Chapter 15 — Key Takeaways
Automated Underwriting: DU, Loan Product Advisor, and Reading the Findings
The core claims
1. An AUS evaluates data, not a borrower. It takes the application plus a credit report, checks the loan against a published rulebook, evaluates the borrower against a proprietary risk model, and returns a recommendation, an eligibility assessment, and a document list. It has never seen a paystub.
2. Half the system is public and half is not. The eligibility rules are in the Selling Guide, the Seller/Servicer Guide, and Handbook 4000.1, and you can look every one of them up. The risk model's weights, thresholds, and interactions are not published — anyone who claims to know exactly how DU trades score against DTI is guessing.
3. The recommendation has two halves. The first is about the borrower (Approve / Accept / Refer / Caution). The second is about the loan (Eligible / Ineligible). All four combinations occur.
4. Approve/Ineligible is a structure problem, not a borrower problem. A creditworthy household on a loan that breaks a parameter — loan amount, LTV/CLTV, property type, occupancy, term, program limit. Read the reason, then fix the structure. It is usually a ten-minute conversation, not a decline.
5. Refer and Caution are routing, not rejection. Manual underwriting, restructuring, the other agency's system, a different program — all still open. Only a lender denies a loan.
6. The verification messages are the document list, and they are the point of the report. They are generated from your own data: variable income produces VOE messages, a gift produces gift-letter messages, a rental history produces a verification message. Convert them into a dated request the day you get them. They are a floor — overlays and the underwriter add to them; nothing subtracts.
7. Re-running changes the answer only if the inputs change. The system is deterministic. Identical data on the same engine version returns an identical result. Two things can move a recommendation without you touching the data: a system version release and a new or refreshed credit report.
8. Occupancy and property type are facts, not levers. Changing them to obtain a different result is misrepresentation, not restructuring (Chapter 27).
9. Garbage in. A mistyped income, a wrong occupancy, an omitted debt, a miscoded property type produce a confident, well-formatted, worthless recommendation — and the system will never tell you. A findings report is only as good as the data underneath it, and the loan officer owns that data.
10. The AUS does not decide. It returns a recommendation against published rules. The lender decides, subject to its overlays (Ch.14 §14.7), its underwriter's review of the actual documents, and the law. "Fannie Mae denied my borrower" is never the correct sentence.
The rules of thumb
| Rule | What it means |
|---|---|
| Read the report backwards | pages 2–3 (what you typed) → messages (the work) → eligibility (the reason) → recommendation (last) |
| Reconcile five figures, every time | income · housing expense · total obligations · verified assets · reserves |
| First word = the borrower. Second word = the loan. | Approve/Ineligible decoded in four seconds |
| A Refer is a route, not a result | manual UW · restructure · other agency · other program · find the typo |
| Look for the data error first | it is free, and it is embarrassing to manually underwrite a typo |
| Sequence by turn time, not list order | VOEs and donor evidence go out on day 1; the insurance binder waits |
| Change the fact in the world, then in the system | never the reverse |
The vocabulary map
| Borrower half | Loan half | Output document | |
|---|---|---|---|
| DU (Fannie Mae) | Approve · Refer | Eligible · Ineligible | Underwriting Findings report |
| LPA (Freddie Mac) | Accept · Caution | purchase eligibility reported | Feedback Certificate |
| TOTAL (FHA) | Accept · Refer | supplied by the AUS it runs through | appears within the AUS findings |
TOTAL is a scorecard, not a standalone system. It runs through DU or LPA.
The numbers from this chapter
| Linden Street, day 6 | Approve/Eligible · 28.89% housing · 42.66% back-end · 706 · 4.16 months reserves |
| Linden Street, day 44 | \$4,479.72 + \$611.00 = \$5,090.72 → 48.48% |
| Linden Street, day 47 re-run | debt removed → 42.66% again; reserves \$7,423.66 = 2.45 months; Approve/Eligible held |
| Harlow Street | \$1,721.57 ÷ \$4,150.00 = 41.48% front · \$2,116.57 ÷ \$4,150.00 = 51.00% back |
| The 31/43 benchmark on Harlow | max housing \$1,286.50** · max obligations **\$1,784.50 → over by \$435.07** and **\$332.07 |
| The typo that flatters | income entered \$12,300 → back-end 36.42% instead of 42.66% |
| The typo that costs | income entered \$10,050 → back-end 44.57% instead of 42.66% |
| 2% down on Linden Street | loan \$377,300 · LTV **98.00%** → Ineligible; **\$3,850** more down reaches 97% |
All illustrative; the Linden Street, Harlow Street, and Fulton Avenue files are constructed. Agency limits, ratio maximums, and waiver criteria change — verify current values at the source.
Key terms
automated underwriting system (AUS) · Desktop Underwriter (DU) · Loan Product Advisor (LPA) · TOTAL Scorecard · findings report · Approve/Eligible · Approve/Ineligible · Refer · Caution · verification messages · appraisal waiver / value acceptance · re-run · data integrity
What you should be able to do Monday morning
Open a live findings report, reconcile five figures on the analysis page against your own worksheet in ninety seconds, decode the two-word recommendation without hesitating, convert the verification messages into a dated document request with the third-party items ordered first — and, if it came back Ineligible, tell the borrower what it will cost to fix the structure before you get off the phone.