What did Plaid just launch?
Plaid unveiled a new generation of AI models for lending, fraud detection, and payments as part of its annual Fall Product Release. The headline claim: its new LendScore 2 credit risk model is 42% stronger at predicting whether a borrower will repay than traditional credit data alone. Alongside it, Plaid introduced LendScore Arc — its first transformer-based credit score — a new AI foundation model for fraud detection, and faster ways for lenders to get cash-flow data.
The details
Plaid, the San Francisco fintech that connects bank accounts to apps, is building its release around three areas:
Lending. Instant Link lets borrowers who connect their financial accounts share cash-flow insights with lenders in under two seconds, cutting friction during applications. LendScore 2 is the new core underwriting model, with specialized versions for specific industries: the auto-lending variant lowered delinquency by 26% among deep-subprime applicants at the same approval rate, while the home-lending variant approved 6.3% more borrowers at the same risk level, according to Plaid. LendScore Arc, the transformer-based model, reads the order and timing of a borrower’s transactions; Plaid calls it its best-performing credit model yet, with early testing showing a 20% predictive lift on deep-subprime borrowers and 24% on superprime.
Fraud. A new AI foundation model trained on patterns and hundreds of millions of data points from Plaid’s network now powers Plaid Protect, the company’s fraud detection product. Instead of analyzing a snapshot of activity, it examines the full sequence of events — which is how fraud often shows up, as a series of routine-looking actions. In internal evaluations, it delivered up to 40% relative improvement over previous baselines.
Payments. Plaid’s sequential foundation model now powers Signal, its ACH payment risk model, reading an account’s transaction history in sequence to predict payment risk. In testing, it helped Signal prevent 26% more ACH returns without increasing false flags. Guaranteed Payments uses that sharper risk read to give businesses more flexible options — including delayed release and partial guarantees — instead of a simple approve-or-decline.
Why it matters
The real story here isn’t any single model — it’s Plaid’s data advantage. With thousands of financial institutions and millions of consumers on its network, Plaid can train sequence-based AI models on real financial behavior at a scale few competitors can match. Cash-flow underwriting has been the industry’s promised land for years, but the bottleneck was always turning that story into something a lender could act on in real time. If Plaid’s claimed numbers hold up in the wild — and note that these are the company’s own internal figures — transformer-based credit scoring could quietly push millions of borrowers who look thin or subprime on traditional credit files into more affordable lending tiers.
FAQ
What is LendScore 2?
Plaid’s new core AI credit risk model, unveiled in its Fall Product Release. Plaid says it is 42% stronger at predicting a borrower’s ability to repay than traditional credit data alone.
What makes LendScore Arc different?
Arc is Plaid’s first transformer-based credit risk score. It learns from the order and timing of a borrower’s transactions rather than static data points.
How is AI being used for fraud detection in this release?
A new AI foundation model trained on Plaid’s network data now powers Plaid Protect. It analyzes the full sequence of account events instead of a snapshot, delivering up to 40% relative improvement over previous baselines in internal testing.
Which companies use Plaid?
More than 9,000 companies rely on Plaid, and it connects to over 12,000 financial institutions across the US, Canada, the UK, and Europe, according to the company.
Sources: Business Wire