A builder in Boise closed on a 2,400-square-foot spec home last spring. (Scenario composited from NAHB member reports; the numbers are representative, not one transaction.) Contract price: $465,000. Spray-foam insulation in every cavity, a 9.6kW rooftop solar array, EV-ready 200-amp panel, triple-pane windows. His buyer’s lender ran an automated valuation model. It came back at $418,000. A $47,000 gap. Her buyer had two options: bring $47,000 extra to closing, or walk. She walked.
That gap wasn’t a market correction. It was an algorithm starving for data.
What Six Agencies Actually Regulated
In July 2024, six federal agencies, the FHFA, CFPB, Federal Reserve, FDIC, NCUA, and OCC, finalized quality control standards for automated valuation models. AVMs are the algorithms that estimate a home’s market value without a human appraiser setting foot on the property. Banks use them for loan origination, portfolio monitoring, and securitization. Fannie Mae and Freddie Mac increasingly accept them in lieu of traditional appraisals.
The new rules require institutions to adopt policies ensuring AVMs produce “credible” estimates, avoid conflicts of interest, undergo random sample testing, and comply with nondiscrimination standards. On paper, sensible. In practice, process requirements without accuracy benchmarks. The rule tells lenders to test their models. It does not tell them what an acceptable error rate looks like for a four-month-old subdivision where nothing has sold yet.
AVMs and the Comp Problem
Every AVM runs on comparable sales. CoreLogic, Black Knight, Zillow. Different vendors, different proprietary adjustments, same foundational dependency: recent sale prices of similar homes nearby. For existing housing stock in established neighborhoods, this works reasonably well. Zillow’s own data shows its Zestimate carries a median error of 2.4% for on-market homes and 7.5% for off-market ones.
New construction breaks this. Here is why, in order of severity:
No comps exist. Phase 1 of a new subdivision has zero closed sales. It reaches for the nearest transaction data, which might be a 15-year-old ranch house a mile away. The algorithm doesn’t know your development exists. It only knows what has already sold.
Upgrades are invisible. Tax records, the primary data feed for most AVMs, capture square footage, lot size, bedroom count, year built. They do not capture the difference between Grade I spray-foam insulation and Grade III fiberglass batts. Or between a standard 100-amp panel and an EV-ready 200-amp service. Or between builder-grade vinyl windows and Marvin Ultimates. An AVM cannot value what it cannot see.
Construction quality is unverifiable remotely. Fannie Mae now permits desktop appraisals for some transactions, including certain new construction. A desktop appraiser reviews photos and data without physically entering the home. They cannot verify that the insulation was installed correctly, that the HVAC system actually achieves the SEER rating on the spec sheet, or that the foundation pour met compressive strength requirements. An AVM has even less visibility than a desktop appraiser: it doesn’t look at photos at all.
How Much This Costs You
When an AVM undervalues your new construction purchase, the math is straightforward and unpleasant. You signed a contract at the agreed price. Your lender will only lend against the appraised value. That delta is yours to cover in cash, or the deal dies.
| Scenario | Contract Price | AVM Value | Gap You Cover |
|---|---|---|---|
| Modest undervaluation | $420,000 | $399,000 | $21,000 |
| Significant (new subdivision) | $465,000 | $418,000 | $47,000 |
| Severe (custom features) | $550,000 | $480,000 | $70,000 |
Assumptions: conventional 80% LTV financing. Gap = contract price minus appraised value. In the “severe” case, the buyer chose $70,000 in upgrades the AVM literally cannot detect. These are not edge cases. The NAHB’s top legislative priorities have included appraisal reform for years because their members report appraisal shortfalls as a recurring barrier to closing new-construction sales.
Median new home price: roughly $420,000 according to Census Bureau data. Note the proximity: the federal appraisal exemption threshold, set in 2019, is $400,000. Below that line, a lender can close with just an AVM evaluation. No human appraiser required. As new home prices cluster near $400,000 in Sun Belt production markets, the regulatory pressure to skip human appraisals and rely on algorithms grows. For the homes most likely to lack comps.
The Honest Case for AVMs
Before dismissing automated valuations entirely, acknowledge what they fix. The CFPB’s PAVE task force documented systematic racial bias in human appraisals. Homes in majority-Black neighborhoods were appraised lower than comparable homes in white neighborhoods, by the same appraisers using the same methodology. Algorithms do not drive through a neighborhood and adjust their estimate based on who lives there. Or at least, they are not supposed to, and the new federal rules explicitly require nondiscrimination testing.
AVMs also save $300 to $500 per transaction in appraisal fees, shave days or weeks off the closing timeline, and address a genuine workforce shortage. With only 77,300 active appraisers nationally, a median age that industry groups estimate is well above 55, and a training pipeline that requires a bachelor’s degree plus years of on-the-job experience, the industry cannot produce enough human appraisers to service every transaction. AVMs will expand. The question is not whether, but how much damage they cause at the margins.
What You Can Do
1. Demand a full appraisal. If you are buying new construction, do not accept an appraisal waiver or desktop appraisal. Yes, it costs $400 to $600. A human appraiser who physically enters the home can identify and value upgrades that an AVM cannot. For a $420,000 purchase, that fee is 0.1% of the transaction. Pay it.
2. Build the comp package for the appraiser. If your home is in a new subdivision with few or no closed sales, the appraiser will struggle to find comps. Help them. Gather the builder’s contract data for other lots in the same phase. If the builder has closed five homes at $430,000 to $470,000 in adjacent phases, those are comps. Appraisers can use pending sales and builder contracts if they are properly documented. Most buyers do not know this. Most real estate agents do not tell them.
3. Document every upgrade with receipts and specifications. Create a binder: solar array proposal with cost breakdown, insulation R-values, window specifications, HVAC SEER ratings, panel amperage. Hand it to the appraiser at the inspection. If the upgrade cost $22,000 and adds measurable energy savings, that is data the appraiser can use. Without it, they are guessing.
4. Know the reconsideration of value process. If the appraisal comes in low, you can file a Reconsideration of Value (ROV) with the lender. Provide specific comparable sales the appraiser may have missed, corrections to factual errors, or documentation of features not captured in the report. ROVs are not appeals in the legal sense. They are requests for the appraiser to review additional data. They succeed more often than buyers think, particularly when the buyer provides comps the appraiser lacked.
5. Budget for the gap. If you are buying in a new subdivision, assume the appraisal will come in 5% to 10% below contract price. Set aside the difference. If the appraisal matches, you have a reserve fund. If it doesn’t, you can close without renegotiating or walking.
What This Analysis Did Not Prove
Quantifying the exact rate of AVM undervaluation for new construction is harder than it should be. AVM accuracy data is self-reported by vendors. Zillow publishes its Zestimate error rates; CoreLogic and Black Knight publish less granular performance metrics. No independent body audits these models against new-construction outcomes specifically.
My gap estimates assume conventional financing at 80% LTV. FHA, VA, and jumbo loans each carry different appraisal requirements and different tolerances for valuation shortfalls. The $400,000 exemption threshold applies to federally regulated lenders, not to all mortgage originators.
Appraiser workforce data from the BLS does not differentiate residential from commercial appraisers. The 77,300 figure includes assessors working in county tax offices, who do not perform individual property appraisals for lending. The residential appraisal workforce is smaller than 77,300, but by how much is unclear.
And the regulatory landscape is moving. CFPB Director Chopra stated that “automated models can embed the very human bias they are meant to correct.” More rulemaking is likely. Whether it addresses the new-construction comp gap specifically, or focuses on the nondiscrimination provisions that attract more political attention, remains to be seen.
Sources
- FHFA et al., “Final Rule on AVM Quality Control” (July 2024): Six-agency final rule requiring quality control standards for automated valuation models
- CFPB, “Algorithms, AI, and Fairness in Home Appraisals” (June 2023): Federal assessment of AVM bias risk and nondiscrimination requirements
- Federal Reserve, “$400,000 Appraisal Exemption Threshold” (2019): Regulatory threshold for residential transactions exempt from appraisal requirements
- BLS, “Appraisers and Assessors of Real Estate” (2024): 77,300 active professionals, median pay $65,420, 4% projected growth
- Fannie Mae, “Desktop Appraisals”: Expanded desktop appraisal eligibility including certain new construction
- NAHB, “Improving the Appraisal Process”: Builder advocacy for appraisal reform citing recurring new-construction valuation shortfalls