Compliance
AVM quality control standards — effective October 1, 2025
The federal banking agencies, the CFPB, NCUA, and FHFA adopted a joint rule requiring institutions that use automated valuation models in certain mortgage credit decisions to maintain policies and controls addressing five specific factors. Compliance became mandatory on October 1, 2025. ValPro was engineered around those five factors, and the summary below is written so a lender can place it directly in a vendor due-diligence file.
01
Ensure a high level of confidence in the estimates produced
Every ValPro valuation reports its own statistical evidence: sample size, dispersion of the comparable set, median absolute percentage error, and a High / Moderate / Low confidence tier with a corresponding value range. Estimates that fall outside our tolerance are flagged rather than presented as settled conclusions, so a reviewer always knows how much weight the number can carry.
02
Protect against the manipulation of data
Inputs come directly from licensed third-party public-record and MLS sale feeds retrieved at the moment of the request. Subject characteristics and comparable sales cannot be typed in, overridden, or re-run for a better answer by the requesting party. Comparable selection, adjustment, and reconciliation are executed server-side by fixed code, and each search is written to an immutable log with the requesting account, timestamp, address, and indicated value.
03
Seek to avoid conflicts of interest
ValPro is compensated per valuation at a contracted rate. Our fee does not vary with the value produced, whether a loan closes, or whether a client is satisfied with the outcome. We hold no interest in the subject properties, take no brokerage or lending position in the transactions we value, and no client may direct a value or request revision of a completed estimate.
04
Require random sample testing and reviews
A random sample of completed valuations is reviewed by certified appraisers on staff for reasonableness of the comparable set and adjustments. In parallel, sampled estimates are back-tested against subsequent recorded closings to track median error and directional bias by submarket. Findings drive documented model changes rather than case-by-case overrides.
05
Comply with applicable nondiscrimination laws
The model is quantitative only. It receives physical property characteristics, location geometry, school attendance data, and recorded sale facts. It never receives, stores, or infers race, color, religion, national origin, sex, familial status, disability, age, or any proxy for a protected class, and no borrower or occupant identity is transmitted to the model at any point.
Data only. No judgment about people.
Bias risk in valuation historically enters through subjective narrative and through neighborhood-level assumptions that stand in for the people who live there. ValPro removes both. There is no free-text opinion, no discretionary neighborhood rating, and no demographic layer in the model. What the model can see is listed below in full.
| Input | Status |
|---|---|
| Living area | Used |
| Bedrooms / bathrooms | Used |
| Lot size | Used |
| Year built / effective age | Used |
| Garage spaces | Used |
| Distance & directional quadrant | Used |
| School attendance area | Used |
| Recorded sale price & date | Used |
| Race, ethnicity, national origin | Never collected |
| Religion, sex, familial status, age, disability | Never collected |
| Neighborhood demographic composition | Never collected |
| Borrower or occupant identity | Never transmitted |
A ValPro estimate is a statistical opinion of value. It is not an appraisal. An appraisal by a licensed appraiser may be required for some federally related transactions. This page describes ValPro's own controls and is not legal advice regarding an institution's obligations under the rule.