AI in Real Estate · Practitioner Guide

The Valuer Isn't Obsolete. The Job Description Just Changed.

By Yishuang (Sherry) Xu · October 2026 · 5 min read

Two stories dominate the conversation about automated valuation and the future of the profession. In the first, AVMs replace human valuers entirely: faster, cheaper, more consistent, and unburdened by subjective judgement. In the second, professional valuers insist that local knowledge, physical inspection, and expert intuition can never be replicated by an algorithm, and that automation is a threat to be resisted.

Both stories are wrong. The role is not disappearing. It is transforming, and the transformation is already well underway.

What AVMs actually changed

To understand where the role is heading, it helps to be precise about what automated valuation has and has not achieved.

AVMs are now very good at valuing typical properties in data-rich markets. A standard residential property in an actively transacting urban area, with abundant comparable sales and well-maintained public records, can be valued by a production AVM with a median error of seven to eight per cent, in milliseconds, at negligible marginal cost. For high-volume, lower-risk applications (portfolio monitoring, initial screening, mortgage pre-qualification), this level of accuracy at this speed and cost is transformative.

But AVMs are not good at everything. They struggle with unusual properties, thin markets, rapidly shifting conditions, and situations where data is incomplete or unreliable. They produce confident numbers that can obscure substantial uncertainty. They inherit biases from historical data. And they cannot inspect a property, notice deferred maintenance, assess the quality of a renovation, or judge the impact of a neighbouring development that has not yet been built.

What this means is that AVMs have automated the routine middle of the valuation distribution while leaving the difficult edges, where the stakes are highest and the judgement calls hardest, to human professionals. The question is not whether human valuers are still needed. It is what they are needed for.

From performing to governing

The shift can be stated simply: the valuer's primary function is moving from performing valuations to governing the systems that perform them.

This is not a demotion. Governing an automated valuation system requires a deeper understanding of valuation principles than producing individual appraisals does, because it demands the ability to reason about when and why a model might fail, not just the ability to value one property at a time.

In practice, governance means several things.

Interpreting confidence, not just value. A modern AVM does not (or should not) produce a single number. It produces a number with an attached confidence score that indicates how much the model trusts its own estimate for that specific property. A valuation professional governing the system needs to understand what a Forecast Standard Deviation of 0.12 means, how it differs from one of 0.06, and at what threshold the automated output should be overridden by a human appraisal. This is statistical literacy applied to professional judgement, and it is a skill most current training programmes do not adequately develop.

Investigating outliers. Production AVMs process millions of properties. A small percentage of those valuations will be substantially wrong, and identifying which ones requires human review. The governance role includes monitoring outlier reports, investigating cases where the AVM's estimate diverges significantly from a subsequent transaction, and determining whether the divergence signals a model deficiency that needs correction or an unusual property that needs manual handling. This is detective work, not data entry.

Testing for bias. Regulatory and professional standards now require that AVMs be tested for discriminatory outcomes. Research has documented that automated valuations produce systematically higher errors for properties in certain demographic contexts, a pattern that persists even after controlling for property and neighbourhood characteristics. The professional governing the system is responsible for ensuring that disaggregated accuracy testing is conducted, that disparities are identified, and that remediation is documented. This is a governance obligation that did not exist a decade ago.

Setting suppression policy. Perhaps the most consequential governance decision is determining when the AVM should decline to produce a valuation. Setting suppression thresholds involves balancing coverage against accuracy, commercial pressure against risk, and speed against reliability. It requires understanding the model's limitations well enough to define the boundary of its competence. This is a judgement call that no algorithm can make for itself.

The regulatory push

This transformation is not happening in a vacuum. Regulatory and professional frameworks are actively reshaping what is expected of professionals who use automated valuation.

The AVM Quality Control Rule finalised by US federal agencies in 2025 mandates that institutions using AVMs in mortgage lending adopt policies for accuracy testing, conflict-of-interest avoidance, ongoing performance monitoring, and explicit non-discrimination compliance. The rule does not ban AVMs or even restrict their use. It requires human governance of automated systems, and it places accountability for that governance on the institution and its professionals.

In the UK, the RICS Red Book standards have increased their emphasis on risk management and transparency in the use of valuation models. The International Association of Assessing Officers (IAAO) has updated its mass appraisal standards with expanded requirements for data standardisation and statistical evaluation. The direction across all three frameworks is consistent: the professional's responsibility is not diminished by automation; it is redirected towards oversight, validation, and accountability.

What this means for career development

If you are a practising valuer, surveyor, or appraisal professional, the implications for your career development are concrete.

Statistical literacy is no longer optional. You do not need to build machine learning models, but you do need to read and interpret their outputs: confidence intervals, error distributions, feature importance rankings, and performance metrics. If your CPD portfolio does not include this, it has a gap that is widening every year.

Governance and compliance skills are becoming core, not peripheral. Understanding the regulatory landscape around automated valuation (what the rules require, what "fairness testing" means in practice, how to document model oversight) is increasingly part of the professional skill set, not a specialism reserved for compliance teams.

The ability to exercise judgement at the margins is your distinctive value. The properties that AVMs handle well do not need you. The properties that AVMs handle poorly, or decline to handle at all, are where your expertise is irreplaceable. Developing deeper competence in complex, unusual, and contested valuations is the highest-value investment you can make in your professional future.

The valuer is not obsolete. But the valuer who cannot engage with the automated systems that now dominate routine valuation is heading towards obsolescence faster than they may realise.