AVMs
One of the most significant ways in which AI is reshaping valuation practices is through the use of Automated Valuation Models (AVMS). While not all AVMs use AI (some are rule-based and follow fixed calculations), many modern AVMs now use machine learning to analyse large datasets and generate property value estimates. AVMs are digital tools that can estimate property values by using machine learning to analyse vast datasets. These data-driven models are increasingly being used to support both lending decisions and professional valuations.
AVMs typically draw on a wide range of data, including:
- Comparable sales
- Historical transaction data
- Property characteristics
- Geospatial and economic data
Some models also generate confidence ratings alongside value estimates.
AVMs are commonly used by mortgage lenders, particularly for remortgages or low-risk loans, as they offer faster, automated assessments that often fall within an acceptable margin of error. Surveyors also use AVMs to support valuations—but always with scrutiny of the inputs, limitations, and appropriateness for the specific case.
It is also important to note that not all AVMs use AI. Traditional AVMs may rely on rule-based calculations which apply fixed formulas to structured data. However, modern AVMs often incorporate machine learning to allow the model to learn from historical patterns and improve accuracy overtime.
To generate accurate valuations, AVMs rely on a wide range of data. The availability of data varies significantly across sectors and across markets globally. When valuing residential properties, data is often publicly available; however, when valuing commercial properties, data is often less widely available and the process tends to be more complex, making AVMs less effective and reliable.
Considering the wide and varied amount of data required to generate the valuation, AVMs are more reliable for standard, homogeneous properties. For example, they typically don't account for property condition or specific features at scale, and they can overlook specific details that a human valuer would identify.
A key limitation of AVMs – and AI in general – is that the output is only as good as the data that the model is trained on. If the foundational data is outdated, incomplete, or biased, the model’s output will have weaknesses.
A final consideration is that AVMs process sensitive information such as financial, geographic, and personal data. If used incorrectly, this could introduce risks related to data privacy, security breaches, or non-compliance with data protection regulations.
When dealing with high-risk activities—such as valuations that inform lending decisions, legal disputes, or investment decisions—AVMs should only be used to support the valuation process, not to replace or undertake it entirely. The expertise of qualified professionals remains critical for interpreting the data, identifying context-specific factors, and ensuring the valuation is accurate, fair, and compliant with professional standards. The RICS Red Book is the benchmark for competent valuation practice, particularly when dealing with complex, high-value, or contentious valuations where a fully qualified valuation professional is essential.
Principal author: Aaliyah Pollock