Plaid CEO Zachary Perret on LendScore — a real-time credit score built on cash flow and network data

Summary

Plaid's fall product release is centered on LendScore, a real-time credit score built on cash flow, income, and network-wide data that Zachary Perret describes as a fundamental upgrade to how lenders assess borrower quality.

The problem with existing scores

Perret's case against traditional credit scores is straightforward: they are backward-looking, slow to update, and ignore large amounts of financially relevant data. A seven-year payment history can't capture someone who just changed jobs and doubled their income. It also can't flag someone who applied for seven loans in the past 24 hours. FICO and VantageScore can do neither of those things in real time. Perret himself was hit by a credit score drop because an optometrist sent a bill to an old address and it went unpaid for years before he noticed.

How LendScore works

LendScore pulls real-time income and expense data to calculate free cash flow, which Perret argues is a more accurate measure of repayment capacity than historical payment records. On top of that, Plaid adds what Perret calls network insights — signals only Plaid can see because it sits across all of its lender customers simultaneously. Loan application velocity is one example: Plaid can detect if the same consumer is applying across multiple lenders within a short window, a pattern traditional bureaus would miss entirely.

The claimed outcome is a roughly 20% lift in the quality of lending decisions, alongside broader credit access for consumers whose real financial strength isn't captured by legacy scores — someone with a high income and a strong rent payment history, for instance, who might otherwise be thin-file.

Go-to-market

Plaid is positioning LendScore as additive rather than disruptive at launch. Lenders can run it alongside FICO or VantageScore rather than replacing them. Perret's expectation is that once lenders see the incremental signal quality, adoption deepens over time. Sophisticated lenders like BNPL companies are likely to take the underlying data inputs and build their own models on top, rather than relying on the LendScore output directly.

The business model mirrors how credit scores are typically sold — per loan application, with lenders as the paying customer. One meaningful difference from FICO is that LendScore is designed as an ongoing score, not a one-time snapshot, which opens up use cases like proactive credit line offers or preferential payment terms as a borrower's financial health improves.

On the consumer side, Perret says Plaid wants consumers to be able to view their credit profile for free, with product launches on that front expected over the coming quarters.

Protect and fraud detection

The second major release is Protect, an expanded fraud suite. It aggregates behavioral signals from across Plaid's network — device fingerprints, phone numbers, IP addresses, app sign-up velocity — into a risk score that lets lenders and fintechs create tiered onboarding flows. Known users get a fast lane: recognized device, known phone number, two-factor and done. Unknown or anomalous users get a slow lane with additional verification steps. Perret frames the goal as letting fintechs sign up more users faster without absorbing more fraud risk.

AI at Plaid

Perret describes Plaid's AI usage as "medium" on token consumption. The primary application is transaction cleaning and categorization — running LLMs over raw transaction text to structure it for downstream models. There is no consumer-facing chat product. Much of Plaid's AI investment is, in Perret's framing, defensive: using AI to detect and counter AI-generated fraud. Categorized insights, like flagging elevated bank fees, are surfaced to Plaid's customers rather than directly to end consumers, who then see it through their budgeting or banking app.