For mortgage compliance leaders, post-close QC managers, and VPs of loan quality, auditing only 10% of closed loans has become the standard operating limit. Fannie Mae’s current post-closing QC rules allow lenders to use a random sample of at least 10% of originated or acquired loans, or a qualifying statistical sample.
But a minimum can become an operating ceiling when every full-file review depends on hours of manual evaluation. In a lender that reviews exactly 10% of production, 90% of closed loans sit outside the random full-file sample. Those loans may receive targeted checks, but they do not receive the same sampled review unless selected later.
Lenders can now keep the same audit team but raise its sample from 10% to 30% or more using Infrrd’s AI platform, Ally for mortgage QC audits, which completes the 10-step audit process in under 30 mins. It helps increase the number of loans the same team can evaluate in accordance with agency rules and lender overlays.
Why 10% Became the Industry Norm
The 10% figure did not originate as a statement that reviewing one loan in ten creates perfect control. It became familiar because Fannie Mae permits a minimum 10% random post-closing sample, provided the selected loans represent the lender’s business. Lenders may also use a statistical method in which variables are calculated at a 95% confidence level, with a 2% precision rate, and a statistical statement window of no more than 6 months (3 months is recommended).
That regulatory floor still requires major effort.
A lender closing 10,000 loans each month would need to complete 1,000 full-file reviews under a straight 10% random method. Each review can include document checks, data validation, underwriting analysis, compliance testing, reverifications, defect grading, and reporting.
Fannie Mae also requires reverification of qualifying file components for random reviews. These components can include income, employment, assets, liabilities, valuation, and occupancy. practical limit is staff capacity. Teams size the sample around the number of loans auditors can finish within the required review cycle.
Sampling creates a second problem. A smaller sample gives the lender less direct evidence about the unreviewed population. The reviewed 10% may reveal defect patterns, but it cannot show the actual condition of every loan in the other 90%.
That untouched share carries the gravity. It is not just work left undone; it is risk that remains less visible until an investor request, a regulator inquiry, a borrower issue, a targeted audit, or a later quality check brings it to light.
The Real Cost of Under-Sampling: A Worked Example
Consider a lender that closes 1,000 loans each month, as earlier.
Under a 10% random sampling model, the QC team evaluates 100 loans. The other 900 loans remain outside that random full-file sample.
Now apply an illustrative 3% defect rate across the monthly population. This is a simple planning assumption, not a claimed industry dataset.
At that rate:
- The 100 reviewed loans would contain about 3 defects.
- The 900 unreviewed loans could contain an estimated 27 defects.
- Across all 1,000 loans, the model projects about 30 defects.
The QC team finds the three defects that appear in the sample. The other 27 may remain unknown at that stage. Some may be minor while others may involve eligibility, income, disclosures, documentation, or investor requirements.
Those defects can surface after delivery. A later finding may lead to remediation, added review work, indemnification, pricing adjustments, or a repurchase request. Freddie Mac’s guide, for example, states that significant defects can lead to a repurchase request or another remedy. The model does not predict which loans will fail. It shows the coverage problem: defect detection rises when direct evaluation covers a larger share of production.
Recalculating the Math at 30% Sampling
Keep the same loan volume and the same illustrative defect rate:
- Monthly production: 1,000 loans
- Assumed defect rate: 3%
- New sample: 300 loans, or 30%
- Loans outside the sample: 700
At a 3% defect rate, the expanded sample contains an estimated nine defects. The unreviewed group contains an estimated 21.
The audit team now catches about nine defects before investor delivery or later review, compared with about three under the 10% model. That is three times as many expected defects identified at the same assumed error rate.
The error rate did not improve in this model. Visibility improved.
That distinction matters. A larger sample does not automatically create better loan quality. It gives the lender more chances to find defects, trace root causes, correct recurring issues, and act before an investor turns the same issue into a putback demand.
The 30% model still leaves 70% outside the random sample. The next step may be higher sampling, risk-based evaluation across the remaining population, or automated checks across all loans. Human full-file review can then focus on selected loans and flagged exceptions.
How Ally (Powered by MortgageCheckai) Makes the Math Work
What It Evaluates, and Against What
MortgageCheckai prepares and validates loan-file data. Ally applies mortgage QC logic to that data and supports automated pre-fund and post-close QC reviews. Infrrd’s mortgage platform supports compliance enforcement, automated reviews, income calculations, missing-document checks, version control, field comparison, and audit evidence. audit teams, the evaluation layer can be configured around:
- Fannie Mae guidelines
- Freddie Mac guidelines
- FHA and VA requirements
- TRID, TILA, and RESPA checks
- Lender-specific overlays
- Internal audit policies and thresholds
Infrrd’s goal is to apply the same repeatable standard across every evaluated loan, then route judgment-based exceptions to an auditor.
This does not replace required QC governance, reverifications, collateral review, or auditor accountability. It removes repeatable manual checking from the document prep process so people can spend more time on findings that need interpretation.
A Complete 10-Step Mortgage Audit in Under 30 Minutes
Infrrd’s AI helps complete a mortgage audit in under 30 minutes, with each stage taking three minutes. It automates 80% of repetitive audit work, reading documents, applying rules, checking evidence, and surfacing exceptions so auditors can focus on analysis.
The 10-step review includes:
- Income & Employment Verification: Validate income, employment, and documentation.
- Credit Review: Review credit history, liabilities, and DTI.
- Asset Review: Verify funds, reserves, and asset sources.
- Appraisal Review: Check valuation and property details.
- Loan Terms Verification: Compare terms, disclosures, and payments.
- Title & Insurance Checks: Verify title, coverage, and risks.
- Fraud Check: Identify inconsistencies and potential fraud indicators.
- Underwriting Compliance: Check guidelines and AUS findings.
- QC & Closing Review: Validate disclosures and settlement details.
- Post-Closing Quality Control: Consolidate findings and prepare the review.
With 80% of repetitive work handled by Ally, auditors can spend the remaining 20% investigating exceptions, validating findings, and making decisions that improve quality and reduce risk.
What It Means for the Audit Team, Not Just the Audit
More coverage changes the team’s role. Auditors spend less time repeating standard checks and more time judging exceptions, studying defect patterns, testing policy changes, and improving controls.
Teams can also model “what-if” rule or threshold changes against loan data before changing production policy. They can see how a new tolerance, calculation method, or overlay would affect open and completed audits.
The audit team gains more than faster processing. It gains broader coverage, consistent evaluation, clearer findings, and more time for work that requires professional judgment.
MortgageCheckai and Ally do more than extract documents or automate workflow steps. They evaluate loans against agency guidelines and lender overlays. That evaluation capacity allows lenders to audit more loans with confidence, rather than simply move more files through a queue.
Beyond Loan Volume: Consistency, Transparency, and Compliance Lock-In
Higher audit coverage creates more visibility, but volume alone does not create a defensible QC program. Mortgage lenders also need consistent checks, clear supporting evidence, and rule logic that reflects current agency and internal requirements. Infrrd applies these controls through repeatable evaluation, explainable findings, and centrally managed rules.
Consistency
Every evaluated loan passes through the same configured guideline and overlay checks. This reduces differences between reviewers when they perform repeatable tests. One auditor may interpret or document a manual check differently from another. Automated rule logic applies the same condition, threshold, calculation, and evidence requirement each time.
Human judgment still plays an important role. Auditors remain responsible for reviewing exceptions, interpreting policy, grading defects, and deciding the proper response.
Transparency
Each finding should show why the system raised it. MortgageCheckai can connect a finding to the relevant data point, source document, comparison, missing requirement, or configured rule. This gives auditors evidence they can review rather than presenting an unsupported pass-or-fail result.
Clear evidence also helps teams explain findings to operations leaders, compliance teams, investors, and regulators.
Compliance That Does Not Go Stale
Mortgage rules do not remain fixed. Agency guidelines, lender policies, thresholds, and regulatory requirements change over time.
A centrally managed rule library allows the lender to update checks for:
- TRID
- TILA
- RESPA
- Fannie Mae requirements
- Freddie Mac requirements
- FHA and VA requirements
- Lender-specific overlays
- Internal audit policies
The lender should still review, approve, test, and document each rule change. It should also preserve the rule version and effective date applied during each audit.
This creates a repeatable compliance record. The audit no longer depends only on what an individual reviewer remembers at the time of evaluation.
Conclusion
The 10% sample was never proof that one loan in ten is the ideal level of post-close scrutiny. It became a common floor and operating pattern because full-file evaluation consumes significant staff time. For many lenders, the sample size reflects capacity as much as risk strategy.
Infrrd changes that equation. MortgageCheckai prepares and validates loan data, while Ally automates nearly 80% of repetitive QC work and helps complete a 10-step audit in under 30 minutes. That allows the same team to review 30% or more of production without scaling headcount at the same rate.
As investor and regulator scrutiny rises, lenders that expand evaluation coverage now create stronger evidence for the future. They can show which loans were checked, which rules were applied, what failed, and how the team responded.
That record matters when a defect appears months or years after closing.
