AI Is Getting a Bigger Role in Home Valuations. Borrowers Need a Better Way to Challenge It

Share this article!

Property valuation has always been one of the stranger parts of the mortgage transaction for consumers. A buyer can negotiate a purchase price with a seller, provide months of financial documentation to a lender and get all the way through underwriting only to have another number suddenly become very important: what somebody — or increasingly, some combination of data and technology — believes the property is worth.

That process is changing quickly.

In March, President Donald Trump signed an executive order directing federal financial regulators, including the Federal Housing Finance Agency, to consider modernizing appraisal regulations and guidance to expand the use of alternative valuation models, desktop and hybrid appraisals, and artificial-intelligence valuation tools. The order also directs regulators to consider reducing appraisal requirements for certain low-risk transactions and setting clearer appraisal timelines.

There are good reasons to pursue some of this. Traditional appraisals take time and cost money. They also depend on human judgment, which is hardly immune from error, inconsistency or bias. Technology can analyze large amounts of property and market data quickly, and the mortgage industry should use better tools when those tools produce reliable results.

My concern isn’t that a computer may play a larger role in determining property value. Automated models are already part of mortgage lending. What concerns me is whether the consumer’s ability to understand and challenge a valuation will keep pace as the process becomes less dependent on a traditional appraisal report and more dependent on models, proprietary data and automated decisions.

That is where I think the next phase of valuation modernization deserves considerably more attention.

This is already happening

Fannie Mae’s current valuation framework includes alternatives to a traditional appraisal. For qualifying loans, Desktop Underwriter can offer value acceptance or value acceptance plus property data. Under the latter option, trained and vetted data collectors perform an interior and exterior property-data collection, but an appraisal is not required. Fannie also permits desktop appraisals in eligible transactions, allowing an appraiser to develop the valuation without a traditional interior and exterior inspection.

Freddie Mac has its own Automated Collateral Evaluation, or ACE. For certain mortgages evaluated through Loan Product Advisor, a lender can receive and accept an ACE offer and originate the mortgage without an appraisal. Freddie says ACE relies on proprietary models, historical data and public records.

None of this means lenders are simply allowing an algorithm to invent a home value. These programs have eligibility requirements and controls, and different properties and transactions can require different valuation approaches.

It does mean the mortgage industry’s definition of property valuation has become much broader than sending an appraiser to a house and waiting for a traditional appraisal report.

That evolution is likely to continue. The administration is now explicitly asking regulators to consider wider use of artificial-intelligence valuation tools. Before we accelerate much further, I would like us to spend as much time thinking about what happens when technology gets an individual property wrong as we spend thinking about the time and money it can save when it gets one right.

Automated valuations are already regulated

Precision matters here because AVMs are not operating in some regulatory Wild West.

Six federal agencies adopted final quality-control standards for automated valuation models in 2024, and those requirements became effective Oct. 1, 2025. The rule applies to mortgage originators and secondary-market issuers using covered AVMs in certain credit decisions and securitization determinations involving mortgages secured by a consumer’s principal dwelling.

Covered institutions must maintain policies, practices, procedures and control systems designed to ensure a high level of confidence in AVM estimates, protect against manipulation of data, avoid conflicts of interest, provide for random sample testing and reviews, and comply with applicable nondiscrimination laws.

Those are meaningful safeguards. There is also good reason regulators adopted a framework that allows institutions to develop controls appropriate to their size, complexity and use of AVMs rather than attempting to prescribe one model-testing formula for a technology that will continue to change.

The regulatory framework, however, primarily addresses what the institution using the model must do. A homeowner encounters the valuation from a different perspective. The homeowner wants to understand what information was used and, more importantly, what can be done when material information about the property is wrong or the resulting valuation appears unsupported.

Those questions become more important as valuation technology becomes more sophisticated.

Borrowers already have a process for challenging traditional appraisals

There is useful precedent here.

Fannie Mae’s current Selling Guide requires lenders, for loans requiring an appraisal report, to maintain policies and procedures for a borrower-initiated reconsideration of value, commonly known as an ROV. The process must allow borrowers to appeal an appraisal when they believe the opinion of value is unsupported, may be deficient because of unacceptable appraisal practices, or reflects prohibited discriminatory practices. The lender remains responsible for ensuring the appraisal report and market-value opinion are reliable and adequately supported.

I think that principle is right. When a valuation can materially affect a mortgage transaction, the consumer should have a reasonable path for identifying relevant information that may have been missed or incorrectly represented.

The complication is that the standardized borrower-initiated ROV process is structured around an appraisal report and an appraiser. An AVM is a different valuation mechanism.

Regulation B already gives consumers an important protection here. For applications covered by the rule, creditors must provide applicants copies of appraisals and other written valuations developed in connection with an application for credit secured by a first lien on a dwelling.

Receiving a valuation, however, and having a standardized process specifically designed to contest incorrect inputs or a materially questionable automated result are not necessarily the same thing.

That is the issue I believe regulators and the mortgage industry should examine as automated valuation expands.

A good model can still be wrong about one house

Discussions about artificial intelligence often become less useful when everyone feels obligated to choose between believing the technology will solve everything and believing it should not be trusted at all. Mortgage lending cannot operate effectively at either extreme.

A valuation model can perform very well across a large portfolio and still materially miss the characteristics or value of an individual property. Human appraisers can make mistakes as well. Neither observation invalidates the underlying valuation method.

For the consumer, though, aggregate model performance isn’t much consolation when the property that falls outside the expected range happens to be theirs.

A home may have undergone renovations that aren’t properly reflected in available data. Public records can contain inaccurate information. Neighborhood conditions can change. Rural or unusual properties may have fewer useful comparable transactions. Any valuation methodology can encounter circumstances where the information available does not tell the complete story.

The objective shouldn’t be to pretend those errors can be eliminated. It should be to build a mortgage process capable of recognizing and addressing them regardless of whether the valuation came from a person, an automated model or a combination of the two.

Responsibility should remain with the institution using the valuation

One thing I would resist as this technology becomes more sophisticated is allowing it to blur responsibility.

A lender shouldn’t be able to treat an automated result as something beyond meaningful review simply because a computer produced it.

The existing federal AVM rule already points in the right direction because it places quality-control responsibilities on institutions using covered models. Technology does not eliminate the institution’s obligation to maintain appropriate controls around the valuation process.

I would carry that philosophy into the consumer experience.

When an automated valuation materially affects a mortgage decision, there should be a clear and disciplined way for the borrower to identify objectively incorrect property information or submit relevant information that may not have been captured.

That doesn’t mean borrowers should be entitled to keep requesting new valuations until they receive a number they like. Any review process needs reasonable boundaries. Mortgage professionals already understand the difference between correcting a legitimate factual problem and shopping for a preferred value.

If the recorded square footage is wrong, there should be a practical way to address it. If material property characteristics are inaccurate, someone should review them. If the available data is thin because the property is unusual, that may be a circumstance where additional human judgment is appropriate.

Exactly how such a process should work is a policy question worth studying. What matters to me is that the question gets answered while the technology is expanding rather than years afterward.

Modernization should make the mortgage process better

There is too much potential in valuation technology to reduce this debate to technology versus appraisers.

Freddie Mac says its ACE program can allow eligible borrowers to avoid an appraisal fee and shorten the time required to close a loan. Fannie Mae’s framework similarly provides several valuation paths depending on the property and transaction.

We don’t necessarily need the most expensive or time-consuming valuation method for every low-risk mortgage simply because that was historically the way the transaction worked. We also shouldn’t assume a faster valuation is automatically a better one because a sophisticated model produced it.

The better approach is to match the valuation method to the risk and the available information. A straightforward property with abundant reliable data may be well suited to greater automation. A unique property in a thinly traded market may require more human judgment. Between those examples are desktop and hybrid appraisals, property-data collection and other methods that combine technology with professional expertise.

The March executive order gives federal regulators an opportunity to accelerate this transition. The existing AVM rule already establishes important expectations around model confidence, data integrity, conflicts, testing and nondiscrimination.

As regulators consider what comes next, I would add another priority: make sure the person whose home is being valued has a practical way to raise a legitimate problem when material property information or the resulting valuation is wrong.

We have already recognized that principle in the traditional appraisal process. There is no reason to abandon it simply because technology is becoming a larger part of determining what a home is worth.