Cash flow underwriting: what to measure, and where the data comes from

Cash flow underwriting judges whether a borrower can repay a loan from the money moving through their bank accounts. The lender looks at what comes in and what goes out over several months, and at the balance left each day. Some lenders use it in place of a credit score. Others use it alongside the credit report. The data comes from PDF bank statements, or from a feed the applicant approves through a service such as Plaid.

How cash flow underwriting differs from a credit check#

A credit report shows how someone has repaid debts in the past. It says little about whether their income covers a new payment today. Bank account data shows that directly. Five federal regulators described the idea in a 2019 joint statement. Cash flow analysis, they wrote, checks whether a borrower can meet new or existing recurring payments. It does so by looking at income and expenses over time.

The data matters most for people the credit system can't score. By the CFPB's June 2025 estimate, 2.7% of US adults had no credit record in December 2020. Another 9.8% had a record with too little history, or no recent activity, to produce a score. For applicants like these, a lender may underwrite on cash flow alone. Other lenders use it as a "second look" for applicants they would turn down on the credit file alone. Others use it next to the credit file for every applicant.

Research supports using cash flow data on its own and alongside the credit file. FinRegLab, a nonprofit research group, studied six non-bank lenders in 2019. It found that their cash flow measures, on their own, generally predicted credit risk as well as traditional credit scores. They also often showed which borrowers were riskier among those the credit scores rated the same. FinRegLab's 2025 consumer study found that adding cash flow data to credit bureau data gave a modest gain. Still, its machine learning model that used both was the most predictive, overall and for every subgroup.

Five measures and what can make each one wrong#

FinRegLab's 2019 study and its 2025 report on mission-based small business lenders found lenders working out the measures below. Visa, Plaid and an agency of the US Commerce Department supported that 2025 report. The table shows how each measure comes from statement lines, and what can make it wrong. For worked numbers, see bank statement analysis.

Two limits apply to every measure. First, a measure is only as complete as the file, so one missing month or one account left out changes the result. FinRegLab found that lenders sometimes have to coach applicants to include every account that matters. Second, no statement shows a debt that's paid from a different account.

Deposits that need a source#

A large deposit needs a closer look before it counts. One big transfer in can raise the average balance and look like revenue or income. That's why lenders separate sales from transfers, loan proceeds and other deposits that aren't revenue.

Lenders that sell mortgages to Fannie Mae follow a written test for large deposits, set out in its Selling Guide. A large deposit is a single deposit over 50% of the total monthly qualifying income for the loan. On a purchase, the test matters when the money is needed for the down payment, closing costs or reserves. The lender must then document that the deposit came from an acceptable source. If it can't, verified funds are reduced by the unsourced amount. On a refinance, the lender doesn't need to document or explain large deposits. It still has to account for any borrowed funds.

At the starting values, $8,000 of monthly income sets the threshold at $4,000. Two of the four unsourced amounts, $5,200 and $12,000, are over it. Without them, verified funds fall from $48,000 to $30,800. That still covers the $30,000 needed, with $800 left over. If you lend outside mortgage, set your own rule for large deposits and write it down before you automate it.

Who uses cash flow data in credit decisions#

Fintech lenders have gone furthest, FinRegLab notes. Banks, mission-based lenders, mortgage lenders and credit score makers now use the data too.

• Online and fintech lenders

  FinRegLab's 2019 study covered six non-bank lenders that used cash flow data for consumer and small business credit. They generally pulled the data through data aggregators.

• Banks

  In the OCC's Project REACh pilot, several national banks used deposit account data to approve first-time credit cards. By October 31, 2023, the pilot had opened over 110,000 accounts. After 12 months, those customers' average FICO score was 680.

• Community lenders

  In a 2025 report, FinRegLab described its interviews with mission-based lenders to small businesses. It uses that term mainly for community development financial institutions (CDFIs) certified by the US Treasury. These lenders use bank data to decide how much to lend, and to see how a business got through months when money was short. Nearly all still let applicants send PDF statements.

• Mortgage lenders

  Fannie Mae's Desktop Underwriter can assess a borrower's cash flow from an asset report. The lender orders this report of the borrower's account data from an authorized supplier. Desktop Underwriter reviews its transaction patterns and balance trends.

• Credit score makers

  Experian, FICO, Nova Credit, Prism and VantageScore offer cash flow scores or attributes, FinRegLab notes. FICO made its new UltraFICO Score, which adds bank data shared through Plaid, generally available in May 2026.

Bank statements or an aggregator feed#

An aggregator is a company that moves account data from a bank to a lender, with the applicant's permission. Plaid, Finicity, MX and Yodlee are examples. The applicant picks their bank from a list, logs in, and chooses which accounts to share.

Feeds are faster where they work. In FinRegLab's 2025 report, mission-based lenders with aggregator connections had quickly made feeds their preferred source. Yet nearly all the mission-based lenders FinRegLab interviewed still accept PDFs. Lenders need PDFs when an applicant won't share access, when a bank shares fewer months than the policy asks for, or when a connection fails.

Accepting both PDFs and feeds has a cost. FinRegLab notes that two systems add work, and that loans underwritten from paper are harder to use later when you test a credit model. Mortgage works differently. For Desktop Underwriter's cash flow assessment, the lender submits an asset report from an authorized supplier, not the statements themselves.

Getting PDF statements into the same rows as a feed#

If you keep accepting PDF statements, make the data from them match the feed. Then every file goes through the same credit rules, whichever way it arrived. Matching the two also leaves you one set of data to test your policy against later.

Docsumo reads the bank statements the applicant provides. It doesn't connect to bank accounts the way aggregators such as Plaid do. Statements arrive by upload, by email to a Docsumo inbox or through the API. Transaction tables that run across pages are joined into one table, with the headers mapped. Docsumo then checks each transaction against the statement's running balance.

Fields Docsumo isn't sure of go to a person, at confidence thresholds you set for each field. The reviewer clicks a field to see the line it came from, and their corrections improve the model. Docsumo supplies the software, not a review team, so your own staff check what it flags.

Cash flow analytics is on the Enterprise plan. On the same plan, cross-document validation can check the statements against the rest of the file, such as the income on the application. A rule your policy needs can run as a workflow step you set up. The step can use an LLM (a large language model) or your own Python code. A step can also change the output so its fields match the ones your feed already uses. Checked data goes to your credit model or loan origination system through the API and webhooks, or out to Excel. The lending page covers the rest of a loan file, such as pay stubs, tax returns and IDs.

• 94%+94%+accuracy on complex bank statements at Grid Finance
• <1 min<1 minper complex statement at Grid Finance

Grid Finance is a Dublin lender to small businesses across Ireland. Its results are in the Grid Finance story.

Our take. Judge any statement tool, ours included, by the corrections your reviewers make on your own files. Run a past month of applications through it, scans and long multi-account files included, and count the rows a person had to fix. Then work out the five measures from the extracted rows, and check them against what your underwriters worked out by hand. One wrong row can move an average daily balance enough to change a decision.

Rules that apply to cash flow data#

Regulators have backed cash flow data, with conditions. The 2019 joint statement said some alternative data may carry no more risk than the data lenders already use. Its example was cash flow data. The statement added that lenders can generally explain their use of cash flow data to the borrower. Two laws may require that explanation, the Equal Credit Opportunity Act and the Fair Credit Reporting Act.

The explanation matters most when you turn an applicant down. Regulation B then requires an adverse action notice. The reasons it gives must be specific, and saying the applicant missed a score cutoff isn't enough. So any cash flow measure that can lead to a decline needs a reason you can state plainly. On fair lending, FinRegLab's 2019 study found the cash flow measures it tested predicted risk consistently across demographic groups. Its results suggest the measures don't create a disparate impact, meaning a rule that applies to everyone but harms protected groups more. The study covered only six lenders, though.

The rules for data feeds aren't settled. The CFPB finalized its personal financial data rights rule in October 2024. The rule would require banks and other data providers to make consumers' account data available to them and to third parties they authorize. In August 2025, the CFPB opened a reconsideration of the rule. In October 2025, a federal court in Kentucky stayed the rule and barred the CFPB from enforcing it until that reconsideration is finished. The rule also wouldn't cover small business accounts, FinRegLab notes. For now, expect to keep accepting PDF statements.

If you already ask applicants for bank statements#

You already hold the data that cash flow underwriting needs. Take one loan product and a few months of past files. Work out the five measures for each file, then set them beside the decision you made and how each loan has been paid since. That shows which measures would have changed a decision, before you change a rule.

Book a demo with a month of your applicants' statements, or start a free trial.

Frequently asked questions#