A model can produce a number. It cannot be responsible for one.
AI is good at finding comparable sales, checking reports for errors and estimating values in bulk. An appraisal still needs a licensed appraiser who looks at the home, supports the value and signs.
Faster computers will not change that. The limit has to do with who answers for the number.
Where models help, and where they struggle
Models do well when the data is plentiful and easy to check. They struggle when it is thin, disputed or never recorded at all.
Finding candidate comparable sales
Sorting thousands of recent sales for likely matches is a job for a machine. It saves an appraiser hours.
Catching inconsistencies
Software can flag a report where the sketch, the room count and the stated living area disagree. Checking rules across many reports is a clear win.
Reading unstructured records
Pulling permits, deeds and listing history into structured fields is tedious work. Software handles it well.
Estimating at portfolio scale
A lender reviewing thousands of properties can use a model that is roughly right on average, even if any one estimate is off.
Condition it has never seen
Two identical houses on one street can be decades apart in condition. The records cannot tell them apart, so the model cannot either.
Thin and unusual markets
Acreage, custom builds, mixed use, anything rare. Few comparable sales means little to learn from. Confidence drops where the stakes are often highest.
Explaining itself
An appraisal has to show why the market supports each adjustment. A number with no reasoning behind it does not meet that bar.
Carrying the consequence
Someone has to answer for a wrong value. Licenses, professional standards and liability attach to a person, never to a model.
The line is accountability, not accuracy
It is tempting to see this as a race that models win once they get accurate enough. Suppose a model did beat appraisers on average. It still could not hold a license, face a state board or defend a value with supported reasoning.
Lending depends on someone being answerable. Better software does not change that.
How the work splits today
The machine searches, extracts and checks. The appraiser selects the sales, adjusts them and reconciles them into one value.
Say software turns up forty candidate sales. The appraiser picks six and explains why those six. The search is mechanical. The choice is not.
Most serious appraisal software already works this way. A free estimate on a consumer website does not. That is a model alone, with nobody standing behind it.
AI is also changing how clients find an appraiser. Some now ask a chatbot who to hire instead of typing into a search box. For appraisers curious about that side of the shift, there is a plain guide to how appraisers show up in ChatGPT and AI search.
When one house needs one answer
A model can tell a lender roughly where a thousand houses sit. One home, with a value someone has to stand behind, needs a licensed appraiser who looks, weighs the evidence and signs. For residential property anywhere in the country, one route is a human appraisal from Pocket Appraisal, a national residential appraisal service.
Common questions
Can AI do a home appraisal?
Are appraisers using AI already?
Why can't a model just learn condition?
Will lenders stop requiring appraisals?
- The federal AVM quality control ruleWhat the 2024 interagency rule asks of lenders, and what it leaves out.
- Can AI sign an appraisal?Why the signature still belongs to a credentialed person.
- Where AI helps an appraiser todayPhoto review, data checks and comparable search, with the appraiser deciding.
- Bias and fairness in automated valuationWhat federal agencies have said about models and fair housing.
Building something for valuation?
Tell us which part of the valuation problem you are working on.