Plaid unveiled Plaid LendScore 2, a new cash flow credit risk model, on 6 October 2026 in its yearly Fall Product Release. Plaid says Plaid LendScore 2 is 42% better at predicting a borrower’s ability to repay than traditional credit data alone. The release also introduced Instant Link for reusing consumer reports, LendScore Arc, a transformer-based credit model, and new AI foundation models for fraud detection and ACH payment risk.

What Plaid LendScore 2 and the new models do

Plaid LendScore 2 is the new core model in Plaid’s LendScore family, and two specialized versions come with it. According to Plaid, Ls2 Auto cut delinquency among deep-subprime applicants by 26% while holding the approval rate steady, and Ls2 Home Lending approved 6.3% more borrowers without taking on extra risk. The family also includes a model built for short-term lending.

LendScore Arc is a different kind of model. It is built on Plaid’s sequential foundation model and learns from when, and in what sequence, a borrower’s transactions happen, rather than just from totals and averages. In early testing, Plaid says Arc showed a 20% lift for deep-subprime borrowers and a 24% lift for superprime borrowers over its core model. Plaid calls Arc its best-performing credit model so far and pitches it at lenders that are ready to run transformer-based credit models.

The same approach now runs in fraud and payments. A new fraud foundation model powers Plaid Protect and feeds its Trust Index scoring, with what Plaid describes as up to a 40% relative improvement over previous baselines in internal evaluations. Plaid says Signal, its ACH payment risk model, used the sequential model to prevent 26% more ACH returns without more false flags. Guaranteed Payments now offers delayed release and partial guarantees, so a business is no longer limited to approving or declining a payment.

Fintechbits analysis

Why the 42% figure for Plaid LendScore 2 is the weakest number here

The 42% headline is measured against “traditional credit data alone.” That is an easy benchmark. Any decent cash flow model should beat a credit file on its own for thin-file applicants, because those applicants have little in their credit file to begin with. Lenders want to know how Plaid LendScore 2 performs when added to a bureau score they already use, and against other cash flow scores. The release answers neither question.

The specialized results are more useful. A 26% cut in delinquency among deep-subprime auto applicants at a fixed approval rate is the kind of figure a credit committee can price. The 6.3% increase in home lending approvals at constant risk is smaller, but mortgage is where cash flow data has had the hardest time, which Plaid itself acknowledges when it calls auto and home lending markets that have historically relied almost entirely on traditional credit data. All of these figures are Plaid’s own testing, and the Arc numbers come from “early testing,” so treat them as claims to verify in a pilot.

Why Instant Link matters more than Plaid LendScore 2

A borrower who has linked accounts through Plaid Consumer Reporting Agency can agree to pass the same cash flow insights to lenders on later credit applications, and lenders receive them in under two seconds. That makes Plaid’s consumer report reusable, like a credit file. The more people who have a Plaid report on file, the less friction there is for the next lender, and the harder it becomes for a lender to choose a different vendor.

That puts Plaid in direct competition with the bureaus. On the same day, Experian launched the Experian Cashflow Data Bureau, a separate FCRA-regulated agency for bank-account data, according to Experian’s release. Experian has the lender relationships and the credit data, while Plaid has the account connections: it says it connects to more than 12,000 financial institutions and serves more than 9,000 companies. Both want to be the one regulated source of cash flow data a lender uses. Plaid LendScore 2 gives lenders a reason to try Plaid, and Instant Link gives them a reason not to switch away later.

Where Plaid LendScore 2 sits in the transformer trend

Plaid is not the first lender or vendor to use transformer architectures on transaction sequences. Affirm put a transformer model behind its checkout underwriting, as Fintechbits reported. The approach is spreading because transaction histories are sequences, and order matters: a paycheck followed by rent is a different signal from rent followed by a payday loan.

Explainability is the harder question. US lenders have to give applicants specific reasons for an adverse action, and those reasons are easier to produce from a gradient-boosted model built on named attributes than from a transformer reading raw sequences. Plaid’s release does not say how LendScore Arc generates reason codes. Regulated lenders will ask about that before they think about lift. Readers new to the plumbing underneath all of this can start with Fintechbits’ explainer on open banking.

What to Watch Next on Plaid LendScore 2

Start with named lenders. If Plaid names an auto lender or mortgage originator using Plaid LendScore 2 in production, the 26% and 6.3% figures become far more credible. Instant Link adoption matters too, since it only works once many consumers already have a Plaid report on file, and so does whether LendScore Arc gets past early testing with a clear way to explain its decisions. If none of these happen by mid-2027, Experian’s pitch that lenders want one bureau for credit and cash flow will start to look like the safer bet.

Questions and answers

What is Plaid LendScore 2?

Plaid LendScore 2 is Plaid’s core cash flow credit risk model, launched on 6 October 2026. Plaid says it is 42% better at predicting a borrower’s ability to repay than traditional credit data alone.

What is Plaid Instant Link?

Instant Link lets borrowers who have connected accounts to Plaid Consumer Reporting Agency reuse their cash flow insights on later credit applications. Plaid says lenders receive them in under two seconds.

What is Plaid LendScore Arc?

LendScore Arc is Plaid’s transformer-based credit model, which reads the sequence and timing of transactions. In early testing, Plaid says it showed a 20% lift for deep-subprime borrowers and a 24% lift for superprime borrowers over its core model.