Threshold

For mortgage brokers & lenders

Threshold moves the file to clear-to-close — without touching the credit decision.

You already run Encompass, Floify, and a borrower portal that still pings the LO at 9pm. Threshold wires AI into that stack safely and keeps the underwriter, the LO, and the examiner comfortable. Done with you, from $499/mo.

7 min read

See pricing

The shift

It now costs more to make a loan than the loan makes. Production expenses hit about $12,579 per loan in early 2025, while the average independent mortgage bank netted just $443 in profit per loan in 2024 — after losing $1,056 per loan in 2023. Rate and volume aren't yours to control. Cost to manufacture is, and that's the one lever where AI actually lives.

Your reality

Your day isn't underwriting — it's chasing the two pay stubs that still aren't uploaded and answering "any update?" for the fourth time. Processing teams burn hours each week just relabeling documents, while borrower satisfaction runs about 40 points higher when a local human is actually advising instead of collecting PDFs. AI should do the chasing so your people can do the advising.

$12,579

to manufacture a loan in early 2025

$443

average profit per loan in 2024

+40 pts

borrower satisfaction with a human advising

73%

of lenders adopt AI for operational efficiency

Wires into the stack your shop already runs

EncompassFloifyBlend1099s & W-2sInvestor overlaysBorrower POS

Sound familiar?

The conditions loop never ends — one missing page, including the blank one, bounces the file and pushes clear-to-close another day out.
Everyone's asking "where's my loan?" and the answer lives in your head, so borrowers and realtor partners flood the LO and processor with status calls.
Your selling hours get eaten by paperwork — referral-partner time and new applications keep losing to document chasing.
Re-keying is hundreds of keystrokes per file, and any one of them can misindex a document and stall the loan.
At $12,579 to manufacture a loan and $443 of profit in it, a couple of stretched processing days is the line between a profitable and a losing quarter.

Where AI fits

Turn a two-hour statement read into a reviewable draft.

Today an underwriter reads bank statements line by line and a processor re-keys hundreds of fields by hand.

AI ingests the pay stubs, W-2s, 1099s, and bank statements, classifies each page, structures the numbers, and drafts a guideline-aware income calculation into your LOS — it never finalizes qualifying income.

Fewer manual keystrokes
Faster statement analysis
Self-employed files less painful

Watch for: A licensed underwriter reviews and signs every income calc · AI prepares, the human qualifies · Low-confidence extractions wait for review

Best for: IMBs with in-house underwriting buried in self-employed and bank-statement files.

Stop relabeling PDFs by hand every week.

Every inbound upload gets manually named and filed to a condition today.

AI auto-classifies the uploads — "2024 W-2, borrower 1," "bank statement, page 3 of 5" — drops each into the right condition, and flags missing pages before they bounce the file.

Less relabeling time
Fewer misindexed files
Missing pages caught early

Watch for: Low-confidence items wait for the processor to confirm · Nothing posts to the file unreviewed · An audit trail on every label

Best for: Broker shops and IMBs where one processor's relabeling backlog gates the whole pipeline.

From conditional approval to clear-to-close with fewer bounces.

The processor reconciles the underwriter's condition list against what's actually arrived by hand, then re-submits and hopes.

AI reads the condition list, matches received documents to each item, and drafts the resubmission package with a clean list of what's still outstanding.

Fewer round-trips to underwriting
Tighter cycle time
Cleaner pull-through

Watch for: The processor approves each submission · The underwriter still clears every condition · AI never marks a condition satisfied on its own

Best for: Any shop where re-issued conditions are quietly eating cycle days.

Answer "where's my loan?" from the live file.

Borrowers and realtors flood the team with "any update?" all day long.

AI answers grounded in the real file state — "we received your pay stubs; underwriting is reviewing two remaining conditions" — and proactively asks the borrower for the specific document holding things up.

Fewer status calls
Borrowers nudged precisely
LO freed to advise

Watch for: Anything touching a TRID date escalates to a human · AI never promises a closing date · Frustrated borrowers get a human with full context

Best for: High-volume LOs and processors buried under status interruptions during peak season.

When rates move, work the back-book before it goes cold.

A 50-75bps drop can multiply a pipeline overnight, but the refi opportunity sits unworked because no one has time to comb past clients.

AI scores the back-book for who's now in the money and drafts personalized outreach for the LO to review and send.

Dormant clients surfaced fast
Personalized drafts ready
Selling hours preserved

Watch for: The LO reviews and edits every message · The LO owns the send · No auto-quoting of rates

Best for: Shops with a real servicing back-book sitting idle between rate cycles.

Does this investor allow a 45-day-old bank statement? With the citation.

Processors lose time hunting through investor guideline matrices, agency overlays, and internal SOPs across disconnected systems.

AI runs grounded, cited search across your guidelines and SOPs and returns the answer with a link to its underlying source.

Faster guideline lookups
Fewer "let me check" delays
Answers a human can verify

Watch for: Every answer cites its source guideline · A person verifies before acting · Never a freestanding ruling

Best for: Broker shops juggling many investors and overlays, where the answer is "it depends."

Safe to start vs. proceed with guardrails

Safe to start now

Auto-classifying and indexing inbound documents, with the processor confirming low-confidence items.
Drafting an income calculation for the underwriter to review and sign.
Matching received documents to open conditions and drafting the gap list.
Answering borrower status questions from the live file state.
Drafting policy-grounded replies to routine questions for human review before send.
Cited internal search across investor guidelines, overlays, and SOPs.

Proceed with guardrails

Anything that becomes a Loan Estimate or quoted figure — a licensed person finalizes it (TRID/RESPA).
Closing-date and TRID-timed commitments — no AI promise on a date that creates a legal obligation.
Any credit decision or adverse-action notice — the CFPB is explicit there's no "fancy-tech" exemption; the principal reasons come from a licensed human (ECOA/Reg B).
Pricing and rate quotes to borrowers — a licensed LO runs the consultation; AI never states approval or quotes a rate.
Borrower PII like SSNs, tax returns, and full financials — GLBA obligations are scoped in before go-live and PII stays inside your governed systems.
Multi-step file orchestration — automate the choreography, but keep human sign-off gates at disclosure send, credit decision, and clear-to-close.

Why do it with us

Hire a consultant	High hourly, a long discovery phase	A deck at the end, then they leave
Hire an AI engineer	A salary your volume can't justify in a slow cycle	Needs the mortgage and compliance context you can't teach
DIY on nights and weekends	No new line item	The conditions loop wins and selling hours slip away
systemlevel.ai	From $499/mo, pause anytime	One senior expert wires AI into your stack, accountable for results

It costs more to make a loan than the loan makes: about $12,579 to manufacture a loan in early 2025, against just $443 of average profit per loan in 2024 (and a $1,056 loss per loan in 2023). Cost to manufacture is the lever AI can actually move.
Borrower satisfaction runs about 40 points higher when a local human is directly, advisorily involved — AI does the chasing so your LO does the advising.
73% of lenders cite operational efficiency as their reason to adopt AI, but only 7% have fully deployed it and 22% are in limited trials. The gap isn't desire — it's the team to choose, integrate, and govern it safely.
Vendor case studies report large drops in per-file processing time when income extraction is automated. We treat those as vendor-reported, not guarantees, and measure your real results in plain numbers — hours per file, conditions cleared, cycle days.

Questions you’re probably asking

Isn't AI in lending a fair-lending lawsuit waiting to happen?: A founded concern — the CFPB has been explicit that there's no "fancy technology" exemption from adverse-action rules. So Threshold deliberately stays on the operational side: document extraction, classification, status, and drafts. It never makes the credit decision or writes the adverse-action notice — those stay with your licensed people, every output is reviewable, and everything decision-bearing carries an audit trail.
This is the most sensitive data on earth — SSNs, tax returns, full financials.: Right, and data security is a top SMB AI concern for good reason. We scope your GLBA obligations into the build before go-live, recommend vendors and models with the right data handling, and keep PII inside your governed systems.
It won't integrate with Encompass, Floify, and our 200 investor overlays.: Integration complexity is the single biggest barrier lenders name. That's exactly why we start by mapping your real stack and wire Threshold into Encompass and Floify rather than asking you to rip and replace — one workflow first, not a migration.
There's no proven track record for a shop my size.: Fair, and it's a top barrier lenders cite. So we start small and measured: one workflow, plain-number results like hours per file, conditions cleared, and cycle days, and we only expand what's working. Done with you, from $499/mo, pause anytime.
I don't have time to figure out where AI even helps.: That's the whole service. A senior expert maps your file flow and ranks the highest-value, lowest-risk candidates for your shop, so you don't go it alone or guess where to start.
My borrowers want a human, not a bot.: Exactly — satisfaction rises about 40 points with a human in the loop. AI absorbs the document chasing and status noise so your LOs spend more time advising people, not less.

Pricing

Operator

$499/mo. Monthly strategy, your first workflows mapped and scoped (start with document intake), vendor and model recommendations that fit Encompass and Floify, a stack assessment, and email support.

Best for: The broker-owner or LO who knows the conditions loop is the problem and wants a safe place to start.

Partner

$999/mo. Everything in Operator, plus bi-weekly working sessions, implementation guidance through deployment, LOS/POS integration review, team training and prompt libraries for processors and LOs, and a direct line for unblocking.

Best for: The IMB or growing shop ready to wire AI through processing and underwriting.