Plaid’s Sequential Foundation Model reads financial behavior as sequences rather than snapshots, powering LendScore Arc, Signal 4, and Cash Advance Index 2.
Key claims (vendor-reported, early testing):
- LendScore Arc: 20% approval lift for subprime borrowers; transformer-based credit risk with ranked reason codes
- Signal 4: 26% more risky ACH dollars caught at 1% decline rate; ~136ms to score 300 transactions in testing
- Fraud Foundation Model: up to 40% relative fraud detection improvement over previous Trust Index models
The sequential foundation model pretrained on behavioral and transactional sequence data before task-specific fine-tuning. Plaid cites network-scale data across thousands of institutions as a training advantage over siloed fraud detection.