Mortgage & lending

What is Desktop Underwriter?

Desktop Underwriter, usually shortened to DU, is Fannie Mae's automated underwriting system. A lender submits a loan file to it and receives a recommendation within seconds, along with a list of the documentation needed to support that recommendation. DU does not approve a loan. It tells the lender what the file looks like against Fannie Mae's rules and what still has to be proven.

What Desktop Underwriter looks at

DU reads a structured version of the loan application together with credit data. Six things carry most of the weight.

The application data

Everything entered on the Form 1003: income, employment, assets, liabilities, the property, and the loan being requested.

Credit

A credit report pulled as part of the submission, including scores, payment history, and the accounts and balances behind them.

The relationship between income and debt

How the monthly obligations compare with the income supporting them. This is usually the single most influential input.

The equity position

How much is being borrowed against what the property is worth, and where the rest is coming from.

Reserves

What the borrower still holds after closing. A file with reserves reads differently from an identical file without them.

The property itself

Type, occupancy and location. A primary residence and an investment property with the same numbers are not the same file.

What the findings actually say

The output is called a findings report, and it is longer than people expect.

The recommendation sits at the top and is the part everyone reads. Underneath it is the more useful part: a list of what the lender now has to document. If DU accepted a stated income figure, the findings say what evidence is needed to support it.

That list is the working document for the rest of the file. Each line becomes a condition, and each condition becomes a request to the borrower.

The findings also record what DU relied on. If a figure changes later, and it often does, the file goes back through and the findings can change with it. A recommendation is tied to the data it ran against, not to the borrower.

The practical read: the recommendation tells you where you stand today, and the documentation list tells you how much work is left.

Approve, Refer and Caution explained

Three outcomes, and the middle one causes the most confusion.

Approve or Accept means the file fits the rules as submitted. Nothing is approved yet in the sense a borrower would understand. The documentation on the findings still has to be produced and verified.

Refer means DU could not reach a recommendation and a human underwriter should look at it. This is not a decline. Files get referred for reasons that have nothing to do with the borrower's strength, including situations the system is not built to assess. A referred file can be manually underwritten and approved.

Caution, or Refer with Caution, means the system identified risk factors. It is the weakest of the three, and it usually means either stronger documentation or a different loan structure.

The word people misread is Refer. Borrowers hear it as rejection. In practice it means the decision moved from a system to a person.

A loan file is submitted to the system, which returns one of three recommendations: Approve or Accept, Refer, or Caution. Each leads to a different next step, and a human underwriter makes the final decision in every case.
What an automated underwriting system returns

Where DU sits in a loan file

Third, and earlier than most borrowers realize.

A file goes through DU once the application is complete and credit has been pulled, usually well before the documents have all arrived. That is deliberate: running it early tells the lender what to collect rather than collecting everything and hoping.

What follows is condition clearing. Each item on the findings has to be satisfied, verified and signed off. This is where most of the calendar goes, and it is the part that stalls when a document is late.

Clear to close comes after the conditions, not after DU. A borrower told their file was approved by DU in week one and still waiting in week six has not had anything go wrong. They are in the normal middle of the process.

Six stages of a mortgage file: application, documents collected, automated underwriting, conditions cleared, clear to close, and funded. Automated underwriting is the third stage and is not the final approval.
Where automated underwriting sits in a loan file

Desktop Underwriter vs Loan Product Advisor

Two systems, same job, different owner.

Desktop Underwriter is Fannie Mae's. Loan Product Advisor, formerly Loan Prospector, is Freddie Mac's. Each evaluates a file against its own rules, and the two do not always agree.

Lenders often run both. A file that refers through one may accept through the other, which is a legitimate reason to submit twice rather than a sign that something is being worked around.

The vocabulary differs slightly. DU returns Approve. LPA returns Accept. The concepts line up, but the wording on the findings will not match if you are comparing the two side by side.

Which system a file goes through affects who the loan can eventually be sold to, which is why the choice belongs to the lender rather than the borrower.

What automated underwriting does not decide

The most common misunderstanding about DU is the size of its job.

It does not verify anything. Every figure it evaluates is one the lender submitted. If income was entered wrong, DU produces a confident recommendation on wrong data, and the error surfaces later when documents contradict it.

It does not read documents. Pay stubs, tax returns and bank statements are not inputs to the run. They are what the findings ask for afterwards.

It does not make the final decision. A recommendation is a recommendation. An underwriter approves the loan, and can decline a file DU accepted.

It does not consider anything outside its rules. A borrower with an unusual but entirely reasonable situation may refer simply because there is no pathway through the system for it.

This is the honest summary: DU is fast and consistent at applying a defined set of rules to submitted data. For the broader category, see automated underwriting. Everything about whether that data is true happens elsewhere.

Why a file gets referred

Income that is hard to structure

Self-employment, commission, bonus, or several part-time sources. The income may be perfectly sufficient and still not fit a field.

Credit that needs context

A recent event on the report, a thin file, or an explanation the system cannot evaluate.

Data that does not line up

A figure entered one way in one place and differently in another. Often a keying issue rather than a borrower issue.

An unusual property

Non-standard construction, mixed use, or a project the system has questions about.

A borrower profile outside the common case

Non-occupant co-borrowers, recent relocation, or foreign income. Common in reality, less common in the rules.

Documentation gaps at submission

A file submitted before it is ready gets a recommendation based on incomplete data, which often means a referral.

What happens after an Approve

The findings become a checklist, and the checklist becomes the loan.

The lender works down the documentation list and collects what each line requires. Income documents, asset statements, employment verification, anything tied to the property.

Each item is verified against what was submitted. This is the step where the file either holds together or does not. A borrower who stated one income and documents another has a file that no longer matches the run.

Conditions get cleared one at a time, and new ones can appear. A document that raises a question generates a condition of its own, which is normal rather than a sign of trouble.

When everything is cleared, an underwriter signs off and the file is clear to close. That is the approval a borrower is actually waiting for.

For lenders: what slows a DU run down

The delay is rarely the run. It is what happens around it.

Files submitted early with placeholder data get a recommendation nobody can rely on, and the resubmission cycle costs more than waiting would have.

Data entered by hand from documents is where mismatches begin. A figure keyed from a pay stub into the 1003 is the most common source of a discrepancy that surfaces two weeks later during verification.

Findings treated as a document rather than a work list mean conditions get read once, actioned partially, and revisited. The lenders who move fastest turn each line into a tracked request the day the findings land.

Resubmissions multiply when nobody records why the last one changed. A file on its fourth run with no history of what moved between them is slow for reasons that have nothing to do with the borrower.

Answers at a glance

Common questions

The data behind the run

A recommendation is only as good as the data submitted with it, and most of that data is keyed by hand from documents the borrower already sent. CliQloan, one of the AmitaSoft platforms, reads those documents and extracts the figures directly, so the file that goes into underwriting matches the file that came in.

Read about CliQloan →