Back to resources
Credit

Chief credit officer, underwriting manager, compliance, BSA/AML, CIO

Prepare cleaner underwriting files before the workday begins

Can we raise file quality without giving the model credit authority?

Morning queue · this walkthrough
9 ready6 exceptions2 held
Last night’s history test · cyan ships
Always count overtimeFailed history
Show DTI both waysPassed
Route large depositsPassed
Dual DTI + deposit mergeMerged
Adaptive income boundaryWinner

Counts are from the walkthrough file below — a sample Monday morning, not a promised customer result.

Overnight Engine gathers the file, verifies what can be verified, calculates DTI both ways when overtime is in question, and parks exceptions on the underwriter’s desk. It does not recommend approve, decline, or counter. Credit authority stays with the people who already hold it.

Overnight Engine

On-premMaya Chen · Underwriter8:02 AM
9Ready
6Exceptions
2Held

Nine files are complete enough to underwrite. Six still have document, income, or exception work. No file carries a credit recommendation.

nCinoExperianSalesforceDocuSignBank statements

UND-10442 · Overtime tagged recurring

Needs attention

Exceptions

UND-10418 · Employer name mismatch

Review

Verify

UND-10390 · File complete for Maya

Ready

File

UND-10355 · $9,400 deposit needs source

Review

Exceptions

Meridian Partners, Monday 8:02 a.m. UND-10442 shows DTI at 41.2% with overtime and 47.1% without. Maya Chen decides whether OT is recurring. The engine will not.

What enters the system

UND-10442 is a purchase file: Sofia Alvarez, 641 W Concord, 20% down.

  1. 01

    Application and docs

    1003, paystubs, W-2s, bank statements, ID, and the purchase contract from nCino.

  2. 02

    Verification services

    Employer / SOS lookup, bureau pull for the file, and any VOE the shop already uses.

  3. 03

    Policy

    Meridian’s written income and deposit guidelines — including that 45% DTI is a guideline, not a cutoff the machine may apply.

What happens overnight

  1. 01

    Completeness

    Required documents are checked in. Missing pages and stale stubs are requested on the approved borrower template.

  2. 02

    Verification

    Employer legal name is compared to the application. “Northline Mfg” vs “Northline Manufacturing Inc.” is an exception, not a fail.

  3. 03

    Income and DTI

    Base and overtime are extracted separately. DTI is shown with and without OT when OT appears on a configured fraction of stubs.

  4. 04

    Exception detection

    Large deposits without a source note, NSF patterns, and occupancy inconsistencies are listed. They are not fraud conclusions.

  5. 05

    File, not decision

    Complete files are placed on the underwriter’s morning desk. Incomplete files stay in exceptions. No credit recommendation is written to Salesforce or nCino.

What the underwriter sees in the morning

Maya Chen has nine files she can underwrite and six that still need work. UND-10442 is in the second group because overtime is tagged recurring and a $9,400 deposit has no source note.

She can open the stub, the dual DTI worksheet, the statement, and the bureau pull. The bureau is attached as evidence for her. The engine does not turn 734 FICO and 41.2% DTI into an approve.

Human authority and lending governance

Can prepare

  • Assemble the document file and completeness list
  • Extract income components and compute DTI scenarios
  • List exceptions with source evidence
  • Stage complete files for the assigned underwriter

Can execute

  • Request a missing document on an approved template
  • Write file-completeness status into nCino
  • Attach the worksheet and exception list to the opportunity

Always needs approval

  • Any credit recommendation or risk rating
  • Approve, decline, or counter
  • Policy exceptions on DTI, LTV, or occupancy
  • Adverse-action notices
  • Closing or funding instructions

How a correction becomes a validated improvement

If Maya excludes overtime, UND-10442 becomes a labeled example: OT on three of four stubs is not enough in this shop. If she includes it, the opposite example is stored — with her reason.

A hypothesis that always treats overtime as recurring fails Meridian history and is discarded. Dual DTI display and large-deposit routing pass. The winning model changes how the file is prepared. It does not change who may lend.

What a 30-day pilot measures

One product (purchase, in this walkthrough). Credit performance is out of scope for the pilot.

MetricHow it is measuredPilot target
Time to a complete fileApplication in → underwriter can decide without chasing basicsMost standard files ready next business morning
Underwriter prep timeMinutes spent assembling vs. decidingAssembly down; decision quality is audited separately
Exception precision% of exceptions Maya agrees belonged on the deskHigh enough that the desk is trusted
Unauthorized recommendationsApprove/decline/counter written by the engineZero. Stop-the-pilot metric. Fair-lending review included.

Implementation

Systems
nCino or Encompass, bureau, CRM, e-sign, and the shop’s VOE / SOS tools.
Data required
Closed files with underwriter conditions, OT decisions, and deposit-source notes — the labeled history.
Deployment
Hardware in the institution. Shadow on historical files, supervised prep, then LOS status writes. Credit memo drafts stay off unless compliance signs a later workflow.
Who operates it
Underwriters own every credit action. Processors still collect. Compliance owns adverse action and fair-lending review. We do not sit on committee.

Security and deployment architecture

Applications, stubs, and bureau data stay inside the institution. The engine calls nCino and Experian the way your processors already do, from your network.

Lending-specific governance — ECOA, FCRA, adverse action, model risk — is why this workflow stops at the file. If someone asks the model “should we lend,” the configured answer is that it will not.

More workflow briefs

Next step

Request an underwriting-file assessment

Document prep and exception detection on one product. No automated credit recommendation.