Definition
Sequential foundation models read financial behavior as ordered event sequences rather than static snapshots. Pretrained broadly on behavioral and transactional data, then fine-tuned for domain tasks like credit risk, fraud, and payment return prediction.
Key Points
- 2026-10-06: plaid deploys sequential foundation models across LendScore Arc, Signal 4, Cash Advance Index 2, and Fraud Foundation Model (2026-10-07-plaid-foundation-models-intelligence)
- Transformer attention weighs events across time; pretrained before task-specific labeling cycles
- Signal 4: ~136ms to score 300 transactions in testing; 26% more risky ACH dollars caught at 1% decline rate (vendor-reported)
Related
- transformer-architecture
- foundation-models
- credit-scoring
- fraud-detection
- ach-payment-risk
- plaid
- fintech-ai