North Dakota has 53 counties. In 2019, more than half of them had zero licensed appraisers: not one appraiser, not understaffed or struggling to recruit, but completely and irreversibly empty. A buyer in Bowman County who needed a home appraised had to wait for someone to drive in from Bismarck, 200 miles east, or from Billings, Montana, 130 miles west, and then pay a travel fee on top of the appraisal fee because no human being with the right credentials lived close enough to look at the house.
That was seven years ago, and it has gotten worse.
According to the National Association of Realtors, the median age of a certified or licensed home appraiser in the United States is 60, and eighty percent are over 50. When industry groups talk about the "appraisal workforce crisis," they tend to frame it as a pipeline problem: the certification requirements are too complicated, the apprenticeship hours too demanding, the pay too low relative to the training investment. All of that is true, but the proposed solution, increasingly, is not to fix the pipeline. It is to build machines that make the pipeline irrelevant.
Two Announcements, Three Days Apart
On May 5, 2026, ATTOM launched an AI-powered automated valuation model built from the ground up, not an incremental update to an existing AVM but a complete rebuild. It replaces the traditional comparable-sales approach with 30-plus years of time-adjusted transaction history fed through machine learning models that, in out-of-sample testing across a decade of sales data, achieved a 2.9% median absolute percentage error. More than 80% of valuations landed within 10% of the actual sale price. It covers 98 million properties across the United States and 160 million property records, and it works in low-transaction rural markets where traditional AVMs choke because there aren't enough recent comps to calculate from.
Twenty-nine days later, on June 3, Veros Real Estate Solutions announced VeroVISION, an AI system that scores a home's physical condition from listing photos. It rates individual rooms, rolls the scores into an overall property condition assessment, and claims a 93% correlation with human appraiser and inspector condition scores. Eric Fox, the company's chief economist, said it plainly: "For a long time, condition was the one variable an AVM could not accurately predict. With VeroVISION, that gap is closed."
Read those two announcements together and the trajectory becomes obvious. ATTOM solved the data problem, Veros solved the eyes problem, and what remains for the human is an open question that nobody in the industry wants to answer honestly.
26% and Climbing
Fannie Mae and Freddie Mac have been answering that question with their wallets. In Q1 2025, the Federal Housing Finance Agency expanded appraisal waiver eligibility for purchase loans from 80% to 90% loan-to-value ratios. Inspection-based waivers, where someone visits the property but doesn't need to be a licensed appraiser, are now eligible up to 97% LTV.
By February 2026, AEI Housing Center data showed that 26% of all GSE-backed loans skipped the human appraiser entirely. Fannie Mae ran a 23% waiver rate, Freddie Mac 29%. For no-cash-out refinances, the number hit 34-35%.
Run that math. With roughly 5.5 million existing home sales and 680,000 new-build sales expected in 2026, a 26% waiver rate means approximately 1.6 million real estate transactions this year where no licensed appraiser will set foot in the house, drive the neighborhood, or physically inspect the roof, the foundation, or the furnace. A machine decides the value, a machine decides the condition, and a machine decides whether the bank's collateral is worth what the borrower says it is.
UAD 3.6 and the Data Extraction Play
Meanwhile, the appraisers who remain are being asked to do something curious. New Uniform Appraisal Dataset 3.6 requirements demand hyper-granular data: room-by-room material ratings, finish classifications, fixture-level detail, micro-condition scoring at a resolution that no previous appraisal form has ever required.
Desiree Mehbod, founder of Appraisers Blogs, called it what it is: "UAD 3.6 turns every full appraisal into a data-mining operation, with the appraiser acting as the human data-collection device for a system that wants our expertise now so it can automate it later." She's describing a classic technology adoption pattern. You hire humans to train the machine, and once the machine is trained, you don't need the humans anymore. In appraisal, the training data is the granular condition and material information that UAD 3.6 now requires, exactly the kind of detail Veros's VeroVISION needs to close that last accuracy gap.
And this is not speculation, because Veros already built the GSE Uniform Collateral Data Portal, the system that ingests every appraisal submitted for government-backed loans. It is, functionally, both the entity demanding more granular human data and the entity building the AI that will replace the humans providing it.
What Happens When the Machine Is Wrong
Before Dodd-Frank, a woman identified as Mrs. R in congressional testimony described what happened when an AVM valued her $34,000 home at $84,000. A lender used that valuation to issue a loan she couldn't afford. She lost the house. No human appraiser visited the property before the valuation was accepted, and the gap between the machine's number and reality was 147%.
Today's AVMs are significantly more accurate, and ATTOM's 2.9% median error represents a different universe from the experience Mrs. R described to Congress, but median error is a misleading comfort. On a $500,000 home, 2.9% is $14,500, a rounding error in a hot market. On the same home, the tail end of the distribution, the cases outside the 80% accuracy window, could mean $50,000 or more in overvaluation. And the tail cases tend to cluster in exactly the markets where the data is thinnest: rural areas, neighborhoods with few recent sales, communities where the housing stock is older and more idiosyncratic.
Those are also the communities where the appraisers disappeared first.
Nobody talks about that part.
Restb.ai is working on one piece of this puzzle. Its computer vision system identifies room materials, ceiling heights, and property condition from photos, and it can blur faces, artwork, and mirrors to reduce racial bias in appraisals, a real and documented problem that human appraisers have demonstrably failed to solve. Tony Pistilli, the company's GM of valuations, frames it as enabling "lenders, appraisers, and homeowners to trust that property valuations are based on hard data rather than personal perceptions."
Fair point. But hard data with a 7% error rate on off-market homes, which is Zillow's Zestimate accuracy when the property isn't listed, means the machine's "hard data" disagrees with reality by $35,000 on a $500,000 house. For a buyer financing that purchase at 90% LTV, that discrepancy is the difference between a viable mortgage and a loan that starts underwater.
An Extinction Timeline
Nobody in the industry publishes one, so I built a rough model. If 80% of the roughly 78,000 active certified and licensed appraisers are over 50, that's 62,400 appraisers approaching retirement age within the next 15 years. Assuming retirement rates track general labor force patterns for aging professionals, and subtracting the trickle of new entrants coming through a certification pipeline that the Appraisal Subcommittee itself admits is broken (the ASC is operating with nearly 30% fewer staff and hadn't met for seven months despite guidance requiring quarterly meetings), the profession hits a critical threshold between 2032 and 2035, not a shortage but a structural absence, the point at which AI appraisal stops being an option and becomes the only option. And every regulatory expansion, every waiver increase, every new AVM launch is accelerating the timeline by making it rational for young workers to choose other careers. Why spend 2,000 hours in apprenticeship for a job that Fannie Mae is systematically waiving? Nobody does.
What Buyers and Builders Should Know
If you are selling a newly built or recently renovated home, AI photo-scoring systems like VeroVISION are beginning to feed into the valuation chain. Your finishes, your materials, your curb appeal are being scored by algorithms trained on millions of listing photos, and that score influences whether a lender accepts an appraisal waiver or demands a full human inspection. Granite countertops and luxury vinyl plank might score differently than you expect when a machine is doing the looking.
If you are buying, particularly with less than 20% down, you should ask your lender whether your loan will use an appraisal waiver or a full appraisal. If it's a waiver, understand that no human professional physically inspected the property's condition. ATTOM's 2.9% median error is impressive, but you should know whether your house is in the accurate 80% or the uncertain 20%, and no AVM currently tells you that with individual-property precision, though ATTOM does provide a per-valuation confidence score that gives directional guidance.
If you are a builder, the appraisal workforce crisis is your problem even if you never think about it. Delayed appraisals hold up closings, inaccurate appraisals kill deals, and as the human appraiser population shrinks, the AVMs that replace them will have less human-validated training data to learn from, which means accuracy in edge cases, custom builds, unusual materials, properties in transitional neighborhoods, could degrade precisely when it matters most.
Limitations
This analysis relies on NAR's median-age statistic for appraisers, which is several years old and may not reflect post-pandemic shifts in workforce composition. Retirement rate projections use general labor force patterns rather than appraiser-specific actuarial data, which doesn't exist publicly. ATTOM's 2.9% median error and Veros's 93% correlation are vendor-reported figures; independent verification of either claim is limited. AEI's 26% waiver rate represents a single month (February 2026) and may not reflect annual trends. A January 2026 arXiv paper found that LLMs approach traditional ML accuracy for house price prediction but produce overconfident price intervals, suggesting that AI valuation certainty may be systematically overstated in ways current error metrics don't capture.
In Defense of the Appraiser
Matt Krodel at Cotality, the company formerly known as CoreLogic, told Working RE that AI will free appraisers to "spend more time on analysis, analytics, and market reporting, because there won't be so much manual work." Jeff Bradford of Bradford Technologies insists that "appraisers can be the architects of the valuation. AI is just an assistant."
Maybe. But neither of those visions addresses the demographic reality. You cannot position AI as an assistant to a workforce that is aging out and not being replaced. What Krodel and Bradford are describing is a transition period, a comfortable decade where experienced appraisers use AI tools to work faster and more accurately, and then they retire, and the tools keep running without them. That's not AI-as-assistant; it's AI-as-successor wrapped in a polite handoff narrative.
Bipartisan legislation introduced by senators from North Dakota and Arizona would add trainee positions and decrease certification fees to attract new appraisers. It is a sincere effort, but even its sponsors seem to understand they are bailing water from a boat that the industry has already decided to scuttle. When Fannie Mae is waiving appraisals at 90% LTV and Veros is announcing 93% correlation with human scores, a bill to make certification slightly cheaper is not a workforce strategy. It is a eulogy dressed up as policy.
Marcus Washington covers workforce and labor in the construction industry. His family has worked union trades for three generations. He worries about the workers.