The application
Everything captured on the loan application: income, employment, assets, liabilities, the property and the loan being requested.
Mortgage & lending
What is automated underwriting?
Automated underwriting is the use of a rules-based system to evaluate a loan file and return a recommendation, usually within seconds. The system reads data the lender submits, compares it against a defined set of rules, and produces both a recommendation and a list of the documentation needed to support it. It does not make the final decision.
What an automated underwriting system evaluates
The inputs are structured data rather than documents. Six things carry most of the weight.
Credit
A credit report pulled as part of the submission, including scores, payment history and open accounts.
Income against obligations
How monthly commitments compare with the income supporting them. Usually the most influential single input.
The equity position
How much is being borrowed relative to what the property is worth, and where the remainder comes from.
Reserves
What remains available after closing. Two otherwise identical files read differently depending on this.
The property and its use
Type, occupancy and location. A primary residence and an investment property are not the same file.
The two systems lenders use
In US mortgage lending, automated underwriting mostly means one of two systems.
Desktop Underwriter, or DU, is Fannie Mae's. Loan Product Advisor, formerly known as Loan Prospector, is Freddie Mac's. Each applies its own rules, and the two do not always reach the same conclusion on the same file.
Lenders often run both. A file that refers through one may accept through the other, which is a normal part of the process rather than a workaround.
The vocabulary differs slightly. DU returns Approve; LPA returns Accept. The concepts correspond, but findings from the two will not read identically.
Which system a file goes through affects who the loan can eventually be sold to, which is why the choice sits with the lender rather than the borrower.
What a recommendation actually is
Three outcomes are possible, and the word people misread is the middle one.
Approve or Accept means the file fits the rules as submitted. Nothing is approved in the sense a borrower would understand. The documentation listed on the findings still has to be produced and verified.
Refer means the system could not reach a recommendation and a person should review the file. This is not a decline. Referred files are underwritten manually every day and many of them are approved.
Caution means risk factors were identified. It usually points to stronger documentation or a different structure rather than an ending.
Underneath the recommendation sits the part that matters operationally: a list of what now has to be documented. That list becomes the conditions, and the conditions become the work.
Automated vs manual underwriting
Both end in the same place. They differ in who reads the file and how much context can be taken into account.
Automated underwriting applies a fixed set of rules to structured data in seconds. It is consistent, it scales, and it treats every file identically. What it cannot do is weigh a circumstance the rules do not describe.
Manual underwriting puts a person in front of the file. It is slower and it costs more, and in exchange it can consider explanation, history and compensating factors that no field captures.
The two are not competitors. Almost every file goes through a system first, and some proportion of them go to a person afterwards. The system is a filter rather than a replacement.
The borrower-facing difference is time. A manually underwritten file takes longer and usually asks for more, because a person is building the picture that the system tried to assemble from data alone.
When a file has to be underwritten manually
Referral is not the only route to manual underwriting, and it is worth separating the cases.
A referred file goes to a person because the system could not reach a recommendation. This is the most common path.
Some loan programs are manually underwritten by design regardless of what any system says. This depends on the program rather than on the borrower.
A borrower with little or no traditional credit history may be underwritten using alternative records instead. Rent, utilities and similar payment histories can substitute where the rules allow.
A file can also be reviewed manually after an accept, if something in the documentation contradicts what was submitted. An underwriter can decline a file the system accepted.
What none of these mean is that something has gone wrong. Manual underwriting is a normal path through the process, not an exception to it.
What underwriting outsourcing is
Underwriting outsourcing is the practice of contracting underwriting work to a third-party provider rather than performing it entirely in house.
It exists because underwriting capacity is difficult to size. Volume moves with rates and with the season, and a team built for a busy quarter is expensive in a quiet one. Outsourced capacity can be added and removed more quickly than headcount.
Arrangements vary. Some lenders outsource overflow only, keeping a core team and sending the excess. Others contract the whole function. A common middle position is to outsource the initial review and keep final decisions in house.
The recurring difficulty is context. An external underwriter reviewing a file has whatever the lender's system contains and nothing else. Where documents are scattered or a condition history lives in email, the outsourced reviewer starts from a worse position than an internal one, and the review takes longer for reasons that have nothing to do with skill.
Regulatory responsibility does not transfer. A lender remains accountable for loans it originates regardless of who reviewed the file, which is why oversight and audit trails matter more in an outsourced arrangement than in an internal one.
What automation has not changed
The recommendation arrives in seconds. The loan still takes weeks, and the reason is worth understanding.
Verification did not become automatic. Every figure the system evaluated was submitted by the lender, and each one still has to be proven with a document that a person reads.
Conditions did not disappear. The findings produce a list, and that list is cleared item by item at whatever pace the documents arrive.
Document collection did not speed up. A borrower still has to locate, produce and send the same records, and the system has no view of whether they have.
Judgment did not move. An underwriter still signs off, and can decline what the system accepted.
Automation compressed the decision from days to seconds. It did not compress what surrounds the decision, which is where the calendar actually goes.
Where the time actually goes
The run
Seconds. This is the part that got fast.
Document collection
Days to weeks. The borrower produces records, and nothing moves until they arrive.
Verification
Days. Each figure is checked against the document supporting it.
Condition clearing
The longest stretch. Items are cleared one at a time and clearing one can create another.
Final sign-off
Hours to days, once everything else is complete.
For lenders: automating around the system
The system is already fast. The gains available now are on either side of it.
Data entry is the first. Figures keyed by hand from documents into an application are the most common source of a mismatch that surfaces weeks later during verification. Extracting them from the document directly removes an entire class of resubmission.
Condition tracking is the second. Findings treated as a document rather than a work list mean items are read once and actioned partially. Turning each line into a tracked request the day the findings land is the single largest operational difference between fast and slow lenders.
Resubmission history is the third. A file on its fourth run with no record of what changed between them is slow for reasons unrelated to the borrower.
Reuse is the fourth. A borrower who verified a document once should not be asked for it again because a different part of the process cannot see it.
None of these touch the underwriting rules. They are all about what surrounds the run.
Answers at a glance
Common questions
Automated underwriting is the use of a rules-based system to evaluate a loan file and return a recommendation within seconds. The system reads data the lender submits, applies a defined set of rules, and lists the documentation needed to support the result. It does not make the final decision.
AUS stands for automated underwriting system. In US mortgage lending it usually refers to Desktop Underwriter, which is Fannie Mae's system, or Loan Product Advisor, which is Freddie Mac's.
No. It produces a recommendation based on the data submitted. Approval comes after the documentation on the findings is collected and verified and an underwriter signs off. An underwriter can decline a file the system accepted.
Automated underwriting applies fixed rules to structured data in seconds and treats every file identically. Manual underwriting puts a person in front of the file, which is slower but can weigh explanation and circumstances the rules do not describe.
Refer means the system could not reach a recommendation and a person should review it. Files are referred for reasons including income that is hard to structure, credit needing context, or an unusual property. It is not a decline.
Underwriting outsourcing is contracting underwriting work to a third-party provider rather than doing it entirely in house. Lenders use it to handle volume that moves with rates and season without carrying permanent headcount.
No. A lender remains accountable for the loans it originates regardless of who reviewed the file, which is why oversight and audit trails matter more in an outsourced arrangement than an internal one.
Yes. A recommendation applies only to the data it evaluated. If income, assets, credit or the loan amount changes, or a figure is corrected, the file is resubmitted and the result can come back different.
No. It evaluates data the lender submits, not documents. Pay stubs, tax returns and bank statements are what the findings ask for afterwards, and they are verified by people.
Because the decision was never the slow part. Collecting documents, verifying each figure against them, and clearing conditions one at a time is where the calendar goes, and none of that was affected by automating the decision.
Related terms
The data going in
A recommendation is only as reliable as the file submitted with it, and most of that file 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 what goes into underwriting matches what came in.
Read about CliQloan →