decision engines
Problem
Nubank's predictive models depended on manually engineered features. Building those features required repeated experimentation for individual tasks.
Approach
Nubank trained transformers on transaction sequences and combined their embeddings with existing tabular features. After acquiring Hyperplane, it integrated foundation-model development into its AI platform. Teams compared challengers with existing models before deploying them in decision engines.
Results
No published figures
Takeaway
Keeping existing inputs and evaluation baselines let teams assess the contribution of a new model architecture.
Categories
- Size
- Large enterprise
- Industry
- Finance and insurance
- Business function
- IT and R&DFinance, risk and legal
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
Nubank explained its transaction-sequence models and their combination with tabular features. It described shared training and fine-tuning infrastructure.
Nubank described integrating foundation models after its 2024 Hyperplane acquisition. It reported deployment in several decision engines following comparisons with existing models.
Full details are in the original sources.
