Solutions
I AM A
I NEED
Company
The real test of QC automation is simple: can an auditor rely on an answer without redoing it? If not, every automated answer becomes a second review. Across the 1.1 million audits a year that run on the AuditGenius platform, Indecomm sees the same six patterns stand in the way.
In a recent HousingWire interview with Allison LaForgia, Indecomm CEO Rajan Nair framed the question this way:
“Can you get an automation solution that the auditors can trust, embrace, and adopt?”
Similar loans come back with different results, or the same loan changes on a rerun, so auditors re-check everything.
A yes or a no with no page in the file behind it sends the auditor looking for the page anyway.
False findings have to be cleared one by one, and missed findings mean nothing can be skipped.
The checklist is answered as worded, while what the guideline actually requires goes unexamined.
Fields that extraction missed or misread, especially on less common documents, have to be looked up by hand.
New checklists, guideline updates and overlays can wait on an engineering release, so manual workarounds fill the gap.
Each one ends the same way: the auditor redoes the work the automation was meant to remove.
Consistency deserves particular attention. Many automated tools have an AI read the file on the spot, during the review, so the same file can produce two different answers with no clear record of how either was reached. That puts QC and compliance leaders in a difficult position when an investor, an agency or an internal auditor asks how a finding was made.
“Too many automated audits operate like a slot machine: The same file can yield a different answer depending on when it’s run,” Nair said. “Consistency is the foundation of a defensible QC program, and it’s where tools that prioritize AI over QC fall short.”
The stakes are rising. Fannie Mae’s Lender Letter LL-2026-04, effective August 6, 2026, sets a governance framework for seller/servicers using AI or machine learning, and Freddie Mac’s AI governance requirements have been in effect since March 3. Lenders remain responsible for the AI their vendors use, and must be able to disclose on request how those systems are used and what safeguards are in place.
A checklist question is shorthand. When an experienced auditor answers “Is the borrower’s income supported by the documentation in the file?”, they check the loan against what the agency guideline and the lender’s overlay actually require. Many automated tools skip that middle step: the AI reads the file and returns an answer, with nothing telling the system what the guideline requires.
AuditGenius | ai starts with the requirements, in three steps:
Each checklist question is tied to the agency, investor and lender guidelines behind it and to the loan data that answers it.
Our mortgage QC specialists review and approve that logic before it is built into the platform, and review it again whenever the guidelines or the logic change.
At review time, the approved logic runs unchanged, so loans with the same facts and circumstances receive the same findings, and every finding links back to the requirement behind it.
“You want to do all the AI work upfront for the authoring or the preparation of the system,” Nair told HousingWire. “You want minimal AI running in production, live in real time.”
Auditors can see exactly how a finding was reached, step by step, from the guideline to the answer.
Each finding links straight to its supporting page, with the rest of the loan file at hand and no system switching.
Adverse findings are presented from the perspective of the auditor, the lender and the investor.
Auditors can ask plain-language questions about the loan, such as which documents support the borrower’s income or why a finding was raised.
Potential fraud and irregularities are flagged for review, and the auditor decides what, if anything, becomes a finding.
Any finding can be overturned, every override is recorded and the auditor closes the file.
Asked what a lender should be able to show an auditor or investor, Nair pointed to a glass box. “Every aspect of the QC that has been done, you must be able to drill down to the last minute data field, essentially.” Asked how quickly that information should be available, his answer was one word: “Instantaneously.”
“When verification is easy, trust follows,” Nair said. “When an auditor opens a file, the checklist is already answered, with the finding, the evidence and the page citation in one place. Instead of hunting through the file, auditors spend their time on judgment calls, and they still decide every outcome.”
That kind of control matters. As Robbie Chrisman wrote in a recent Chrisman Commentary, “the mere presence of a human reviewer should not be conflated with meaningful oversight. Effective governance requires that individuals possess not only visibility into an AI system’s reasoning, but also the authority, expertise, and institutional support necessary to challenge its conclusions.” Glassbox provides the visibility, and recorded overrides give auditors the authority to act on what they see.
The requirements and the logic that evaluates them are drafted with AI, approved by Indecomm’s QC specialists and reviewed again whenever the guidelines or the logic change.
The approved logic runs unchanged on every loan, so loans with the same facts and circumstances get the same findings.
Extracted values are checked against the document itself, against arithmetic that has to add up and against the other documents in the file. A model’s confidence score is an input, never the gate.
AuditGenius | ai builds on 25 years of mortgage review work and 10 years of building automation for the mortgage industry. The AuditGenius platform supports 75,000 users running 1.1 million audits a year, and AuditGenius | ai carries that track record, along with more than four years of QC data, into its early-adopter preview ahead of a broader release.
Read HousingWire’s coverage of the launch for more on the preview.
As Chrisman put it, “Trust is not created by automation alone; it emerges from thoughtful workflow design, transparent decision-making processes, and clearly defined accountability structures.”
For QC leaders, the measure of automation is not how many checklist questions it answers. It is how many of those answers an auditor can rely on without doing the work again.