Automated underwriting

A modern working-capital or cash-advance decision is largely mechanical. Bank data arrives through a read-only aggregator connection or parsed statement uploads, and a model extracts deposit counts and amounts, average and minimum balances, negative days, returned items, existing daily or weekly debits that reveal other funders, and the seasonality of receipts. That is joined to credit bureau data, business filings, UCC records, industry codes and prior performance with the funder.

The output is not just approve or decline. It sets the amount, the term, the payment frequency and the price tier, and it flags exceptions for a human to look at. Most funders operate a hybrid: automated for the standard range of their credit box, manual review above a size threshold or when the file trips a rule.

Two consequences for applicants. The decision is only as good as the data, so a missing month, a newly opened account with no history, or an aggregator connection that fails will produce a worse answer than the business deserves. And the model reads patterns, not explanations. Seasonality that any human would recognize looks like decline to a model that has three months of statements from your slow quarter.

Regulation lags the practice. The Equal Credit Opportunity Act and Regulation B do apply to business credit, including notification of action taken, though the requirements differ by applicant revenue and are lighter than on the consumer side. Whether ECOA reaches transactions structured as purchases of receivables rather than credit is contested. The CFPB rule collecting small-business lending data under Section 1071 has been through repeated litigation, delay and re-proposal, so its status should be checked rather than assumed.