Will AI Replace Appraisers? What AVMs Can and Can't Do

Will AI replace appraisers? You hear the question at the closing table, from your nephew, and maybe from yourself at 10pm after the third revision request this week. It's a fair question. Lenders already skip appraisals on some loans. Automated valuation models produce a number in seconds. Software vendors (us included) say "AI" a lot. Here is an honest answer from people who build appraisal software and grew up around an appraisal firm: AI will take some work away, it will change a lot of work, and it will not replace the person who has to stand behind a value.

Key takeaways

• An automated valuation model (AVM) predicts a price from data. It doesn't inspect, verify or reason about a specific house.
• AVMs and appraisal waivers already handle some low-risk loans. That trend predates the current AI wave.
• A federal rule on AVM quality control took effect October 1, 2025. It covers lenders and issuers, not appraisers developing appraisals.
• The BLS projects appraiser and assessor employment to grow about as fast as average through 2034.
• The safe bet for appraisers: let software do the clerical work, and build your practice around the judgment a model can't supply.

What an automated valuation model actually is

An AVM is a statistical model. It takes in what it can find about a property (address, lot size, GLA, beds, baths, year built from public records, recent nearby sales, sometimes listing data) and outputs a predicted price, often with a confidence score.

The federal rule defines it simply: any computerized model used by mortgage originators and secondary market issuers to determine the value of a consumer's principal dwelling collateralizing a mortgage. Older AVMs used hedonic regression. Newer ones use machine learning. The idea is the same: find the pattern in many past sales and apply it to one more house.

What AVMs do well

• Speed and scale. A lender can value a whole loan portfolio overnight.
• Typical homes in active markets. A 1990s three-bedroom ranch in a subdivision with 40 sales in the last six months is a pattern problem, and models are good at pattern problems.
• Screening. Flagging loans that need a closer look, or catching a value that looks out of line.

What AVMs can't do

• See condition. If the tax roll says the house is fine, the model thinks the house is fine. It doesn't know the basement floods, the roof is at the end of its life, or the kitchen was gutted last month.
• Verify the data. Public records can be years out of date. An unpermitted addition, a converted garage or a finished basement may not be there at all. The model trusts its inputs.
• Handle thin markets. Rural property, unique homes, mixed-use lots, the house on 40 acres with a pole barn: few comparable sales means weak patterns.
• Explain itself in a way a reviewer can test. An appraiser can say why comp 3 got a $15,000 view adjustment. A model's reasoning is spread across many variables.
• Exercise judgment at a turn. When a market shifts quickly, the model is learning from sales that already happened. An appraiser who talks to agents and watches pendings sees the turn sooner.
• Take responsibility. An appraisal is a credentialed person's opinion, developed under USPAP, with a signature and a license behind it.

Note: A comp three miles out with a big location adjustment, explained in two sentences of your commentary, is something a reviewer can test. An AVM's "confidence score" is not the same kind of support.

The 2024 interagency AVM rule

For years the AVM was a black box from the appraiser's point of view. That changed when six agencies (the OCC, the Federal Reserve, the FDIC, the NCUA, the CFPB and the FHFA) issued a final rule implementing the AVM quality control standards Congress required in the Dodd-Frank Act. It was published August 7, 2024, and took effect October 1, 2025.

Lenders and secondary market issuers that use AVMs for credit decisions must maintain policies, practices, procedures and controls designed to:

1. ensure a high level of confidence in the estimates,
2. protect against the manipulation of data,
3. seek to avoid conflicts of interest,
4. require random sample testing and reviews, and
5. comply with applicable nondiscrimination laws.

The FDIC's financial institution letter on the rule is a short, readable summary if you want the regulator's own words.

For appraisers, two things matter. First, the CFPB's compliance guide says the rule doesn't cover "the development of an appraisal by a certified or licensed appraiser." Your work stays under USPAP and your state board. Second, the rule is a signal. Regulators are treating automated value as something that needs testing, independence and fairness checks, the same values appraisers have been held to for decades.

Where automation has already replaced appraisals

Be clear-eyed: some loans that once got a full interior appraisal don't anymore.

• Value acceptance (appraisal waivers). On eligible loans, Fannie Mae's DU and Freddie Mac's ACE can offer to skip the appraisal. The lender relies on the GSE's data and models.
• Value acceptance + property data. Fannie Mae's VA+PD sends a trained, vetted data collector to observe the interior and exterior. No appraisal is obtained, and Fannie Mae accepts the lender's value estimate. Freddie Mac's ACE+ PDR works on similar lines.
• Desktop and hybrid appraisals. An appraiser still forms the opinion, but someone else (or no one) does the physical inspection. We cover these scopes in hybrid, desktop and bifurcated appraisals.

Notice what these programs keep: a person on site when condition matters, and an appraiser when the risk is higher. The automation is in deciding when an appraisal is needed, not in producing one.

The jobs numbers

The Bureau of Labor Statistics projects employment of property appraisers and assessors to grow 4 percent from 2024 to 2034, about as fast as the average for all occupations, with about 6,300 openings a year on average over the decade. In BLS projections, most openings in a field like this come from replacing people who retire or change careers, not only from growth.

Projections aren't promises, and volume swings hard with interest rates. Anyone who lived through the 2020 refi boom and the 2022 rate shock knows that. But nothing in the official outlook describes a profession being automated out of existence. The bigger near-term problem in many markets is getting trainees through to licensure. See the appraiser shortage and the trainee bottleneck.

What will change for appraisers

AI won't take the signature. It will change the work around it.

The pattern is the same one appraisers lived through with forms software, digital photos and UCDP. The clerical work shrinks. The judgment stays. The overview of AI appraisal software in 2026 goes deeper on which tools help and which ones overreach.

How to make yourself hard to replace

1. Be the person for the hard assignments. Rural, unique, high-value, multi-unit, condition-driven, litigation. These are exactly where models are weakest.
2. Support your adjustments with data. "Market derived" with nothing behind it reads like a rule of thumb. Paired sales, grouped data and your own history read like analysis. Our guide to data-driven adjustments from your own past reports shows how.
3. Get fast at the clerical part. Let software fill parcel data and check the UAD 3.6 compliance rules, so your hours go into analysis.
4. Learn the new scopes. Hybrid and desktop work is real work. Knowing how to read a property data report critically is a skill.
5. Write commentary a reviewer can test. Short, specific, tied to the data. That's the opposite of what a model produces.

Tip: When a lender or borrower quotes an online estimate at you, don't argue with the number. Explain what you saw that the model couldn't: the condition, the location influence, the comps it didn't have.

The honest answer

Will AI replace appraisers? Not the appraiser who inspects, analyzes, reconciles and signs. It will replace some appraisals on low-risk loans, as waivers already have. It will replace hours of retyping and checking, which is good news for anyone paid a flat fee. And it raises the bar for what an appraiser's report needs to show, because the reader can now get a number for free. The value of an appraisal is in the reasoning behind the number.

That's how we built Carat: automation for the clerical work, AI-powered checks and adjustment history from your own reports, and the value opinion left to you. If you want software that works that way, you can create an account.