The term means two different things
Model-based cash-flow underwriting uses electronic transaction data from deposit accounts, and sometimes accounting software, as a direct input to a score or model that assesses repayment capacity. The data is typically pulled through an aggregator with the applicant's permission, and the output sits alongside or partly replaces a bureau score. This is what the term means in a fintech context, and it is what most articles on the subject are describing.
Cash-flow analysis in credit underwriting is the traditional discipline: computing debt service coverage, global cash flow and debt service capacity from tax returns, financial statements and bank statements. Every commercial lender in the country does this. It is not an innovation, it is the core of commercial credit, and it has been for decades.
The distinction matters because the honest answer to "which lenders use cash-flow based underwriting" is that essentially all commercial lenders use the second and a much narrower set uses the first. A bank asking whether it should adopt cash-flow underwriting is often already doing rigorous cash-flow analysis and is really being asked whether it wants to buy a transaction-data scoring model, which is a different purchase with different obligations.
The rest of this article covers the first meaning, since that is what the question usually intends. For the definitional groundwork, see what is cash flow-based credit scoring.
Who actually uses it today
Adoption in the United States is uneven, and it is spreading faster in small business lending than in consumer lending, a pattern FinRegLab has documented consistently across several years of research.
| Segment | Where it stands | What it is used for |
|---|---|---|
| Fintech small business lenders | The most advanced adopters, and effectively the origin of the practice | A primary underwriting signal, with owner credit history often secondary rather than dominant |
| CDFIs, MDIs and mission-based lenders | Active and publicly documented pilot programmes | Extending credit to thin-file owners and young businesses within a mission mandate |
| Non-bank consumer lenders | Some adoption, notably slower than the small business side | Thin-file and near-prime expansion where bureau data is insufficient |
| Community banks and credit unions | Uneven, and mostly the traditional discipline rather than the model-based one | Debt service coverage and global cash flow from documents; some deposit-data pilots |
| Large banks | Experimentation, often through structured industry working groups | Machine learning model management and consumer underwriting research |
| Commercial and CRE lenders generally | Universal, in the traditional sense | DSCR, global cash flow, debt service capacity. Not new, and not what the fintech term describes |
On named institutions, the most useful public source is FinRegLab's work on technology and data adoption in mission-based lending, which documents pilots using cash-flow data for small business lending at Allies for Community Business, Ascendus, LiftFund, Ponce Bank and Texas National Bank, alongside other community-based lender initiatives. Its 2026 work on managing machine learning underwriting models reflects discussions with banks participating in a technology working group convened under the OCC's Project REACh.
That is a meaningful list precisely because it is not a list of fintechs. Mission lenders and community institutions adopting this is the more interesting signal, because they operate under the same constraints most readers of this article do.
What the evidence actually shows
The research base here is better than in most areas of lending technology, largely because FinRegLab has done independent empirical work rather than vendor-sponsored studies.
On small business, its June 2025 paper with Sabrina Howell and Siena Matsumoto at NYU Stern analysed more than 38,000 small business loans originated by two fintech lenders between February 2015 and January 2024. Adding cash-flow information substantially increased the predictive signal of models that had relied mainly on the owner's personal credit score and firm characteristics. The effect was larger for owners with low scores whose businesses were less than five years old, which is precisely the population most often declined on thin file.
On consumer credit, its July 2025 study found that adopting machine learning techniques and incorporating cash-flow data increased predictiveness and expanded access without increasing default risk. A finding worth noting for smaller institutions: staging the two changes rather than making both at once still delivered measurable gains, so this does not have to be an all-or-nothing modernisation.
The market context is the reason any of this matters. FinRegLab has estimated that 45 to 60 million US consumers lack sufficient credit history to generate a reliable score, with many more constrained by low scores. Cash-flow data is one of the few inputs that speaks directly to current capacity rather than past borrowing behaviour.
Why banks and credit unions lag
Four reasons, and only one of them is inertia.
Resource and technology hurdles
FinRegLab has repeatedly flagged that smaller traditional lenders face particular cost and capability constraints here, despite community banks historically playing an outsized role in small business lending. Buying a model is not the hard part; integrating and governing it with a lean team is.
The data rail is unsettled
Underwriting on accounts you do not hold depends on aggregator access, and the CFPB's Section 1033 rule is currently enjoined and being rewritten, so access rests on commercial arrangements rather than a data right.
It is a model, with everything that follows
A cash-flow scoring model sits inside model risk management scope, which since April 2026 means SR 26-2 and its parallel OCC and FDIC issuances. Lighter than SR 11-7 in places, but you still classify, document and defend.
Adverse action does not get easier
A decline driven by a cash-flow model still requires specific and accurate principal reasons, and regulators have been explicit that algorithmic complexity does not reduce the obligation. If the model cannot say why, it cannot be deployed on declines.
The current position on open banking access is covered in dynamic credit limits tied to cash flow, and the model classification question in how risk decisioning software works.
The advantage banks and credit unions overlook
Here is the part that gets missed in this debate. A fintech lender needs an aggregator because it does not hold the applicant's deposit account. A bank or credit union underwriting an existing member or customer very often does.
That is a structural advantage, and it inverts the usual framing. The institution with the deposit relationship has the cleanest, longest and most current view of that applicant's cash flow, without a consent flow, without an aggregator contract, and without a dependency on how Section 1033 is eventually resolved. It is sitting in the core.
The constraint is rarely access to the data. It is that the data is not assembled into anything a credit decision can use, and that the documents which would complete the picture, tax returns, financial statements, statements from accounts held elsewhere, still arrive as PDFs that somebody re-keys.
Where Uptiq fits
Uptiq does not sell a cash-flow scoring model, and an institution's credit policy and models should be its own. What Uptiq does is the input layer that both meanings of cash-flow underwriting depend on. The Bank Statement Analysis and Cash Flow agents turn statements and transaction documents into structured, categorised figures with each value cited back to its source page. The Financial Spreading agents compute debt service coverage and global cash flow using your definitions rather than a generic template, across entities and guarantors. Document AI classifies and extracts the tax returns and financial statements that complete the picture, so the analysis reflects the whole borrower rather than the part that is convenient to gather. Confidence thresholds route items into an exception queue, overrides are retained with reason and user, and the credit decision stays with your credit authority. The agents run above existing core, origination and document systems through more than 100 native integrations. Related reading: global cash flow analysis software and how to calculate DSCR.
How to approach it
Decide which version you are actually buying
A transaction-data scoring model and faster, more complete traditional cash-flow analysis are different purchases with different governance consequences. Most institutions asking the question want the second and have been shown the first.
Start with the accounts you already hold
If the applicant banks with you, the transaction history is in your core. Assembling it into a usable view requires no aggregator, no consent flow and no dependency on how open banking rules land.
Fix the document side before the model side
A model fed by re-keyed figures inherits every transcription error, and the same documents that feed the model feed the file an examiner will look at.
If you do adopt a model, settle adverse action first
What the reason codes will say, who configures the thresholds, and whether the drivers behind a decline can be reproduced months later. This constrains the design, so it belongs before selection.
Stage it
The consumer research suggests that adopting one change at a time still produces measurable gains, which is a more realistic path for an institution without a dedicated modelling team than attempting both at once.
Frequently asked questions
Which lenders use cash-flow based underwriting?
In the model-based sense, adoption is led by fintech small business lenders, followed by CDFIs, minority depository institutions and other mission-based lenders running documented pilots, with some non-bank consumer lenders and experimentation at large banks. Community banks and credit unions are uneven adopters of the model-based form. In the traditional sense of analysing debt service coverage and global cash flow from financial statements, essentially every commercial lender uses cash-flow underwriting and always has.
Are there publicly named examples?
Yes. FinRegLab's research on technology and data adoption in mission-based lending documents small business cash-flow data pilots at Allies for Community Business, Ascendus, LiftFund, Ponce Bank and Texas National Bank, among other community-based lender initiatives. Its later work on managing machine learning underwriting models draws on discussions with banks in a working group convened under the OCC's Project REACh.
Does cash-flow underwriting actually predict better?
The independent evidence says it adds real signal. FinRegLab's 2025 study with NYU Stern researchers, covering more than 38,000 small business loans from two fintech lenders, found that adding cash-flow information substantially increased predictive power over models built mainly on owner credit scores and firm characteristics, with the largest effects for low-score owners of businesses under five years old. Separate consumer research found gains in predictiveness and access without increased default risk.
Why has adoption been slower in consumer lending than small business?
FinRegLab has observed this pattern directly. Small business lending has a longer tradition of examining bank statements and financial reports, so cash-flow data extends an existing practice rather than replacing one. Consumer underwriting is more heavily standardised around bureau scores, carries a denser set of compliance obligations, and involves privacy expectations that are particularly sensitive where personal and business accounts overlap.
Do we need open banking data to do this?
Not if you hold the account. A bank or credit union underwriting an existing customer already has the transaction history in its core, without an aggregator or a consent flow. Open banking access matters for accounts held elsewhere, and that access currently rests on commercial arrangements because the CFPB's Section 1033 rule is enjoined and under reconsideration.
What are the compliance implications of a cash-flow model?
It is a model, so it sits inside model risk management expectations, which for banks now means the April 2026 interagency guidance rather than SR 11-7. Declines require specific and accurate principal reasons, and regulators have stated that algorithmic complexity does not reduce that obligation. Fair lending testing applies as it would to any underwriting model. Confirm your own position with your compliance and model risk functions.
Research findings described here are drawn from publicly available FinRegLab publications and are summarised rather than reproduced; readers should consult the original reports for methodology and limitations. Named institutions are identified as participants in publicly documented pilots and research, and their inclusion does not imply any relationship with Uptiq. Regulatory descriptions reflect publicly available sources as of September 2026 and are subject to change. Nothing here is legal, compliance or regulatory advice; confirm with your own legal, compliance and model risk functions. Performance figures are Uptiq platform benchmarks across production deployments and are not a guarantee of results at any individual institution.
Start with the data you already hold
Tell us what your cash-flow analysis looks like today and we will run your own statements and returns through it, with every figure traced back to its page.
