The build kit

What we're replacing — and what we have to beat

The goal is not to fill a gap between LendingPad and Floify. It is to build the platform that replaces both, beats Blue Sage, and can be sold to other lenders. That ambition only survives contact with reality if we are honest about how good the incumbents actually are. Every claim below carries its confidence level.

How to read the confidence tags. VERIFIED found in a primary or independent source · VENDOR the vendor's own marketing — directionally useful, not evidence · UNVERIFIED reasoning or operator testimony, not yet sourced. Nothing on this page should be shown to an investor as fact unless it is tagged VERIFIED.

⚠️ The finding that is urgent — and it is not competitive

Fannie Mae Lender Letter LL-2026-04 took effect on August 6, 2026 — six days ago. VERIFIED

Every Fannie seller/servicer using AI or machine learning in origination or servicing must maintain a written, active AI/ML governance framework addressing six specified elements — reviewed at least annually — and must, on Fannie's request, promptly disclose the types of AI used, the purpose and manner of use, and the safeguards in place. Freddie has set parallel standards.

What this means for Nacho: the AI guardrail charter in Part 7 of the Atlas is no longer a design preference — it is a delivery requirement. "AI built in, not bolted on" is the pitch; being able to answer "for this loan, what did the model see, what did it output, who reviewed it, and what did the human do with it" is now the price of selling the loan. Any platform sold to other lenders must let them satisfy this too — which makes model governance a product feature, not internal hygiene.

This is arguably the single most valuable thing this research turned up, and it has nothing to do with Blue Sage.

Blue Sage — the honest profile

The instruction was to beat them. So this is written to be useful, not flattering to us. Blue Sage is a genuinely strong product, and pretending otherwise would produce a plan that loses.

DimensionWhat's trueConfidence
Architecture100% cloud-native, API-first, multi-channel — retail, wholesale, correspondent — with integrated borrower, LO and broker portalsVENDOR
LeadershipFounder/president Carmine Cacciavillani, ~30 years building loan origination technologyVERIFIED
ScopeOrigination and servicing — a Digital Servicing Platform alongside the lending platformVERIFIED
POS"LION" — and they sell it standalone, to run on any LOS the lender already hasVERIFIED
AI (June 2026)SageVision — extraction and validation across borrower, income, asset, collateral, purchase agreement, title and insurance docs, cross-document mismatch detection, and confidence scoring that triggers exception workflows. AI Studio — document intelligence + guideline analysis + condition automation. Voice AI — servicing calls with sentiment analysisVENDOR
Pricing~$200–800 per user per month, volume/configuration dependent; no published rate card; quotes customizedUNVERIFIED third-party estimate
Target marketMid-to-large lenders, ~500–10,000+ loans/yearUNVERIFIED
ImplementationOperator testimony: ~$35k and 3–4 months to configure, limited customization afterUNVERIFIED — Sharon's account
CustomersPrimeLending ($14.5B); Royal Credit Union (MN/WI); customer base more than doubled over two yearsVERIFIED
WeaknessesNewer player, smaller installed base, integration ecosystem still growing vs. Encompass; very little public customer review dataVERIFIED

Three things this changes about the plan

1 · The business model is proven

Blue Sage already sells its POS separately to lenders running a different LOS. That is the Nacho commercialization play, validated by a competitor: a modular piece, sold to the whole market. It also means their architecture is genuinely separable — which is the bar Nacho must clear to be sellable, not merely usable in-house.

2 · "AI" is no longer a wedge by itself

As of June 2026 Blue Sage ships document extraction, cross-document mismatch detection, confidence-scored exception routing and condition automation. "We'll have AI" is table stakes, not a differentiator. The Atlas needs to say that plainly — the wedge has to be somewhere they are not.

3 · The open door swings one way

Their POS runs on anyone's LOS. I found no statement that a third-party POS can replace LION on their LOS. If that asymmetry holds, it is a real wedge: Nacho open in both directions — and it is also the question that decides whether a Blue Sage contract at Aspire would strand the Nacho POS. Ask them directly.

Where the genuine wedges actually are

WedgeWhy it holdsAtlas
⭐ Trench-built worksheetsFHA streamline with UFMIP refund and pennies dropped · VA IRRRL comparison + NTB · TX A(6) 2% cap · escrow holdback · VVOE. Nobody ships these because no competitor's product manager has balanced a wire at 4:50pm. This is the most defensible thing in the whole build5.1
⭐ Owning both halves of the seamA borrower front end and a back office built to one data model with the boundary written down. Blue Sage sells a POS for other people's systems; nobody sells a matched pair with a published contractPart 11
Configurability without a services engagementIf configuring Blue Sage really costs ~$35k and 3–4 months with limited customization after, then an admin console a Director of Operations can drive herself is a real product claim6.5
Operator-built depthThe 2015 at 5–10 minutes instead of 20–25. Vesting checkboxes with defined document consequences. A funding screen that actually works. These come from people who do the job4.6–4.9
Governance as a featurePost-LL-2026-04, a platform that can produce the AI audit trail its customers owe Fannie is selling compliance, not just software7.5, 7.6

The window nobody is talking about

ICE is forcing its own customers through a migration right now. VERIFIED The Encompass SDK sunset was pushed from Oct 2025 to December 31, 2026; no new SDK features since November 1, 2025; after the deadline, transitional SDK access requires special approval and incurs monthly fees. Lenders and vendors are being moved to Encompass Partner Connect — more than 70% of service orders already flow through it.

Why it matters commercially: a large installed base is being forced to touch its integration layer anyway. The cost of evaluating a replacement is never lower than when you are already being made to migrate. That is a market timing argument for the Nacho commercialization plan — and it has a deadline on it.

The economics — what the market will bear

NumberValueSource
Cost to originate — independent mortgage banks$12,209 per loan (2025) — down 5% from 2024, down 15% from the $14,381 study high in 2023VERIFIED MBA
Cost to originate — depositories$16,320 per loan (2025)VERIFIED MBA
Total loan production expense336 bps in Q1 2026, up from 323 bps in Q4 2025VERIFIED MBA
IMB production profit$727 per loan (16 bps), Q1 2026VERIFIED MBA
Industry AI condition clearingSome lenders report auto-clearing 70–75% of credit, income and asset conditions without underwriter touch, targeting 85%+UNVERIFIED vendor-adjacent
Read those two numbers together. It costs an IMB $12,209 to make a loan and they earn $727 on it — and expenses went up last quarter. The margin is roughly 6% of the cost. That is the entire commercial case for this build stated in two figures: at those margins, a platform that removes touches is not a nice-to-have, and a per-seat licence that taxes headcount growth is a genuine strategic problem. It also means any buyer of Nacho will do this exact arithmetic — so the pitch must be cost-per-funded-loan, not features.

⚠️ The counter-argument, taken seriously

A plan that hasn't survived its own counter-argument isn't a plan. This is the least comfortable section here, and it belongs in front of leadership rather than in an appendix. UNVERIFIED — industry analyses and vendor-adjacent research, directionally reliable, individually unconfirmed.

FindingWhat it means here
~30% of large digital transformation efforts succeedThe base rate is against big replacements. Nothing about this one exempts it
Custom LOS ≈ 3–5× SaaS cost over three years; 18–24 months, 8–15 engineersConsistent with our own 12–24 month estimate. It is not the cheap path and must never be sold as one
Enterprise LOS migration: $5–10M+ — data, re-integration, retraining, disruptionCuts both ways — the most strategically useful number on this page
Integrations built for current-state ops become liabilities in 18–36 monthsWhy the seam is a versioned contract and an adapter, not point-to-point wiring
Building is defensible with large dedicated engineering, a genuinely novel product, or data-sovereignty constraintsTwo of the three apply — a dedicated vendor team, and a product intended to be sold
⭐ Read the $5–10M switching cost twice.
Against us commercially: every lender we'd sell to is entrenched by that same cost. It's why inferior systems survive. So the pitch can't be "ours is nicer" — it has to land in a forced-migration moment (the Dec 31 2026 window above) or at a new lender with nothing to migrate.
For us operationally: it's also the reason not to sign a three-year contract for a system we intend to replace. Switching costs are paid on the way in and on the way out.
The honest conclusion. The build case does not rest on cost. It rests on three things: capabilities that genuinely cannot be bought (the worksheets, the seam), a per-seat pricing model that taxes the growth we're pursuing, and converting build spend into a product asset. If those three stop being true, the build case stops being true with them.

What is still open in this research

Deliberately listed rather than glossed. Still to close: Blue Sage's real implementation timeline and contract shape from an actual customer rather than operator hearsay · whether a third-party POS can replace LION on their LOS (the D-01 question — ask them directly) · documented customer complaints, which are genuinely scarce in public sources · MeridianLink, Dark Matter Empower, Byte and Mortgage Machine at this depth · what failed LOS replacement projects specifically got wrong · and the certification and approval path for licensing a platform to other lenders (SOC 2, agency and investor integrations, support model).

The white-label question is answered: the intent is to white-label eventually, with an eligible in-house instance running first. That sequencing is right, and it is now Decision D-15 with the commercialization requirements attached to it.