Plaid’s AI Models Enhance Financial Risk Assessment and Lending
Two borrowers can share identical incomes, account balances, and overdraft histories but still pose different credit risks. Traditional models often fail to distinguish between them. Plaid believes foundation models can bridge this gap.
This year, the company introduced a two-layer foundation model architecture for financial data. The transaction model interprets individual financial events, while the sequential model analyzes their order, timing, and relationships. This approach mirrors how large language models understand words in context rather than isolation.
A Richer Understanding of Financial Behavior
This deeper understanding is enhancing lending, fraud detection, and payment risk across Plaid’s network of 12,000 financial institutions and 9,000 apps. The transaction model improved income classification accuracy by 48%, while the sequential model reduced credit default risk by 13.6% without lowering approval rates.
This foundation also powers products like LendScore, Plaid’s real-time cash flow-based credit risk score. LendScore reduced lending risk by 41% compared to traditional benchmarks, maintaining approval rates. This innovation earned Plaid the Data Innovation Award at Tearsheet’s AI Innovation Awards 2026.
From Connectivity to Intelligence
Tearsheet spoke with Suddu Seshadri, Plaid’s Head of Data & AI, about the company’s evolution. Seshadri explained that Plaid initially focused on data connectivity, enabling access to financial data across thousands of institutions and apps. As the digital financial ecosystem matured, the focus shifted to intelligent finance, where AI analyzes financial data to answer critical questions about fraud, payment risk, and credit.
“From our connectivity, we’ve built dimensionality around identity, connections, and transactions,” Seshadri said. “This helped build an incredibly deep network and helped us answer critical questions around fraud, payment risk, and credit. The compounding effects of this network are fueling our next evolution.”
Why Foundation Models Matter
Seshadri highlighted the growing consumer demand for self-driving money. A 2026 Harris Poll revealed that 86% of U.S. adults already use AI to manage their finances, and they want AI to do more. Traditional models, while effective for specific tasks, often lack a reusable understanding of transactions, income, cash flow, and behavior.
Foundation models address this by learning a reusable representation of financial activity, adapting it across tasks. This is particularly valuable given the challenges of financial data: cryptic transaction descriptions, limited labeled data, and timing-dependent event meanings.
Plaid’s Unique Approach
“We are not asking a general-purpose model to guess from raw bank text,” Seshadri explained. “Our transaction model learns the economic meaning of individual events, while our sequential model learns order, timing, and changes in behavior over time.”
This shared understanding improves product-specific systems. For example, the transaction foundation model enhanced income classification by 48% and loan payment detection by 14%.
A Shared Foundation for Diverse Products
Lending, fraud detection, and payments rely on common concepts: income, recurring obligations, cash-flow stability, account behavior, and changes over time. Plaid’s shared foundation turns these concepts into reusable representations, which products then combine with their own data, labels, thresholds, and controls.
“That means an improvement in the underlying representation can benefit several products, without treating a lending decision, a fraud decision, and a payments decision as the same problem,” Seshadri noted.
Plaid’s AI Innovation Award and Future Vision
The conversation focused on Plaid’s shift from data connectivity to building foundation models for finance, driven by the need for more intelligent financial solutions.
Redefining “Good Financial Data”
While the fundamentals of good financial data remain—permissioned, accurate, current, reliable, and traceable—AI has shifted the focus to contextual value. A transaction gains meaning when understood alongside prior activity, timing, relationships, and behavioral changes.
“Sequence and relationships were always important,” Seshadri said. “Modern models make it possible to use them at a much greater scale.”
