For non-QM and bank-statement lenders

The loan file that can
answer for itself
two years later.

LoanCite extracts every figure, cites its source, and runs the rules — so when a note buyer or regulator asks why, you hand them one record instead of excavating inboxes.

Book a demo See how it works
Scroll to explore
0%
Faster from raw docs
to underwriter queue
0%
Figure accuracy
across bank-stmt, P&L, 1099 extraction
0%
Lower ops cost
per funded loan vs manual processing
Weeks
not quarters
Typical time to first pilot
rules configured, not coded
How it works

From raw documents to a defensible decision.

01 / Upload
Drop the file. It classifies itself.
Bank statements, P&L, CPA letters — LoanCite recognizes them on ingest. No manual sorting, no template mapping.
02 / Extract
Every number gets a citation.
$13,828/mo comes from 12-mo bank statements, pages 3–14, at a 50% expense factor. Every figure works that way — source, method, result.
03 / Report card
A report card on every rule.
Not a summary score — a verdict on each rule: pass, advisory, or blocking. Nothing bundled, nothing hidden. Your underwriter sees exactly what needs attention.
04 / Binder
One record from first doc to final call.
Export the binder anytime — cited figures, guideline version, exception resolutions, audit hash. Two years from now, you're not excavating inboxes.
LN-2026-04417 · upload
Document ingestion
bank_stmt_12mo.pdf
2.4 MB · Bank Statement
Parsed
cpa_pl_2025.pdf
890 KB · P&L Statement
Parsed
cpa_letter.pdf
210 KB · CPA Letter
Parsing…
Zero manual sorting required
LN-2026-04417 · extraction
Cited figures
Qualifying monthly income
12-mo bank stmt · p.3–14
$13,828
Avg monthly deposits
Statements · 12 mo
$27,656
Expense factor
CPA P&L · 2025
50%
Total liquid reserves
Stmt · p.22
$48,200
Every figure cited to its source page
LN-2026-04417 · rule check
Rule-by-rule report card
ATR documented · income, assets, obligations Pass
Bank-statement method · 12 mo deposits Pass
Reserves below 6 months Advisory
CPA letter dated 118 days Blocking
2 exceptions flagged for underwriter review
LN-2026-04417 · binder export
Decision-ready binder
Every figure with its cited source 42 figures
Guideline version applied matrix v4.2
Each exception and how it cleared 2 resolved
Decision audit hash sha256:9f2c…
sha256:9f2c4e1b…a71b · rules v4.2
Why the file isn't enough

Faster isn't the problem.
The problem is two years later.

Buyback demands are figure-level

When an investor samples your loan, they don't ask about the process — they ask about the number on page 7. "We ran it through AI" is not a defense.

Income is a calculation, not a field

Bank-statement, P&L, 1099, asset-depletion — each method involves steps a reviewer can dispute. If the math isn't shown, the number isn't defensible.

Guidelines change mid-pipeline

An investor updates a matrix on a Tuesday. You have 40 files in flight. You find out when a loan gets kicked back, not before.

Sending every page to a model is a liability

Most AI tools send the full document to an external model. That's borrower PII leaving your environment, and a per-page bill you can't predict.

What makes it different

Reading a document is easy.
Defending the number is hard.

01 / Built for non-QM

Knows the instruments, not just the fields

Bank-statement, P&L, 1099, asset-depletion — each has its own income method, its own edge cases, its own audit exposure. LoanCite runs the math the way an underwriter does and shows every step.

Bank-statement, P&L, 1099, and asset-depletion — each cited and reproducible.
02 / Defensible by design

The file answers for itself

Every figure is cited to its source page. Every rule has a verdict. Every exception is logged and resolved by a human. The full decision chain is hash-chained and exportable on demand.

Trace any number back to its source doc and the guideline version applied.
03 / Fits your shop

Your credit box, your environment

Rules are config — your toggles, severities, and thresholds, not ours. Recognized documents parse without an AI key. Everything stays inside your deployment. A real engineering team handles the setup.

Deterministic parser for recognized docs. Model only sees scans.
Income math — shown and reproducible
The number, and exactly how it was reached
12-mo personal deposits$663,744
less non-qualifying transfers($331,872)
eligible deposits$331,872
÷ 12 months$27,656
× 50% expense factor$13,828 / mo
How we compare

Speed is table stakes.
Defensibility isn't.

Generic extraction tools get data out of documents. LoanCite gets data out and proves where it came from.

Full Partial Not offered
Capability LoanCite Per-page Agents Platforms
Built for non-QM instruments (bank-stmt, P&L, 1099, asset-depletion)
Income computed with cited, reproducible math
Citation on every figure, traced to source line
Human in the loop on every exception
Hash-chained, tamper-evident audit trail
Decision-ready binder, figure-level & guideline-version pinned
Key-less extraction, no per-page meter
GSE LL-2026-04 vendor-governance packRoadmap

Compared to common categories of AI lending tools, not specific vendors.

Our agents

Two agents. One defensible file.

Start with document processing or jump straight to underwriting. Add the other when you're ready.

M
01 / Processor
Marcy
Pulls deal files from email, portals, and cloud storage — organized, classified, and validated before anyone touches them.
Document ingestion Classification Validation Completeness check
J
02 / Underwriter
James
Runs the income math, cites every figure to its source, produces the rule-by-rule report card, and populates your underwriting model. Zero re-keying.
Income math Citation engine Rule-by-rule report LOS export
T
03 / Asset Manager
Tyler
Tracks covenants, surfaces risks before they become problems, and generates on-demand portfolio reports — without you having to go looking.
Coming soon
Marcy · Automated Processor
Captures all deal files across channels
email · portal · cloud storage
Pulls key information across the loan package
42 fields extracted · 3 docs
Ensures completeness and data quality
Checking CPA letter date…
4
Prepares structured inputs for underwriting
waiting
James · Supercharged Underwriter
Standardizes financial and collateral data
bank-stmt · p&l · 1099 normalized
Runs cited income math
$13,828/mo · cited to p.3–14
Produces rule-by-rule report card
2 exceptions flagged · 1 blocking
Populating your underwriting model
Exporting to Encompass…
Tyler · Proactive Asset Manager
Coming soon
Covenant tracking, intelligent alerts,
and on-demand portfolio reporting.

Each agent connects to your existing LOS, cloud storage, and email. They run independently or together — start where the pain is biggest.

Fannie Mae LL-2026-04 · Aug 6, 2026

LL-2026-04 lands in August. Your vendors are included.

Fannie Mae's new lender letter requires an AI governance framework — and it explicitly covers the vendors behind your tools, not just your own use. LoanCite ships the governance pack, audit trail, and human-in-the-loop layer out of the box, with an exportable record you can hand to an investor or regulator on the spot.

Informational, not legal advice. LoanCite does not claim that human-in-the-loop or any specific tool is legally mandated.

Versioned rule governance

Every rule has an owner, a version, and a change history. Config-only — no code changes required.

Human signs off on every exception

Blocking exceptions route to your underwriter. Nothing auto-clears. The approval is logged to the record.

Exportable governance pack

A single PDF + JSON export covering your AI use, your vendor (us), and every decision on a given file.

"Finally, income math that shows its work. Our underwriters actually trust the numbers now."
Head of Credit$400M+ annual origination
"When a note buyer samples a file, we export one record instead of excavating inboxes. Game-changing."
Chief Compliance OfficerNon-QM Lender
"LoanCite catches discrepancies I would have missed at midnight. The citation trail is what makes it defensible."
Senior UnderwriterRegional Mortgage Shop
"Setup took weeks, not quarters. Rules config instead of code was the right call for our team."
VP of OperationsSpecialty Finance Lender
"Finally, income math that shows its work. Our underwriters actually trust the numbers now."
Head of Credit$400M+ annual origination
"When a note buyer samples a file, we export one record instead of excavating inboxes. Game-changing."
Chief Compliance OfficerNon-QM Lender
"LoanCite catches discrepancies I would have missed at midnight. The citation trail is what makes it defensible."
Senior UnderwriterRegional Mortgage Shop
"Setup took weeks, not quarters. Rules config instead of code was the right call for our team."
VP of OperationsSpecialty Finance Lender
Data stays where it belongs

Private by default,
not as an add-on.

Your environment, your data

LoanCite runs inside your infrastructure. Borrower documents never touch our servers unless you explicitly configure that.

No AI key required for standard docs

Recognized document types parse deterministically — no external model call, no per-page cost, no PII leaving your environment.

Isolated per lender

Each lender gets their own extraction config, rule set, and branding. Isolation is architectural, not just access-controlled.

We don't over-claim certifications

Built to SOC 2 standards. We'll tell you exactly what we have and what's in progress — not what looks good in a deck.

See it for yourself

See it on a real file,
not a slide deck.

The live demo runs an actual bank-statement loan through extraction, income math, and rule checking — in real time. Then book time to see it against your own guidelines and document types.

Talk to the team

We'll run it on a file that looks like yours, against rules that match your credit box.

    We reply within one business day. No sales deck, just the demo.

    Questions

    Questions we get before the demo.

    Does this replace my LOS?+

    No — and that's intentional. LoanCite sits alongside Encompass, ICE EPC, and most other systems. The file we produce lands in your LOS; your team works the same way they do today, just with a complete, cited package already in front of them.

    Does LoanCite make the credit decision?+

    Never. LoanCite prepares the file and reports on the rules. The underwriter decides. That's not a limitation — it's the design. Human sign-off is what makes the decision defensible.

    Where does borrower data go?+

    With private deployment, it stays in your infrastructure — full stop. The key-less parser handles standard documents without calling any external model. The AI only sees what can't be parsed deterministically, which is typically scanned or non-standard pages.

    Can you match our specific credit box?+

    Yes. Rules are config-only — your thresholds, your severities, your toggles. For custom income methods or document types we haven't seen before, our engineering team handles the build. We've done this for shops with unusual P&L structures and non-standard investor matrices.

    We're not sure if LL-2026-04 applies to us.+

    It likely does if you use any AI or ML in your origination process — and it covers your vendors, not just your own tools. LoanCite ships the governance framework, audit trail, and human-in-the-loop layer as part of the standard product. We can't give legal advice, but we can show you exactly what we provide and let your counsel decide.

    How long does it take to go live?+

    Most pilots are running in two to four weeks. Standard income methods and common document types are configured out of the box — custom work takes longer, but we scope that before you commit.