Open banking underwriting is the practice of assessing a borrower's creditworthiness from bank-account and transaction data that the borrower has chosen to share, instead of relying only on bureau files and tax documents. For credit unions, community banks and lenders, it turns income deposits, recurring expenses and payment patterns into a current view of repayment capacity, provided consent, data quality, fair lending and model governance are handled with the same discipline as any other credit input.
This guide walks through how the data moves, what it adds to a credit decision, what the research shows, which US rules apply, and how to put the process into production without creating exam findings.
TL;DR
Automate Regulatory Model Risk Governance
Examine models against 32 qualitative criteria and resolve risk Tiers with pre-deployment checklists per OCC 2011-12 guidelines.
- Open banking underwriting uses consumer-permissioned account data, delivered through APIs and data aggregators, to evaluate income, expenses and payment behavior alongside or in place of traditional credit data.
- Research from FinRegLab and the NBER finds that cash-flow data is generally at least as predictive as traditional scores, and that combining the two tends to improve risk prediction and approval rates.
- The CFPB's Section 1033 rule sets consent, purpose-limitation and one-year authorization requirements for third parties, and ECOA, Regulation B and FCRA adverse action duties still apply to the decision.
- Implementation comes down to four controls: documented consent, data-quality checks, model risk management with independent validation, and explainable decisions with audit trails.
Table of Contents
- What open banking underwriting means in the US
- How lenders use account and transaction data
- What the evidence shows
- The US rules that shape the process
- Putting open banking underwriting into production
- Where credit unions and community banks have an edge
- Turning bank statements into decisions with RiskInMind
- FAQ
- Sources
What open banking underwriting means in the US
Open banking lets consumers share bank-account data, such as balances and transaction history, with third-party service providers through APIs that data aggregators facilitate. The Kansas City Fed's payments briefing describes the typical flow: the consumer authorizes the third party, the third party requests the data from an aggregator through an API, the aggregator obtains it from the bank through another API, and then forwards it on.

In underwriting, the third party is often the lender itself or a vendor acting for the lender. The same briefing notes that a potential lender may use transaction history, including income deposits, recurring expenses and payment patterns, as an alternative or addition to traditional credit data when assessing creditworthiness and setting loan terms. It also notes this may expand credit access for consumers with no or limited credit histories.
Two features separate the US from regulated open banking regimes elsewhere. The first is that the US system has historically been a de facto one, built on private agreements between banks and aggregators. The NBER authors cited below describe it exactly that way, in contrast to the regulated version in the UK. The second is that the CFPB's rule under Section 1033 is moving the US toward a defined, standardized framework, which we cover later in this guide.
Why API access matters to a lender
The Kansas City Fed notes that open banking is more secure, efficient and timely than screen scraping. Consumers' credentials are shared with only a small number of data aggregators, and APIs let banks act as gatekeepers over the amount and type of data that leaves the institution. For a lender, that translates into a cleaner audit story: you can document where the data came from, what the borrower authorized and when.
How lenders use account and transaction data
Cash-flow evaluation is not a novelty. The Federal Reserve, CFPB, FDIC, NCUA and OCC said in their 2019 Interagency Statement on alternative data that evaluating a borrower's income and expenses to help determine repayment capacity is a well-established part of underwriting. They added that better measurement of income and expenses through cash-flow evaluation may be particularly useful for consumers with reliable income patterns from several sources rather than a single job.
In practice, underwriters and models extract a consistent set of signals from account data:
- Income stability. Frequency, amount and source of deposits, including gig, seasonal and multi-employer income that a pay stub cannot capture.
- Fixed obligations. Rent, loan payments, utilities and subscriptions, which show real debt service capacity rather than reported capacity.
- Liquidity behavior. Average and minimum balances, and the buffer between income arrival and obligation due dates.
- Distress indicators. Overdrafts, returned items and sudden changes in deposit patterns.
An NBER working paper from January 2025, analyzing business checking account data, found that measures of business health and revenue are associated with lower chances of default, while distress measures such as overdrafts predict higher chances. That mirrors what experienced credit officers look for when they read a statement by hand. The difference is that the data now arrives structured, so the same logic can be applied to every application in seconds.
Small business and commercial borrowers
Cash-flow data is not limited to consumer loans. FinRegLab's small business fact sheet describes cash-flow data as including company bank account records, feeds from accounting software and transaction information from e-commerce platforms and payment processors. It says this gives a more detailed and timely picture of how applicants manage their finances than traditional credit reports, and is particularly valuable for startups and owners without strong credit histories.
For community institutions where small business lending is a core line, this is where permissioned data tends to earn its place first. Our guide to SMB lending AI for banks and credit unions covers how those workflows are structured, and the Business Loan Qualifier shows how a borrower's financial profile maps to qualification.
What the evidence shows
The strongest US evidence comes from studies that compared models built on bureau data, cash-flow data and both together, then tested them against actual repayment.
Combined models performed best. FinRegLab's empirical white paper built models from an anonymized dataset that linked one of the three nationwide credit bureaus with bank account information compiled by a data aggregator, and evaluated them against the performance of new accounts opened in 2018 and 2019. The machine learning model that combined bureau and cash-flow data was the most predictive overall and across all subgroups. It also had the highest approval rates overall and for most subgroups at most risk thresholds, with relatively low false positive rates.
Cash-flow variables hold their own. In FinRegLab's research with Charles River Associates, using data from six non-bank financial services providers, cash-flow variables and scores were generally at least as predictive as traditional credit scores. Combined models frequently improved risk prediction across traditional score bands. Where data was available, roughly 45 to 50 percent of borrowers had traditional scores below about 650, which is the population where an added signal matters most.
Gains are uneven across borrowers. The NBER paper, using data from three fintech lenders, found that timely information about ability to repay from business checking statements improves default prediction most for younger business owners: a 2.6 percentage point gain in ROC AUC for owners under 40, against 0.08 percentage points for owners over 40. Access to cash-flow-intensive underwriting also raised approval rates for younger entrepreneurs relative to older ones. The practical reading for a lender is that the value of the data depends on how thin the file would otherwise be.
Data quality is the constraint. The Federal Reserve Board's October 2025 publication cautions that unreliable access to transaction data, inconsistent quality and improperly structured data can reduce the effectiveness of new technologies and practices. It adds that economies of scale and profitability can be a challenge, particularly for small-dollar credit. That is a useful framing for pilots: choose a portfolio segment where the data is dense and the loan size justifies the analysis, then expand.

The US rules that shape the process
Open banking data enters a credit decision that is already governed by fair lending and consumer reporting law. Three layers matter.
Section 1033 and consumer data access
The CFPB's Personal Financial Data Rights final rule (12 CFR part 1033), published in the Federal Register on November 18, 2024, requires banks, credit unions and other financial providers to make consumer data available on request in a secure and reliable manner, and in a standardized, machine-readable format when it goes to authorized third parties. The rule as published set data-provider compliance dates beginning April 1, 2026 and phasing through April 1, 2030. It also prohibits providers from satisfying access requirements by letting third parties screen scrape, and prohibits fees for consumer and third-party data access.
Because the rule has been the subject of ongoing agency attention, confirm the current status and compliance schedule on the CFPB's official site before you set your own timeline.
For lenders acting as authorized third parties, the rule's consent and use provisions are the part that touches underwriting directly. An authorized third party must obtain the consumer's express informed consent through a signed authorization disclosure. It must limit collection, use and retention of covered data to what is reasonably necessary to provide the requested product or service, and limit collection under an authorization to a maximum of one year.
FCRA and the question of consumer reports
Whether aggregated account data constitutes a consumer report depends on who assembles it and for what purpose. The OCC's FCRA handbook defines a consumer report as communication by a consumer reporting agency bearing on creditworthiness, credit capacity, character or mode of living, used or expected to be used to establish eligibility for credit. Your compliance team and counsel should decide how each data source and vendor relationship fits that definition, and document the conclusion.
ECOA, Regulation B and adverse action
The Interagency Statement says creditors' use of cash-flow data can generally be explained and disclosed to the borrower as ECOA and FCRA may require, and that firms should consider adverse action notice requirements under Regulation B and FCRA. Under Regulation B, a creditor must notify an applicant of action taken within 30 days after receiving a completed application. An adverse action notice must be in writing and include either a statement of specific reasons or a disclosure of the right to request them within 30 days, if the request is made within 60 days of the notification.
The Federal Reserve's 2011 adverse action rule explains that FCRA Section 615(a) requires a notice when adverse action is based in whole or in part on information in a consumer report, and that certain Regulation B model notices cover the content required by both statutes. FinRegLab's April 2026 framework summarizes the practical consequence for model-driven decisions: ECOA requires individualized disclosure of the principal reasons for a rejection, and FCRA requires a list of key factors that negatively affected a credit score when that score was a factor.
The lesson for cash-flow models is simple. Every feature that can drive a decline, such as overdraft frequency or deposit volatility, must be translatable into a reason a borrower can understand. Choose and document features with that in mind from the first design meeting. Our overview of CFPB consumer credit AI rules goes deeper on how these expectations apply to automated decisioning.
The Congressional Research Service, as summarized by Orrick, notes that alternative data has shown promise for scoring previously unscorable borrowers and improving existing scores. It also notes that data security, privacy and fair lending concerns remain, and that institutions must stay compliant with GLBA and ECOA.
Putting open banking underwriting into production
A workable rollout follows the order in which risk arises.
- Define the use case and segment. Pick a product where cash flow is the natural lens: small business term loans, lines of credit, or thin-file consumer loans. Set the decision the data will support, whether a full decision, a referral to manual review, or a pricing input.
- Design consent as a control. Capture a clear authorization that names the purpose, the data categories and the duration, and set retention and refresh rules that stay within the one-year limit for authorizations. Store the record where an examiner can retrieve it by application.
- Validate the data before the model sees it. Test for completeness, duplicate accounts, categorization errors and gaps in history. The Fed's warning about inconsistent and improperly structured data is a reminder that bad inputs create bad decisions at scale.
- Build the model under model risk management. FinRegLab's framework describes analyzing whether models rely on conceptually sound relationships in the data and assessing stability as data conditions change. It also describes three lines of defense: business and operational staff, compliance and risk-control staff handling independent validation, and internal audit. Map your cash-flow model to those lines before launch.
- Test for fair lending outcomes. Compare approval rates and pricing across protected groups, and look for less discriminatory alternatives when disparities appear. Keep the analysis with the model documentation.
- Make decisions explainable. Link each model feature to an adverse action reason, and test that notices are generated within the Regulation B timeline.
- Secure the pipeline. Review API authentication, logging, data residency and vendor access. Our guides to API security for U.S. banks and banking data residency show how to turn GLBA and FFIEC expectations into exam evidence.
- Monitor after launch. Track default performance by segment, drift in income and expense patterns, and override rates, and schedule revalidation.
Speed deserves its own note. Cash-flow data supports fast decisions, but speed without controls is what draws scrutiny. Our piece on real-time credit decisions explains how lenders keep that pace safely, and the AI loan underwriting audit trail guide covers governance-first design. If you want the broader credit assessment framework around the cash-flow view, the borrower creditworthiness evaluation process is the companion read.
Where credit unions and community banks have an edge
Smaller institutions often assume open banking underwriting is a large-bank or fintech advantage. The evidence suggests otherwise. The populations where cash-flow data adds the most, such as thin-file borrowers, newer businesses and people with variable income, are populations community lenders already serve and know by name. A lender that pairs that relationship knowledge with a structured view of the account data can lend with more confidence to borrowers a bureau score alone would leave in the gray zone.
The constraint is capacity. Reading statements by hand does not scale, and building validated models in house is a heavy lift for a small risk team. That is the gap automation is meant to close, provided the automation stays inside your governance framework and your examiner can follow every step.
Our perspective
We see open banking underwriting as a question of control before it is a question of speed. The data is only useful if you can show where it came from, what the borrower consented to, how it moved through the decision and why the outcome was what it was. That is why we built RiskInMind around in-house AI processing with zero third-party data exposure, granular audit trails and SOC 2® aligned controls, so a credit union or community bank can use cash-flow signals without sending sensitive borrower data to outside parties.
Our approach pairs specialized AI agents for credit risk, regulatory compliance and document analysis under a central director, Ava. Transaction analysis, credit memo drafting and regulatory checks run in one workflow with response times under half a second, and every step is logged. For risk managers and compliance officers, that means the analyst sees the cash-flow picture, the examiner sees the evidence, and the borrower gets a decision that can be explained.
Our goal is simple: let institutions of every size lend on the full picture of a borrower's finances, with the governance an exam demands.
Raj
Turning bank statements into decisions with RiskInMind
If your team is ready to bring cash-flow analysis into underwriting, our AI Bank Statement Analyzer reads account data into a structured cash-flow view, and the Loan Application product carries that view into the application workflow. Credit memos can be drafted with the AI Credit Memo Generator, so the analysis behind each decision is documented as it is made.

We offer Starter, Professional and Enterprise plans; see the pricing page for details, or explore the full platform on the RiskInMind site to request a demo.
FAQ
What is open banking underwriting?
It is credit assessment that uses bank-account and transaction data a borrower has authorized the lender to access. The lender evaluates income deposits, recurring expenses and payment patterns to judge repayment capacity. It can supplement traditional credit data or, for thin-file borrowers, carry more of the decision.
Is open banking data more predictive than a credit score?
In FinRegLab's research, cash-flow variables and scores were generally at least as predictive as traditional credit scores. Models combining both frequently improved risk prediction across score bands. The strongest result in its empirical white paper came from the combined machine learning model.
How long can a lender use a borrower's authorization under Section 1033?
The CFPB's final rule limits collection under a single authorization to a maximum of one year. Third parties must also limit collection, use and retention of data to what is reasonably necessary for the requested product or service. Check the CFPB's site for the rule's current status before finalizing your process.
Do adverse action notice rules apply when a cash-flow model declines an applicant?
Yes. ECOA and Regulation B require written notice with specific reasons, or a disclosure of the right to request them. FCRA notices apply when a consumer report or score contributed to the action. Your model's features should therefore map to reasons that can be stated plainly to the applicant.
Does open banking underwriting work for small business loans?
Yes. FinRegLab describes business bank account records, accounting software feeds and e-commerce or payment processor data as a timely view of how applicants manage their finances. The NBER study found it particularly improved default prediction for younger business owners, a segment that often has limited credit history.
Sources
- Open Banking: A Critical Piece in the Future of Financial Services, Kansas City Fed
- Interagency Statement on the Use of Alternative Data in Credit Underwriting
- NBER working paper w33367, January 2025
- FinRegLab fact sheet: cash-flow data in underwriting small business credit
- FinRegLab: Advancing the Credit Ecosystem, machine learning and cash-flow data
- FinRegLab fact sheet: cash-flow data in credit underwriting
- Federal Reserve Board consumer and community context, October 2025
- CFPB Personal Financial Data Rights final rule, Federal Register, November 18, 2024
- OCC Examination Handbook 1300: FCRA, CAN-SPAM and TCPA
- eCFR: Regulation B, 12 CFR part 1002
- Federal Register, July 15, 2011: adverse action notices
- FinRegLab: Framework for Managing Machine Learning Models in Consumer Credit Underwriting, April 2026
- Orrick summary of the Congressional Research Service report on alternative data
