A borrower with no credit history is not the same as a borrower with a bad one. In practice, many underwriting decisions still treat them the same way.
Thin-file borrowers represent one of the lending industry's biggest untapped revenue opportunities. In the US alone, 19 percent of adults have no conventional credit score: 28 million who are credit invisible, and a further 21 million who are unscoreable [1]. Research by PYMNTS Intelligence found that nearly a third of US consumers are credit insecure [2], and that repeated rejection pushes many of them from credit marginalised into credit avoidant, shrinking the addressable market further with every decline. These are not high-risk borrowers. They are borrowers the data cannot see.
The problem is not the borrowers. It is the data.
That gap has a name: cash flow underwriting, and it is growing fast. Lenders across consumer credit, BNPL, and multi-line banking are adopting it because the bureau alone is no longer enough [3]. The gap is where mispriced risk lives, and it is larger than most credit committees realise.
The Bureau File Records History
Bureau scores have underpinned credit decisions for decades. They work well for borrowers with long, established credit histories. So why is the lending industry leaving money on the table?
Because the bureau answers one question and one question only: has this person repaid debt before? It is a reasonable question. But it is not the same as asking whether they can repay debt now. For borrowers with thin files, most standard underwriting models offer no mechanism to answer the second question. That is where the gap begins.
A bureau score captures historical borrowing behaviour. It updates monthly and lags a borrower's actual financial position by weeks or months [4]. For anyone newer to credit, newer to a country, or simply someone who manages their finances without borrowing, the record is thin, silent, or missing entirely. Two applicants can present with identical scores and represent entirely different risk profiles: one financially stable, the other under pressure the monthly update cycle has not yet caught.
That distinction has shaped lending for longer than most credit teams would care to admit.
Cash flow underwriting answers the second question, from data that is current rather than historical. The two approaches are not competing versions of the same assessment. They answer different questions, and a lender who only asks the first one is working with half the picture.
Three Approaches, Three Answers
Most lenders today are working with one of three data approaches, and each tells a different part of the story.
The bureau file captures how a borrower has behaved in the past. It is well-established, widely understood, and useful for borrowers with long credit histories. Its limitation is time: it records what happened, not what is happening now.
Open banking data is a step forward. It gives lenders access to real-time transaction information directly from a borrower's bank account, including recent inflows, outflows, and balances, without waiting for the monthly update. Open banking has improved the picture [5]. But raw transaction data does not interpret itself. Knowing that money arrived last Tuesday is not the same as knowing whether that income is regular, stable, and sufficient. Open banking gives a lender access to the transaction stream. It does not read what that stream means.
Behavioural signals, the layer Neumetria provides, read the pattern behind the raw data: whether income lands on a predictable cycle, how much is already committed to recurring obligations, whether capacity has actually opened up, and whether spending has remained stable or shifted. This is the interpretation layer, and it is what determines whether the transaction data in front of a lender describes a reliable borrower or a vulnerable one.
These three approaches work together where all three are in play. Each one answers a question the others cannot. A bureau file, open banking data, and behavioural signals give a lender a far more complete picture of an applicant than any one of them provides alone.
What the Transaction Stream Already Knows
The data to answer the second question has always existed. It sits in the transaction stream a lender already pulls. What has been missing is not the data. It is the question.
Cash flow underwriting reads what the bureau cannot, from data the lender already holds. For lending decisions, the key questions it answers are:
Is this income real and regular? Not just whether money arrives, but whether it arrives on a consistent cycle and in predictable amounts. The transaction stream reads income rhythm and income stability directly: how regular the cycle is, and how consistent the amount. A borrower with twelve months of regular, stable inflows is a different proposition from one with irregular, variable income, even if the annual total looks similar. Irregular income is one of the strongest predictors of early-stage default, and one that does not appear in a monthly bureau update until a payment is already missed [6].
What can this borrower actually afford? Stated income figures and bureau-implied capacity are both proxies. The transaction stream shows what is genuinely left after recurring obligations go out each month. This is the number that determines whether a borrower can service a new obligation. The gap between stated capacity and actual capacity is where lenders absorb losses they never saw coming.
How much of this income is already committed? Two borrowers with the same income can carry very different recurring loads: the share of income already committed to fixed obligations. A borrower whose recurring load is high has little room for an additional obligation, regardless of what their credit history implies about their quality.
Has anything changed recently? Consistent spending signals financial resilience, while shifting patterns signal pressure. Spending stability, measured by how consistent patterns are over time, is one of the signals the transaction stream surfaces and a static monthly file cannot. The transaction stream shows which picture is accurate. A snapshot taken last month does not.
The consequences of that gap are larger than they first appear.
Four Places the Credit Decision Gets Sharper
These signals do not replace the bureau file. They sit alongside it. For lenders, four things change.
Income figures stop being guesswork. Most lenders verify income through declarations or payslips. The transaction stream shows what actually lands, how often, and whether it is consistent. For borrowers with variable or non-traditional income, this distinction changes the decision.
Actual capacity replaces stated capacity. Debt-to-income ratios built on declared figures can be significantly wrong in either direction. A transaction-based capacity window is built from what is actually moving through the account. It tells a lender what the borrower can genuinely service, not what they claim to earn minus what the file thinks they owe. Experian reports that its Credit + Cashflow Score — a model combining bureau, cash flow and alternative data — improves predictive accuracy by over 40 percent against conventional credit models [7]. That is a vendor's figure for a vendor's product and should be read as one. What it demonstrates is not that a particular score is better, but that the transaction stream carries predictive signal the bureau file does not.
The decline pile becomes accessible revenue. A thin file does not mean a high-risk borrower. It means someone whose financial life has not passed through the credit system in a traceable way.
The cleanest causal evidence sits in small business lending, where a National Bureau of Economic Research study compared applicants routed to lenders using different underwriting models. Entrepreneurs under 40 assigned to a cash-flow-intensive lender were approved 2.4 percentage points more often — a 12 percent lift on the mean approval rate for that group — with the effect concentrated among low-FICO applicants and no deterioration in subsequent business survival [8]. That is SME credit rather than consumer credit, and the population is deliberately narrow. The mechanism it isolates is not: when a model reads cash flow rather than file thickness, applicants the bureau could not price become approvable without the portfolio absorbing more risk.
Cash flow data gives lenders a way to assess these applicants on what they actually do with their money, rather than on the absence of a borrowing record.
Deterioration surfaces before a missed payment. A borrower under financial pressure will show it in spending and cash flow patterns before a missed payment appears in the file. Catching that signal before origination is the difference between a portfolio that performs as modelled and one that underperforms from the start.
Most lenders already hold the data to answer this. The question is whether they are asking it.
Monday Morning
Two applications arrive. Both request the same amount. Both show similar wages, a near-prime score, and a thin tradeline. Under a bureau-only model, they are close to identical.
Read the transaction stream and they separate. One has received the same salary on the same day every fortnight for eleven months, carries a moderate recurring load, and has a capacity window that comfortably covers the requested obligation. The other's inflows have turned irregular over the last quarter, the recurring load has crept up against them, and discretionary spending has tightened in the way it does when someone has started drawing on a buffer.
Neither reading is a prediction. Both are descriptions of where the applicant stands today, and they are different descriptions. The bureau file records the same thing for both, because on its own terms nothing has gone wrong yet.
These are not two different opinions on the same applicant. They are two different questions. One asks what the borrower has done. The other asks what they can do. Both matter. Only one of them is being asked at the point of decision.
Signal In. Context Out. Lender Decides.
Neumetria reads the transaction stream a lender pulls and surfaces these signals before origination. Integration works directly with the transaction data a lender already holds, through a documented API. There is no new data source to acquire. The signals inform the decision. The lender makes it.
This is not a credit decisioning engine. It does not approve or decline. It gives the lender's existing underwriting model a fuller picture of the applicant in front of it, the kind of picture that the monthly file, however accurate, was never designed to provide.
For decades, underwriting has relied on records of borrowing to estimate future repayment. Today, lenders can observe financial behaviour directly. The question is no longer whether that data exists. It is whether underwriting models are prepared to ignore it.
References
[1] Experian and Oliver Wyman, January 2022: "Nineteen percent of American adults do not have a conventional credit score. This includes 28 million adult Americans who are credit invisible and 21 million who are unscoreable." experianplc.com
[2] PYMNTS Intelligence / Sezzle, 2023: "Nearly one-third of US consumers were credit insecure. Repeated rejection often pushes consumers from being credit marginalised into becoming credit avoidant." pymnts.com
[3] Alternative credit scoring market valued at USD 1.8 billion in 2026, growing at 23.1% CAGR through 2035. Market.us, February 2026. market.us
[4] Experian: lenders typically update credit bureaus once per month. Updates appear on reports within 30 to 45 days. experian.com
[5] Cloudkaptan Insights, February 2026: "Open banking data is now used to understand income stability, spending rigidity, and remaining repayment capacity in near real time." cloudkaptan.com
[6] Finezza Blog, November 2025: "Irregular salary credits, missing bonuses, or changing jobs are among the strongest predictors of repayment risk." finezza.in
[7] Experian plc, 10 November 2025: the Experian Credit + Cashflow Score improves "predictive accuracy by over 40% when compared to conventional credit models." experianplc.com
[8] Hair, Howell, Johnson and Matsumoto, "Modernizing Access to Credit for Younger Entrepreneurs: From FICO to Cash Flow," NBER Working Paper 33367: "Assignment to a cash-flow-intensive lender increases approval chances for entrepreneurs under 40 by 2.4 percentage points (12 percent of the mean approval rate)," concentrated among low-FICO applicants and validated against business survival data. nber.org