Automated Underwriting Explained: What DU and LPA Findings Actually Mean

The short answer: the engine recommends, a person decides

No, a computer does not approve your mortgage. An automated underwriting system returns a recommendation based on the data entered about you, and a human underwriter still verifies the documents, clears the conditions, and signs off on the file.

The engines you will hear named are DU and LPA. Neither one issues a loan approval. Both issue a finding.

Here is the mechanism end to end:

• Your loan file goes in as data: income, assets, credit, property, loan terms.
• The engine returns a finding, such as DU’s Approve/Eligible or Refer with Caution, or LPA’s Accept or Caution risk class.
• The same finding lists the documents the lender must obtain to support every figure it relied on.
• A human underwriter compares those documents to the data that produced the finding.
• Conditions are issued, cleared, and re-checked by that human before the file reaches clear to close.

Our take: borrowers hear “the system approved us” and assume the file is finished. The finding is the starting gun, not the finish line.

What an automated underwriting system actually does

It scores data, not documents

An automated underwriting system (AUS) takes the loan application as structured data, including income figures, asset balances, the credit report, property type, occupancy, loan amount, and term. It runs that data against the investor’s risk model and its product eligibility rules.

It does not read your bank statements. It reads what was typed about your bank statements. Fannie Mae puts the responsibility squarely on the lender to confirm the accuracy of the data it submits, and to make sure nothing was left out that might have changed the recommendation had it been known.

That single distinction explains most of what confuses borrowers later in the process. The engine is scoring the same four Cs of mortgage approval a human underwriter would apply: credit, capacity, capital, and collateral. It simply does it from data someone else keyed in.

What comes back is a finding, not a decision

The output is a multi-page findings report. It contains the recommendation, the eligibility result, the risk factors the model weighed, and the documentation requirements attached to each figure. In DU, those requirements appear in the findings report under Verification Messages and Approval Conditions, and the Selling Guide treats them as the minimum: a lender may always require more documentation when the file warrants it.

That documentation list is the part that actually drives what happens next. It is where a file either moves or stalls.

Our take: the borrowers who feel blindsided by underwriting are often the ones who never saw the findings documentation list that was generated on day one.

DU, LPA, and the vocabulary problem

Different engines, different dialects

Desktop Underwriter is Fannie Mae’s engine, and its rules live in the Fannie Mae Selling Guide. Loan Product Advisor is Freddie Mac’s engine, documented in the Freddie Mac Single-Family Seller/Servicer Guide.

Worth saying plainly, because the confusion is constant: Fannie Mae and Freddie Mac are government-sponsored enterprises, or GSEs. They are not government agencies. FHA, VA, and USDA are the government programs; Fannie and Freddie are chartered private companies that buy loans on the secondary market.

The GSE engines use different words for the same idea, and neither word is a final approval.

The vocabulary decoded

Finding	Engine	What it means	What happens next
Approve/Eligible	DU (Fannie Mae)	The data as entered met the credit risk standard and the eligibility criteria	Underwriter verifies documents against that data
Approve/Ineligible	DU (Fannie Mae)	Credit risk standard satisfied, but an eligibility criterion is not	Correct or restructure the file and resubmit, or underwrite manually
Refer with Caution	DU (Fannie Mae)	The file does not meet Fannie Mae’s credit risk standards for DU loans	Lender may underwrite manually where the product allows it
Out of Scope	DU (Fannie Mae)	DU does not evaluate this loan or transaction type	Underwrite manually under the applicable guideline
Accept	LPA (Freddie Mac)	Freddie Mac’s risk class for a file its model accepts	Same path: human verification of the documents
Caution	LPA (Freddie Mac)	Freddie Mac’s other risk class; not an Accept	Seller manually underwrites for the loan to be eligible for sale

Notice that DU pairs two separate results, a credit risk recommendation and an eligibility assessment, which is why Approve/Ineligible exists as a recommendation of its own. Fannie Mae describes that outcome as satisfying its credit risk standards but not its loan eligibility criteria: the risk profile cleared, but something structural, such as loan amount, occupancy, or property type, does not fit. Loan Product Advisor reports a risk class and a purchase eligibility result separately for the same reason.

What a Refer actually triggers

Refer is a routing instruction, not a denial

A Refer routes the file to manual underwriting, where a person reads the file against the written guideline. That is the whole meaning of the word.

Under the Fannie Mae Selling Guide, a loan casefile that receives a Refer with Caution has not met Fannie Mae’s credit risk standards for DU loans, and the lender may manually underwrite it in accordance with the Selling Guide where the loan product or transaction otherwise allows manually underwritten loans. Freddie Mac applies the same principle to a Caution risk class: the Seller manually underwrites the mortgage for it to be eligible for sale.

Government programs work on the same principle. HUD states that the TOTAL Mortgage Scorecard is accessed through an automated underwriting system and is not an automated underwriting system itself, and that an Accept means FHA will insure without a manual underwriting review, while a Refer means the loan must be underwritten by an FHA Direct Endorsement underwriter. USDA’s Guaranteed Underwriting System renders Accept, Refer, or Refer with Caution under HB-1-3555, Chapter 5, Origination and Underwriting Overview, and a Refer or Refer with Caution goes to the lender’s underwriter for further review.

VA files have their own version of this conversation, and a Refer on a VA loan is not the end of the loan. VA publishes its credit standards in the VA Lender’s Handbook (VA Pamphlet 26-7); how a particular lender routes a VA file through an automated system varies by lender, so confirm the current process with yours.

What manual review changes

Manual underwriting generally means tighter documentation, closer attention to reserves and housing payment history, and a person weighing compensating factors instead of a model. The specific overlays a lender layers on top of the agency guideline vary by lender, so ask yours which ones apply before you assume the worst.

What a refer-type finding does not mean: that you have been denied, that the file has to start over, or that a human underwriter cannot approve it.

From file to finding to decision

The sequence, in order

1. Step one: the file is assembled and submitted as data, with the supporting documents behind it.
2. Step two: the AUS returns a finding and a documentation list.
3. A human underwriter picks the file up in queue order and reads the documents against the data.
4. Conditions are issued, worked, and re-reviewed, often more than once.
5. The underwriter clears the last condition and the file moves to clear to close.

The engine’s finding is step two of a process a human underwriter finishes.

Step one is the half you actually control. If the income figure keyed into the application does not match the pay stubs, the engine still returns a clean-looking finding, because it scored the number it was given. The mismatch surfaces later, as a condition, on the underwriter’s clock instead of yours.

Why re-running the AUS is normal, not a red flag

Runs are fast and repeatable

Here is the broker-side observation most borrowers never hear: submitting the same file to the engine more than once is routine work, not a warning sign. Fannie Mae’s own guidance contemplates resubmission, including reviewing the loan data for accuracy, updating it, and resubmitting the casefile for an updated recommendation.

Restructuring and re-running is how a file gets optimized. Common reasons a run gets repeated:

• A down payment or loan amount changes and the eligibility result changes with it.
• A debt gets paid off or documented differently, changing the capacity picture.
• An income figure is corrected after the documents come in.
• The product changes, for example from a conventional structure to an FHA structure.
• The initial run returned an ineligible result on a structural item that can simply be fixed.

Any per-run cost sits with the lender and varies by lender and by engine, so we will not quote a figure.

What a re-run does and does not do

A re-run does not by itself pull new credit; resubmitting the file and ordering a new credit report are separate events. Practices vary by lender, so ask your loan officer to confirm what a re-run does on your file and to tell you if a new credit pull is actually needed.

Our take: if a loan officer tells you they cannot re-run the file to test a different structure, ask why. Testing structures before submission is ordinary craft, not a favor.

What the engine cannot see

An automated underwriting system evaluates patterns in data. It has no way to evaluate context, and context is exactly what most real files are made of.

Things a model cannot weigh, and a human can:

• Why a deposit that looks irregular is actually a documented, sourced transfer.
• Why a gap in employment happened and why the new position is stronger than the old one.
• Whether a self-employed borrower’s business is stable in a way the tax returns understate.
• Whether a derogatory item was a one-time event with a paper trail behind it.

This is why the documentation list matters more than the recommendation. The finding tells the underwriter what to prove; your file either proves it on the first pass or generates conditions.

What the finding means for your timeline

Two clocks, not one

A finding comes back quickly, but how long underwriting takes is governed by two different clocks. The first is underwriter turn time, which is how long the file sits before a human opens it. The second is condition-clearing time, which is how long you and your loan officer take to satisfy what the underwriter asks for.

Those two clocks are not the same, and confusing them is why timelines feel unpredictable. Turn time is queue position at that specific lender in that specific week. Condition clearing is file completeness, which is the actionable half. Both vary by lender and by file, so treat any timeline you are given as general and confirm current expectations with your loan officer.

Turn time also shifts by program: the underwriting timeline by loan type differs because FHA, VA, USDA, and conventional files run different checklists and different scorecards.

Where the finding fits

A file that clears the engine typically reaches conditional approval first, with a list of items to satisfy. That is a normal outcome, not a setback, and it descends directly from the documentation list the AUS generated on day one.

Because we are a broker, the submission choice is ours to make: the same complete file can go to different wholesale lenders whose queues and overlays differ. That is a structural fact about how brokers work, and it is not a promise about speed, approval, or pricing. No one can promise you a queue position.

How to give the engine a clean file on day one

Completeness on day one is the single largest variable you control. A practical checklist:

• Give complete statements, all pages, including the pages that look blank.
• Provide the most recent pay stubs and W-2s or full tax returns, not summaries or screenshots.
• Disclose every debt, including the ones you plan to pay off, so the entered data matches the credit report.
• Explain unusual deposits up front rather than waiting to be asked.
• Confirm the exact spelling of your name, employer, and address as they appear on documents.
• Ask your loan officer to walk you through the findings documentation list the day it is generated.

Every one of those items exists to make the data you were scored on match the documents that prove it. When those two agree, the underwriter clears conditions instead of chasing them.

Frequently asked questions

Does an automated underwriting system approve my loan?

No. It returns a recommendation and a documentation list. A human underwriter at the lender verifies the documents, issues and clears conditions, and makes the actual credit decision. Fannie Mae’s Selling Guide is explicit that lenders must employ prudent underwriting judgment in assessing whether a loan casefile should be approved.

What is the difference between DU and LPA?

DU is Desktop Underwriter, Fannie Mae’s engine; LPA is Loan Product Advisor, Freddie Mac’s engine. They use different vocabularies for similar outcomes, DU returning recommendations such as Approve/Eligible and Refer with Caution, and LPA returning a risk class of Accept or Caution alongside a separate purchase eligibility result. Both are followed by human underwriting.

Is a Refer a denial?

No. A refer-type finding means the engine did not return an automated approval and the file is routed to a person for manual underwriting against the written guideline. In DU the wording is Refer with Caution; in LPA it is a Caution risk class; under FHA’s TOTAL Scorecard it is a Refer. It is a routing instruction, not a decline.

Does re-running the AUS hurt my credit or look suspicious?

Re-running a file is routine and is how loan officers test different structures. Fannie Mae’s guidance itself contemplates correcting data and resubmitting for an updated recommendation. A re-run does not by itself pull new credit, and a new credit pull is a separate event. Practices vary by lender, so ask your loan officer to confirm.

Are Fannie Mae and Freddie Mac government agencies?

No. They are government-sponsored enterprises, chartered private companies that buy loans on the secondary market. FHA, VA, and USDA are the government programs, and each has its own scorecard or underwriting requirements.

Can a loan get an automated approval and still be denied later?

Yes. The finding is based on the data entered. If the documents do not support that data, or if something changes during the process such as employment, debt, or the property itself, the human underwriter can reach a different conclusion. This is why file completeness on day one matters more than the initial recommendation.

Do FHA, VA, and USDA loans use the same systems?

Not exactly. FHA files are scored by the TOTAL Mortgage Scorecard, which HUD says is accessed through an automated underwriting system and is not one itself; an Accept means FHA will insure without a manual underwriting review, and a Refer must be underwritten by an FHA Direct Endorsement underwriter. USDA uses its own Guaranteed Underwriting System, which renders Accept, Refer, or Refer with Caution. VA publishes its credit standards in the VA Lender’s Handbook, and the automated path varies by lender. In each case, a non-accept result routes the file to a person rather than ending it.

719 Lending Inc., NMLS #1601989 · Equal Housing Opportunity

719 Lending Inc. is not affiliated with or endorsed by HUD, FHA, VA, USDA, CHFA, or any government agency.

Last updated: August 2026