Automated valuation models already produce property estimates in seconds, and lenders use them routinely. Yet the profession has not collapsed. Federal projections put appraisers and assessors of real estate at 3 percent growth from 2025 to 2035, about as fast as average, taking the occupation from 68,400 positions to roughly 70,800. Median pay was $67,960 in 2025, with a bachelor’s degree as the typical entry requirement.
The gap between capable technology and stable employment is the whole story here, and it comes down to what a valuation is actually for.
The short answer
An automated model estimates value from data it already holds: recent comparable sales, property characteristics on record, and market trends. A licensed appraiser produces a defensible opinion of value that a lender, a court or a tax authority can rely on, having inspected the property and reconciled what the records say against what is actually there.
Those are different products. The first is fast and cheap and adequate for low-risk decisions. The second carries professional liability and is required by regulation for the decisions that matter most. Automation has taken a large share of the first category, which is why growth is modest rather than strong. It has barely touched the second, which is why the profession is growing at all.
Who this is for
- People considering appraisal as a career, who want to know whether the licence is worth the time and cost.
- Working appraisers deciding where to specialise as automated valuation absorbs routine work.
- Anyone in property or lending trying to understand where a human valuation is still required and where a model will do.
- Career changers from adjacent fields such as real estate, surveying or lending, who already have relevant knowledge and want to know how much of it counts toward licensing.
Key benefits of the profession as it stands
- Licensing creates a genuine barrier. Appraisal is regulated, and lenders are required to use licensed appraisers for many transaction types. That requirement is written into rules rather than chosen by preference.
- Employment is stable, projected at 3 percent growth with roughly 2,400 additional positions to 2035.
- Pay is solid at $67,960 median, with substantial variation by specialism and region.
- The work is varied. Residential, commercial, agricultural, litigation support and tax assessment are meaningfully different practices.
- Independence is available. A large share of appraisers work for themselves or in small firms, which suits people who want control over their workload.
- The knowledge transfers. Understanding how property is valued is useful in lending, investment, development and tax work, so the licence opens doors beyond appraisal itself.
How it works: valuation and where models fit
Understanding the division of labour requires knowing what an appraiser does that a model does not.
An automated valuation model takes what is recorded about a property and compares it with recent sales of similar properties nearby, adjusting for differences the data captures. Given good data and a homogeneous market, this works well. In a suburb of similar houses built in the same decade with frequent transactions, a model produces estimates close to what an appraiser would conclude, at a fraction of the cost.
A licensed appraisal starts from the same comparables and then does several things a model cannot.
It verifies the property exists as described. Records are frequently wrong. An extension may be unpermitted. A stated bedroom may be a converted garage with no window. A property may have been altered substantially since the last recorded transaction.
It observes condition. Two houses with identical recorded characteristics can differ enormously in value because one has been maintained and the other has not. Condition is not in any dataset, and it is often the largest single adjustment.
It reads the local context. Whether a road is about to be widened, whether a neighbouring plot has planning consent, whether a street has a known drainage problem. Much of this comes from local practice rather than data, which is why appraisal remains a local profession.
It reconciles conflicting evidence. When comparable sales point in different directions, the appraiser decides which are genuinely comparable and explains why. That reasoning is the deliverable, not the number.
It carries liability. A signed appraisal is a professional opinion the appraiser can be held to. Lenders, courts and tax authorities rely on that accountability. No vendor of an automated model has offered to accept it.
That last point is worth dwelling on, because it explains the shape of the whole market. A lender using a model for a low-risk refinance has decided the exposure is small enough to absorb. A lender financing a commercial property, or a court dividing an estate, needs an opinion someone will defend under questioning. The technology has not changed which decisions carry that weight. It has only made the cheap end cheaper.
Where automation has taken share
Being honest about this matters. Automated models now handle a substantial volume of work that once went to appraisers: portfolio revaluation, lower-risk refinancing, initial estimates, tax assessment screening and property tracking. Several lending programmes permit appraisal waivers for qualifying transactions, which removed a category of routine residential work outright.
That is the reason growth is 3 percent rather than 10. The routine, homogeneous, low-risk end of the market has largely gone, and what remains is denser.
There is a consequence for anyone entering the field. The work that used to build experience has thinned. Straightforward suburban appraisals were how new appraisers accumulated supervised hours, and there are fewer of them. The entry route now runs through firms doing more complex work, so finding a supervisor matters more than it once did.
Proof, examples, and objections
The clearest evidence is where models fail rather than where they succeed.
Unusual properties. A converted barn, a house on a large irregular plot, a mixed-use building, a property with an unusual easement. Models depend on comparable sales. Where genuine comparables do not exist, the estimate degrades badly while still returning a confident number, which is the most dangerous failure mode a valuation tool can have.
Thin markets. Rural areas with few transactions give a model almost nothing to work from. This is precisely where appraisers remain busiest, and it is worth noting that model confidence often does not fall as sharply as accuracy does in these cases.
Condition extremes. A fire-damaged property and a fully renovated one may have identical records. Only inspection separates them.
Contested valuations. Divorce, probate, tax appeals and litigation all require a valuation someone can defend under questioning. A model output is not evidence in the way a signed appraisal is.
Rapidly moving markets. Models rely on completed sales, which lag by weeks or months. When conditions shift quickly, an appraiser reading current listings sees the change first.
A common objection is that models will improve until these gaps close. The gaps are not primarily about model quality. Condition and unpermitted alterations are not recorded anywhere, so no amount of modelling recovers them without someone visiting. And the liability question is legal rather than technical.
A second objection is that appraisal waivers show regulators are comfortable removing the human. The waivers are conditional. They apply to lower-risk transactions with strong data coverage, which supports the division described here rather than contradicting it.
A third objection deserves a straight answer. If a model estimate costs almost nothing and an appraisal costs several hundred, will clients not take the cheaper option? Many already do where they control the decision. But the party demanding the valuation is often not the party paying. Lenders, courts and tax authorities set the requirement, and their tolerance for error decides which product is acceptable.
Comparison with adjacent work is instructive. Claims adjusters, appraisers, examiners and investigators are forecast to decline 6 percent over the same decade. Real estate appraisal grows while insurance claims work declines, and the difference is that property valuation retains a regulatory requirement for a licensed opinion in the highest-value transactions.
For context on how exposure is measured, the Bureau publishes AI exposure categories for 831 occupations and states plainly that exposure “does not imply job loss, productivity gains, automation probability, or wage effects.”
One further point is worth making. The appraiser’s deliverable is shifting from the number toward the reasoning. When a client can obtain an estimate instantly and for nothing, what they buy from a professional is an explanation of why that estimate is or is not reliable for their property and their decision.
Your next step
If you are considering this profession, look at the licensing requirements in your jurisdiction before anything else. Appraisal is locally regulated, the training and experience requirements differ considerably, and the supervised hours are the real cost of entry. Find out whether supervisors are actually available where you live, because in some markets that is the binding constraint rather than the coursework.
It is also worth asking what kind of work local firms actually do. A market dominated by high-volume residential lending faces more automation pressure than one with substantial commercial or litigation practice.
Whichever route you take, the durable skill is knowing what an automated estimate is actually computing and where it goes wrong. An appraiser who can explain precisely why a model’s number is unreliable for a given property is doing something the model cannot, and that explanation is increasingly what clients are paying for. If you want a structured way to build that understanding, explore Coursiv AI lessons and check current plan details on the official site.
If you already practise, the useful move is away from the work a model does adequately. Routine residential valuation in data-rich suburbs is where automation competes hardest on cost. Complex properties, commercial work, litigation support and thin rural markets are where a defensible human opinion is still required, and where the fee reflects that.