Plaid Rolls Out AI Foundation Models for Credit Underwriting, Fraud Detection and Payment Risk
Plaid Rolls Out AI Foundation Models for Credit Underwriting, Fraud Detection and Payment Risk #
Plaid, the San Francisco-based open finance infrastructure company, announced a set of new artificial intelligence models on 6 October as part of its annual Fall Product Release. The models are designed to support lenders, fraud teams, and payment risk managers.
The main credit product is LendScore 2 (Ls2), an updated version of the company’s cash flow credit risk score. Plaid says Ls2 delivers 42% greater predictive power than traditional credit data alone by combining real-time cash flow signals with behavioural data from across the Plaid Network. Plaid also introduced LendScore Arc, its first transformer-based credit risk model. The transformer architecture is the same class of machine learning design used in large language models; Plaid says it allows the model to detect more subtle patterns in a borrower’s financial history.
Plaid additionally launched specialised LendScore variants for auto, home, and short-term lending markets, including buy now, pay later providers. A new consumer-facing tool called Instant Link lets eligible borrowers share cash flow data with lenders in seconds by connecting their bank accounts to Plaid’s Consumer Reporting Agency.
“Cash flow data tells a more complete story about a borrower,” said Michelle Young, Credit Product Lead at Plaid. “The next generation of LendScore and specialised models close that gap at scale.”
On fraud, Plaid announced a purpose-built AI foundation model trained on data from the Plaid Network. The model is intended to power its existing Protect fraud detection product and draws on behavioural signals across users, devices, and accounts rather than relying solely on individual transaction history.
Plaid also announced improvements to Signal, its payment risk product, with changes designed to help companies assess the risk of ACH transactions more precisely and reduce failed and fraudulent payments.
With connections to more than 12,000 financial institutions and roughly 9,000 applications using its services, Plaid has broad network data to train risk and fraud models at scale. The Fall Product Release is part of the company’s shift from connectivity infrastructure toward what it calls an intelligence layer for financial services. Traditional credit scoring relies on historical repayment data and has been criticised for excluding consumers with thin or no credit files; Plaid positions its cash flow models as a supplement that can identify creditworthy borrowers whose profiles do not appear in bureau data.
Copy link