Why mortgage W-2s come back high risk twice as often as bank statements

TL;DR: Ocrolus production data from Aug. 21 to Sept. 22, 2026, found that 11.4% of the mortgage W-2s Ocrolus processes come back high risk, roughly double the rate for bank statements and pay stubs over the same period. W-2 is one of the document types Ocrolus Detect fully screens, receiving both a content check and Resistant AI forensic analysis of file metadata and pixels. Genway Home Mortgage and JMAC Lending, two wholesale mortgage lenders, use Detect to screen documents their underwriters used to check by eye.

Ocrolus production data from Aug. 21 to Sept. 22, 2026, found that 11.4% of the mortgage W-2s Ocrolus processes come back high risk. That is roughly double the rate for bank statements and pay stubs over the same period. A W-2 is a single page from an employer, issued once a year on a standard IRS form. A bank statement runs for pages and documents months of transactions.

What Ocrolus data shows about W-2 risk

Ocrolus processes mortgage income and asset documents for lenders across Encompass, MeridianLink, the Ocrolus dashboard and the API. Of the W-2s scored by Ocrolus Detect between Aug. 21 and Sept. 22, 2026, 11.4% came back high risk, about twice the rate for bank statements and pay stubs. All three document types were scored by Ocrolus Detect in the same production environment over the same window, so the gap comes from the documents themselves.

Why W-2s get the same review as bank statements

Detect runs three layers. Ocrolus checks that the math adds up, balances, totals and dates, on supported document types. Resistant AI forensic analysis checks file metadata and the pixels of the page, including signs of AI generation, on any document type. Supported documents get all three layers. W-2 is one of the document types Detect’s content check supports, alongside bank statements and pay stubs, so a submitted W-2 gets the full three-layer review, the content check plus both of Resistant AI’s forensic layers. A document that reads as consistent on screen can still fail at the file level: an edited figure, a document built from a known template, a pixel pattern invisible to a reviewer. None of that depends on how many pages the document has.

Resistant AI’s forensic analysis can also flag when a single document is built from a known template, such as a publicly available pay stub template site, and identify which template it matches. That signal works document by document today. It does not yet link a flagged document to other files, loans or borrowers across applications, a capability Ocrolus has said is coming in early 2027.

What that looks like in wholesale and broker files

Genway Home Mortgage, a mortgage lender operating in wholesale and delegated channels, used to catch alterations by eye. “Detecting document fraud was manual; the underwriter looked at the document to see if there were any altered fonts, any altered balances and if the balances didn’t add up,” said Chinelo Nwanze, VP, strategic initiatives at Genway Home Mortgage.

Nwanze’s confidence in the product predates Genway. At a previous employer, she sat on a team evaluating vendors and tested them by submitting documents she had altered herself. “I duplicated and manipulated documents to see if they caught the fraud, and those third parties didn’t,” she said. “I uploaded the same documents to Ocrolus and they caught it right away.”

“When it comes to document fraud detection, Ocrolus is the winner,” Nwanze said.

That gap matters most where documents arrive from outside a lender’s own walls. JMAC Lending, another wholesale lender, stopped nearly $2 million in fraudulent loan volume after moving from selective manual review to Detect running on 100% of its files. “We have zero tolerance for fraud,” said Dana Reminder, director of underwriting at JMAC. “Ocrolus Detect is a partner to our underwriters to be their second set of eyes. You’re not replacing them. You’re enhancing them.”

Files that arrive already merged by a broker can score differently than files a lender originates directly, and that difference is expected rather than a flaw. The signals that matter are the ones that catch edited data inside the document itself, not the structure of how a file was assembled before it reached the lender.

Altered documents do not typically surface during underwriting. An altered pay stub or W-2 that looks clean on screen tends to surface later, after a loan has sold, when it is harder and more costly to resolve. Screening every W-2 with the same signal-level check as every bank statement, on arrival rather than on suspicion, puts that review where a flagged file can still be corrected. See how Ocrolus Detect screens every bank statement, pay stub and W-2.

Key takeaways

• Ocrolus production data from Aug. 21 to Sept. 22, 2026, shows 11.4% of mortgage W-2s processed come back high risk, roughly double the rate for bank statements and pay stubs over the same period.
• W-2 is one of the document types on Ocrolus Detect’s supported list, so it receives the same three-layer review as a bank statement: a content check plus Resistant AI’s file metadata and pixel-level forensic analysis.
• Resistant AI’s forensic layers, including checks for signs of AI generation and known document templates, run on any document type Ocrolus processes, not only the ones on Ocrolus’s own supported list.
• Genway Home Mortgage, a wholesale and delegated-channel lender, moved from manually checking fonts and balances by eye to Detect flagging altered documents before a file moves forward.
• JMAC Lending, also a wholesale lender, stopped nearly $2 million in fraudulent loan volume after running Detect on 100% of its files instead of relying on selective manual review.

FAQs

What percentage of mortgage W-2s come back high risk?

Ocrolus production data from Aug. 21 to Sept. 22, 2026, found that 11.4% of the mortgage W-2s Ocrolus processes came back high risk, roughly double the rate for bank statements and pay stubs over the same period.

Does Ocrolus Detect check W-2s the same way it checks bank statements?

Yes. W-2 is one of the document types Detect’s content check supports, alongside bank statements and pay stubs. Supported documents get all three of Detect’s layers: a content check for math that does not add up, plus Resistant AI forensic analysis of file metadata and the pixels of the page, including signs of AI generation.

What is Resistant AI’s role in Ocrolus Detect?

Resistant AI performs forensic analysis of file metadata and the pixels of a document’s page, including signs of AI generation, on any document type Ocrolus processes. Ocrolus’s own content check, which verifies that balances, totals and dates add up, runs on a defined list of supported document types. Supported documents, including W-2s, get both.

Why do broker-originated mortgage files sometimes score differently in fraud detection?

Files that arrive already merged by a broker can score differently than files a lender originates directly, which is expected rather than a flaw. The signals that matter most are the ones that catch edited data inside the document itself, not the structure of how the file was assembled before it reached the lender.

Can Ocrolus Detect link a fraud signal on one document to other files or loans?

Not yet. Resistant AI’s forensic analysis can flag a single document built from a known template, such as a public pay stub template site, and identify which template it matches. Linking a flagged document to other files, loans or borrowers across applications is a capability Ocrolus has said is coming in early 2027.