FinCrunch
← All coverage

Plaid launches Instant Link and a new LendScore 2 model family for cash-flow underwriting

Instant Link lets returning borrowers share permissioned cash-flow data with lenders in under two seconds without reconnecting their bank accounts, while LendScore 2 adds a transformer-based model built on transaction sequences.

Priya RaghunathanFintech Desk
High-contrast black and white photograph of a laptop screen displaying an abstract financial data interface with flowing bank-account connection lines and credit score graphs

Plaid, the financial data network that connects bank accounts to lending and fintech apps, launched two new products Tuesday: Instant Link, a faster reconnection flow for returning borrowers, and LendScore 2, an expanded family of cash-flow underwriting models.

Instant Link is built for borrowers who have previously connected a bank account through Plaid and need to share updated cash-flow data with a lender again later. Rather than re-authenticating the full account-linking flow, the product lets a returning, permissioned user share current cash-flow insights with a lender in under two seconds, according to the company.

The LendScore 2 family centers on a core model Plaid calls Ls2, which the company says offers 42% greater predictive power than traditional credit-bureau data in its internal testing, along with specialized versions tuned for auto, home and short-term lending. Plaid also introduced LendScore Arc, a transformer-based model — the same general architecture used in large language models — that analyzes the sequence of a borrower's transactions over time rather than treating cash-flow data as a static snapshot.

Cash-flow underwriting, which scores borrowers using actual bank-transaction history instead of relying solely on bureau tradelines, has gained traction as lenders look to extend credit to borrowers who are thin-file or credit-invisible but have steady, verifiable income and spending patterns. Plaid's launch adds a transformer-based entrant to a field that has mostly used simpler statistical and gradient-boosted models to date.

Plaid did not disclose which lenders are using the new models at launch or publish third-party validation of the 42% predictive-power figure beyond its own internal comparison.