Every math teacher has said it at least once: show your work. The right answer earns partial credit. The right answer with every step on the page earns full credit, because it proves you understood how you got there.
Mortgage lending runs on the same rule. It simply grades harder.
The repurchase request lands on a Tuesday. An investor has flagged a loan your team closed eight months ago, and they want to know how your quality control review handled it. You pull the file. The audit was AI-assisted. The item in question is right there, marked as cleared. Now someone has to explain why.
If the answer is “the system said so,” you have a problem. If the answer is a click to the exact document, page and field that supported the decision, with the auditor’s sign-off beside it, you have a defense.
That difference is what Indecomm calls the glass box. It has guided how we build our automated underwriting platform,
DecisionGenius, and it is the foundation of how we approach AI-driven quality control with
AuditGenius | ai today.
Every finding gets questioned eventually
The investor asks. The examiner asks. The warehouse lender asks. Sooner or later, every decision in a loan file meets someone who wants to see how it was made.
In a recent conversation about AI in mortgage QC (quality control), Indecomm CEO Rajan Nair was asked what a lender should be able to show in that moment. His answer was immediate.
“They should be able to show them a glass box. You look at the findings, and you need to be able to get right down to the document, to the page, to the field. You need to be able to show them this is what drove this particular decision or outcome.”
Then he went further, to the part of an audit that rarely gets airtime.
“Every issue that you point out, you need to be able to drill down. But also not just issues. What was cleared by the auditor? Why was that cleared? Were there false positives?”
Cleared items are where the hard questions hide. A defect on a report draws attention. A clean item draws trust, and that trust holds only as long as someone can show why the item was clean. False positives add another layer. AI models raise flags that turn out to be fine, and an auditor needs to see why a flag was dismissed as clearly as why it was raised.
In other words, the glass box has to be clear on all four sides: what was found, what was cleared, what was flagged in error and why.
The industry is writing the same sentence
Read the mortgage trade press this year and you will notice a refrain. Different authors, different outlets, one message.
In a
HousingWire article related to building AI agents that compliance teams will trust, Sandeep Shivam distilled it to a single line: “If a loan document was marked insufficient, there should be a reason.” He argued that each surfaced exception needs a trail showing which policy rule, document fact or workflow signal produced it.
Transparency now has a deadline
What began as best practice is now written into the rulebook
. Fannie Mae issued Lender Letter LL-2026-04 on April 8, 2026, establishing an AI and machine learning governance framework for seller/servicers that took effect August 6, 2026. Writing in
National Mortgage Professional, James Brody and Ronald Gapp of BRODY | GAPP LLP explained that Fannie Mae and Freddie Mac have written AI governance directly into their Seller/Servicer Guides, which turns it into a contractual obligation for lenders with GSE approvals. Among the expectations they highlighted: validating model explainability where AI supports borrower-facing decisions.
The obligation extends to every vendor in the chain. The letter requires seller/servicers to govern vendor AI to a standard “no less protective” than their own, and to disclose on request the types of AI they use, how they use it and the safeguards in place
. In a separate National Mortgage Professional analysis, Brody framed the practical test for lenders: whether a vendor’s disclosures are sufficient for the lender to support its own governance and examination obligations. A platform that shows its reasoning makes that test easy to pass.
A platform that shows its reasoning by design answers both concerns. The clarity is there on day one, and it is still there on day five hundred.
We chose glass from the start
When
Indecomm released DecisionGenius in March 2021, we made a deliberate choice. Automated underwriting systems were already common, and many returned a decision with little visibility into the reasoning behind it. We wanted underwriters to see every step.
DecisionGenius delivers risk-based analysis across credit, assets, income and collateral with full transparency into each finding, showing underwriters exactly how it reached its conclusion.
We have held that line as the platform has grown. When
DecisionGenius expanded to FHA lending in 2025, the factors behind each recommendation remained visible to Direct Endorsement underwriters, and the recommendations came backed by the Indecomm Promise warranty, except where an underwriter overrides them.
The person who signs needs to see
Underwriters and auditors put their names on decisions. When the questions come, they are the ones who answer.
Hand them a recommendation they cannot verify, and they face an uncomfortable choice. They can accept it on faith, which introduces risk no one can measure. Or they can redo the work by hand, which erases the productivity the AI was meant to deliver. Faith on one side, rework on the other, and confidence on neither.
A glass box opens a third path. The auditor sees the finding, follows it to its source, confirms it and moves on to the judgment calls that call for human expertise. Verification builds trust. Trust drives adoption. Adoption delivers the return the lender invested in. Each step depends on the one before it, and the first step is being able to see.
Inside the glass box: AuditGenius | ai
That same thinking now powers AuditGenius | ai, which Indecomm previewed at the MBA Annual Convention & Expo 2026.
In a recent
HousingWire video feature, Nair described the work that happens before an auditor ever opens a loan. The AI analysis is complete and staged in advance, built on what he calls scaffolding. Each investor requirement and lender overlay is tracked, mapped to the specific documents and fields that satisfy it, and mapped back into the audit checklist.
“When the auditor logs in to their system on that loan, they see here are the findings, here are all the things that were cleared. They’re able to click all the way down to the document.”
The scaffolding stays out of sight. The results stay in full view.
The loan arrives already audited. Each result is complete, sourced to the page and decided by the auditor, with the same answer every run. Indecomm’s mortgage QC specialists review and approve the audit logic before it enters the platform, and review it again whenever guidelines or logic change. The data comes from
IDXGenius | ai, Indecomm’s own extraction platform covering more than 1,400 pre-mapped mortgage document types, which lets AuditGenius | ai reach every data point in the loan file.
Now return to that Tuesday. The investor wants to know why the item was cleared. With AuditGenius | ai, the answer is a few clicks away: the source document on screen, the requirement it satisfied and the auditor’s decision on record. The work is shown.
Built by people who do the work
There is a reason the glass box looks the way it does. Rajan makes that clear:
“We’ve been at this for years now, and we’ve seen where the pitfalls are and where the adoption challenges come in. We’re fortunate because we have a very large QC operation ourselves. That’s what we’re now able to talk to lenders about and show them working.”
Our own auditors told us what they needed to trust an AI finding. We listened, and we built it.
Five questions to ask any AI provider
Before you trust a platform with your underwriting or QC, hold it up to the light:
- Can an auditor trace each finding to the exact document, page and field behind it?
- Can the system show why items were cleared, as well as why they were flagged?
- Will the same loan produce the same result every time it runs?
- Who reviews and approves the logic behind the outputs, and how often?
- Can the outputs support a GSE disclosure request?
Full credit
AI will keep getting more capable. The question from auditors, investors and regulators will stay exactly the same: show me how you got there.
The right answer still earns partial credit. In mortgage lending, full credit goes to the answer that shows its work, all the way down to the page.
Since
DecisionGenius launched,
Indecomm has built its Genius AI Suite to earn full credit. With AuditGenius | ai, the glass box now extends to quality control, so every finding your auditors sign can stand up to any question that follows.
