Plaid
Plaid connects financial accounts to lenders and fintechs. Plaid Check, its consumer reporting subsidiary, turns permissioned bank account activity into reports and credit risk inputs. Consumer Report offers cash flow and network insights, income analysis, and a home lending asset report. The original LendScore ranks the likelihood of repayment on a 1 to 99 scale and returns the five factors that most reduced the score. Lenders can combine it with a traditional credit report or use underlying cash flow measures such as volatility and loan payments. Instant Link can retrieve previously permissioned insights for eligible returning applicants without asking them to reconnect their bank accounts.
The broader Plaid Network connects to more than 12,000 financial institutions. Personify uses Plaid transaction data with Prism Data in underwriting more than 90% of its applicant traffic and reported up to 8% more approvals in underserved segments. Fragile reported 26% lower delinquency among eligible users when it used LendScore and Instant Link attributes while maintaining approvals. Mortgage lenders can share eligible asset reports with Fannie Mae and Freddie Mac. Plaid Check also lets consumers obtain or dispute their report directly. The lender applies its own credit policy and makes the decision.
Plaid sells fraud and payment risk products on the same network. Plaid Protect scores fraud risk in real time at onboarding, login, and account linking, drawing on more than 10,000 signals from 500 million linked accounts, device intelligence, and consortium reports shared among Plaid customers. Fraud teams can query patterns in plain language, turn findings into live rules, and see whether a person or an AI agent is driving a session. Signal Transaction Scores estimates the chance that a US ACH payment will be returned, weighing more than 1,000 risk factors, and businesses set approval thresholds in a no code rules dashboard. Signal is billed as a flat fee per evaluation.
Capability Axes
Capability grades
15 of 15 axes rated · 13 graded A or B
LendScore version 1 uses gradient boosted decision trees to turn cash flow and network behavior into a credit risk score. The model combines bank transaction information with signals such as connections to services that may fall outside a bureau report. It estimates the risk of becoming at least 90 days past due within 12 months. Plaid also links financial accounts and delivers underlying records and asset reports. Those services remain useful when a lender uses its own credit model or buys scores from another provider.
An underwriting report delivers a score and reason codes after Plaid Check establishes which accounts the lender can access. Lenders apply their own credit policy and can use the score alongside bureau data or other inputs. The report flow can return an error when a score cannot be computed. Each version 1 score includes the five factors that most pulled it down, which a credit team can match to its adverse action process. Plaid’s consumer reporting subsidiary provides the inputs and the lender retains decision authority.
Version 1 uses Shapley additive explanations to measure the contribution of individual features. The five largest negative contributors become codes suitable for mapping to a lender’s adverse action process under the Equal Credit Opportunity Act. Plaid assesses model separation using the Kolmogorov Smirnov statistic and reports results for loans and revolving credit. Model inputs have directional constraints, so for example regular inflows cannot move estimated risk the wrong way. The reason codes show which inputs influenced a given score.
Fragile reports that using LendScore and Instant Link attributes reduced delinquency by 26% for eligible users while maintaining approvals. Personify, which pairs Plaid bank transaction data with Prism Data analytics in its underwriting stack, expanded its use of cash flow data to more than 90% of applicant traffic without harming conversion. It increased approvals by up to 8% in underserved segments while holding credit costs steady.
Version 1 was trained on 1.44 million credit tradelines using Plaid transaction data and bureau performance labels. The consumer reporting subsidiary can draw on affiliate data to develop, test, and improve services. Its privacy terms also allow use of data stripped of personal identifiers or combined into aggregates for new products and research. Those terms describe what may be reused, but leave institution specific separation and retention of material used to train the scoring model unspecified.
Plaid Check’s privacy policy covers the bank account data used in its commercial consumer reports. It explains how the consumer reporting subsidiary collects and shares information for legally permitted purposes and sets retention by legal, contractual, and operational need. It allows continued retention for circumstances including an active lender connection, fraud prevention, or legal compliance. Consumers can revoke access or exercise applicable rights through the service center. The policy distinguishes the subsidiary’s responsibilities from those of the lender and Plaid Inc.
The Plaid Trust Center lists its SOC 2 Type II examination and ISO 27001 and ISO 27701 certifications. Schellman performs the annual SOC 2 audit covering security, availability, and confidentiality. Plaid Check’s own privacy terms address protection of consumer report information and how its service uses identity verification. The public summaries do not say whether the Plaid Check report workflow sits inside those examinations, which the reports and certificates themselves would settle.
Plaid Check is the trading name of Plaid Consumer Reporting Agency, Inc. It provides reports for permissible purposes under the Fair Credit Reporting Act. Plaid is also listed by Fannie Mae as an authorized supplier of asset verification reports through Desktop Underwriter. The home lending report can be used in Fannie Mae Day 1 Certainty and Freddie Mac’s Asset and Income Modeler. That authorization concerns asset verification reports. It does not confer agency approval on LendScore or on a lender’s credit decisions.
FairPlay independently reviewed version 1 of LendScore for disparate impact and whether the training data was representative. Plaid says the review found relatively low correlation between scoring features and protected attributes and no disparate approvals at the thresholds it describes. The model also constrains the direction in which a change in a feature can move a score. Five adverse action reason codes accompany each response. The findings come without a demographic breakdown of approval rates or the full independent test package.
A borrower can obtain their Plaid Check report, dispute an incomplete or inaccurate entry, or request reinvestigation through the consumer service center. Consumers can also freeze their file or place a fraud alert. Those options apply to the consumer report that supplies credit inputs. When the question is why an application was denied, the lender that made the credit decision handles it. The dispute process gives the borrower a route back to the reporting subsidiary when an input is wrong.
XGBoost is the basis for the version 1 score, which Plaid trained using its network’s transaction data and credit bureau performance labels. FairPlay performs independent fairness review, a separate role from supplying or hosting the model. Check also offers a Partner Insights option with Prism Data, which provides a separate cash flow assessment. The bureaus whose performance labels trained the score go unnamed.
The Consumer Report interface retrieves report contents and product results for the accounts covered by a specific report. Home lenders can share eligible reports with Fannie Mae and Freddie Mac, and Plaid describes token based sharing for eligible Experian use cases. The partner connection passes the report through an authorized channel. Lenders can integrate rating inputs through the same product family as their home mortgage asset reports, while selecting which report modules and account history the workflow consumes.
The lending service is hosted and accessed through Plaid Link and its interfaces. The broader Plaid end user policy describes United States regions on Amazon Web Services for certain cross border transfers. The Plaid Check service handles bank account data for reports and controls access to eligible partners. It has no separate region selection or data residency commitment stated for the hosted lending product. The lender receives the report and stores its own copy according to its policies.
Module pricing and bundle pricing are both available for Plaid Check. The Underwriting bundle combines the Base Report, Cash Flow Insights, Network Insights, and LendScore. The Income bundle pairs the Base Report with Income Insights. Under the bundle structure, the first billable report retrieval for a given user triggers the charge. With individual product pricing, most Check modules charge when the lender retrieves that product’s output. Plaid quotes the rate for each configuration. A general Plaid sandbox account allows testing before production access.
Consumer lenders use the score and cash flow measures for credit applications, and mortgage lenders have an asset verification report for home lending. Underwriting tools cover personal credit and include analysis of income stability, cash flow, and previous financial app connections. The servicing product refreshes an existing report after origination for an ongoing lender relationship. Plaid also offers dedicated materials for banks and credit unions. Personify and Fragile illustrate lender use. Plaid Protect and Signal carry the same network into fraud screening and ACH payment risk for lenders, payment apps, and crypto platforms.
Alternatives to Plaid
Other products that do the same kind of work as Plaid, and what each one does.
Upstart operates an artificial intelligence lending marketplace through which more than 100 banks and credit unions use its underwriting models and cloud applications to originate consumer credit.
CreditXpert helps mortgage lenders and brokers see how an applicant's mortgage credit score could change before the loan is priced or underwritten.
Prism Data turns bank transaction data into credit scores for lenders. CashScore, its main product, is a three digit score trained on pooled data from many lenders.
CredoLab scores creditworthiness from smartphone device metadata for banks, consumer finance companies, auto lenders, online and mobile lenders, insurers and retailers, aimed at applicants with no credit file.
Traive supplies credit risk assessment and asset qualification across the agricultural finance chain, serving lenders, crop input manufacturers, traders, cooperatives and capital markets participants rather than farmers directly.
Trusting Social, operating through Trust IQ Pte. Ltd. in Singapore, scores consumers with little or no formal credit history for more than 130 financial institutions across India, Indonesia and Vietnam, and partners with UNO Digital Bank in the Philippines.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | Data Protection Terms | Implementation | Source |
|---|---|---|---|---|
|
Not published. Plaid Check rates depend on the contracted modules or bundle.
|
The Income bundle includes a Base Report and Income Insights. Underwriting includes the Base Report, Cash Flow Insights, Network Insights, and LendScore. The first billable report retrieval for a user triggers the bundle charge. With individual product pricing, most modules generate a charge when the lender retrieves their output. | Plaid Check has product specific consumer privacy terms under federal consumer reporting and financial privacy law. | Plaid provides a sandbox for testing its interfaces. Setup fees for a production Check integration are not specified. | Vendor Published |
Plaid Check sells consumer report products individually or through Income and Underwriting bundles. In a bundle, the initial billable retrieval for a user triggers the bundle fee. Individual modules are generally charged when the lender retrieves the module’s output. The lending rate is set by agreement for the products in use.