Cash-Flow Data Is Not a Shortcut Around Underwriting Discipline
Bank statements, accounting feeds and card data can sharpen SMB underwriting, but only when funders document source quality, fraud risk and explainable decision logic.
The fight over small-business underwriting data is not really a fight over who has the newest feed. It is a fight over which records actually explain repayment capacity, which records can be manipulated before funding, and which records can be explained after a decline, pricing decision or fraud miss. FinRegLab put the core promise plainly in its Small Business Spotlight on cash-flow underwriting : electronic bank-account records, accounting feeds, e-commerce data and payment-processor data sit at the center of efforts to make small-business credit faster and more predictive. The harder question for MCA and alt-finance operators is whether those signals are being treated as evidence, or as a faster way to automate the same thin file. Bank data Cash timing Deposits, withdrawals, volatility, NSFs and stacking clues. Accounting feeds Context Revenue categories, expenses, receivables and business-model detail. Card rails Sales pulse Gross card volume, returns, chargebacks and merchant concentration. Fraud signals Evidence quality Synthetic revenue, circular transfers, altered statements and undisclosed debt. The strongest file combines signals instead of crowning one winner Bank statements are immediate, but they do not explain gross margin or job profitability. Accounting software can explain categories, but the books may be stale, owner-maintained or cleaner for mature firms than for thin-file businesses. Processor data sees card activity, but not cash, checks, ACH, marketplace payouts or off-platform sales. Reviews and digital presence can verify that a business exists, but they can also reflect marketing budget, neighborhood, language and platform behavior. “Electronic cash-flow information, such as companies’ bank account records, feeds from accounting software, and transaction information from e-commerce platforms and payment processors, is at the core of these efforts.” FinRegLab, Small Business Spotlight on cash-flow underwriting The operating mistake is to turn a new feed into a magic variable. A serious underwriting file starts with an industry prior, then updates it with transaction behavior, owner liquidity, customer concentration, seasonality, existing obligations, lien position and local knowledge. Two contractors can share a NAICS code while one has signed jobs, clean draws and verified receivables and the other is funding payroll from a stack of short-duration advances. Speed creates its own fraud surface The faster a funding desk can approve a file, the more valuable manufactured signal becomes. Fraud risk is not limited to fake identities. It includes altered PDFs, circular transfers, synthetic revenue, borrowed bank accounts, phantom invoices, undisclosed stacks, chargeback games, platform manipulation and owners who understand which screenshots trigger a fast yes. That is why direct-source data helps, but does not end the review. Open-banking feeds reduce some statement-tampering risk. They do not prove that deposits are durable operating revenue, that processor volume is margin-positive, that the owner disclosed every position, or that the same merchant is not being shopped through multiple brokers with slightly different file packages. Opinion The cleaner operating model is not “more data.” It is a source-of-truth hierarchy. Direct bank connections beat uploads for cash timing. Processor feeds beat borrower summaries for card sales. Accounting records explain what bank data cannot, but only when update cadence and category quality are visible. Every model output should point back to the evidence it used. Explainability is not only a bank problem FinRegLab’s machine-learning policy analysis says ML underwriting already affects hundreds of thousands of consumer and small-business applications each week. It also warns that stakeholders are still debating whether post hoc explanation tools are sufficient for model governance, adverse-action notices and fair-lending review. “Every week, machine learning underwriting models are determining the outcomes of credit applications submitted by hundreds of thousands of consumers and small business owners.” FinRegLab, Explainability and Fairness in Machine Learning for Credit Underwriting Alt-finance operators should not hear that as a compliance paragraph and move on. It is a capital-markets paragraph. If a funder cannot explain why a model penalized a merchant, why a broker cohort is deteriorating, why a processor feed overrode a bank-statement warning, or why a renewal was offered despite rising NSFs, the model becomes hard to trust in diligence, servicing and dispute resolution. Section 1071 adds another reason to keep the file clean. The CFPB describes Section 1071 as requiring covered financial institutions to compile, maintain and submit certain small-business credit application data. Coverage details can change by product and institution, so operators should verify obligations with counsel. But the practical point is broader: application data, decision reasons and source evidence are becoming harder to treat as informal desk notes. Market context: more online credit, more borrower friction The Federal Reserve Banks’ 2026 Report on Employer Firms found that 60% of small employer firms applied for financing in the prior year, while online-lender use continued to matter for borrowers outside traditional bank channels. The report also cautions that the survey is a convenience sample, not a random sample. That caveat matters. Operators should use these figures as market context, not as a portfolio benchmark. “The 2025 SBCS was fielded from September 3 to November 14, 2025. It yielded 6,525 responses from a nationwide convenience sample of small employer firms with 1-499 full- or part-time employees across all 50 states and the District of Columbia.” Federal Reserve Banks, 2026 Report on Employer Firms The FTC’s small-business financing perspective is useful because it links alternative data to borrower-facing cost and product design. It notes that payment processors and technology platforms offer financing based partly on data they already possess, including receipts or sales data. It also notes that MCA providers buy future receivables and that the factor is often 20% to 50% of the amount advanced. “Some base their underwriting on financial technologies and alternative data.” Federal Trade Commission, Small business financing staff perspective What belongs in the data-control file Signal What it answers What to verify Bank transactions Cash timing, deposit volatility, NSFs, debt service and daily liquidity. Direct connection status, excluded accounts, circular transfers and undisclosed positions. Accounting data Revenue categories, margin, receivables, payables and expense behavior. Bookkeeping cadence, owner edits, reconciliations and category quality. Processor and platform feeds Card sales, returns, chargebacks, marketplace dependence and real-time activity. Off-platform revenue, refunds, concentration and whether volume is margin-positive. External validation Business existence, liens, licenses, reviews, website history and local context. Bias, spoofing, stale registrations, paid-review artifacts and geography effects. The goal is not to slow every file into a bank-style credit memo. The goal is to make the fast file explainable. A renewal desk should know which signals improved after the first advance. A syndication partner should know whether broker cohorts differ by loss severity, not just approval rate. A servicing team should know which pre-funding warning signs were present before ACH failures started. For Plutus readers, the practical answer is a layered file: source-of-truth hierarchy, direct-feed preference, fraud exceptions, decision reason codes, renewal outcome tracking, cohort-level broker review and a clear caveat whenever a signal is directional rather than dispositive. Cash-flow data can make underwriting better. It can also make weak underwriting faster.