In many markets, from suburban housing to high-value urban assets, deal volumes have thinned from recent peaks. The slowdown has not removed the need for pricing. Loans still need to be underwritten, land still needs to be assessed, and portfolios still need to be marked. What has changed is the reliability of the signals that traditionally informed those decisions.
For years, valuation has leaned heavily on comparable sales. It relies on recent transactions of similar properties as the clearest signal of current value, and tends to work well in more active markets. When transactions slow, that signal weakens. In segments where properties are less standardised, including luxury or design-led developments, it can become difficult to establish meaningful benchmarks at all.
This gap is beginning to shape how valuation technology is being rebuilt.
A shift driven by constraints in transaction data
Companies working in property analytics are responding to the same underlying constraint. With fewer recent transactions to rely on, models are being pushed to extract value from broader and deeper datasets.
US-based property data and analytics company ATTOM has framed its latest automated valuation model around this problem. Rather than depending primarily on recent comparable sales, the company is leaning on more than three decades of transaction history, combined with local market patterns and property-level characteristics.
The model is designed to interpret how neighbourhoods have evolved over time and translate that into present-day estimates, particularly in markets where recent sales data is limited. The company reports a median error rate of 2.9% based on historical testing, with most valuations falling within a relatively narrow band of actual sale prices.
That approach reflects a broader move away from treating valuation as a near-term snapshot. Instead, it places more weight on long-term patterns and relationships within the data.
Different models, different assumptions
Other players in the space are addressing the same issue, but through different technical frameworks.
Cotality (earlier known as CoreLogic), which is deeply embedded in mortgage and underwriting systems, has focused on combining multiple valuation models rather than relying on a single method. Its approach blends statistical techniques and machine learning outputs, with an emphasis on consistency and risk assessment for lenders. In this context, accuracy is not only about proximity to sale price, but also about how reliably the model performs across varying market conditions.
HouseCanary has taken a more explicitly predictive route. Its models extend beyond current valuation to estimate future price movements, incorporating a wide set of variables including market trends, property attributes, and external data signals. This positions valuation as part of a forward-looking decision framework rather than a static estimate.
At the consumer end of the spectrum, Zillow continues to refine its widely used valuation model through large-scale data inputs and user interactions. The system benefits from continuous feedback, as listing updates and user engagement provide a steady stream of corrections and adjustments. While its outputs are not typically used for formal underwriting, they illustrate how valuation models can evolve through real-time data loops.
Across these approaches, there is also a degree of convergence. Most leading models now draw from expanded datasets that include historical transactions, property attributes, market trends, and, in some cases, imagery and behavioural signals. The differences lie less in the data itself and more in how it is prioritised. Some models place greater weight on long-term time-series analysis, others on forward-looking prediction, and some on integration into underwriting and risk workflows. Models that combine depth of historical data with direct integration into decision systems are likely to see stronger adoption in institutional settings.
As valuation models become more data-rich and incorporate deeper machine learning techniques, they also face growing pressure to remain explainable within regulated lending environments. In markets where automated valuation models influence credit decisions, lenders are required to justify how outputs are derived and validated. This creates an ongoing tension between predictive sophistication and the need for transparency in regulated financial systems.
From comparable sales to data-rich modelling
What is changing is less about replacing comparable sales altogether and more about reducing their dominance within the valuation process. In markets where transaction data is thin, models are increasingly required to interpolate rather than simply reference recent deals.
This has led to a greater reliance on time-series analysis, multi-variable modelling, and confidence scoring. Instead of producing a single figure in isolation, valuation systems now often generate estimates alongside indicators of reliability, allowing users to gauge how much weight to place on the output.
For lenders, this can support more calibrated underwriting decisions. For investors, it enables more continuous portfolio assessment. For developers, it offers a way to evaluate projects in locations or segments where comparable evidence is limited. In practice, this shifts valuation from a reference point to an active input in risk assessment and capital allocation.
Implications for the built environment
The impact of these changes extends beyond financial modelling. In parts of the market where properties are highly differentiated, traditional valuation methods have often struggled to capture nuance. Data-driven models, which can incorporate a wider range of inputs, may provide a more consistent basis for assessing such assets.
At the same time, increased reliance on algorithmic systems introduces new questions. Transparency remains a concern, particularly in regulated environments where valuation decisions need to be explained. Model performance can vary depending on data quality and local conditions, and no system fully removes the need for human judgement.
A more model-led phase of valuation
The current shift reflects a broader transition in how real estate data is used. As transaction activity becomes less uniform across markets, valuation is moving towards a more model-led approach, where historical data, real-time inputs, and analytical frameworks work alongside traditional methods.
In that context, the emergence of new valuation models is not simply a product cycle. It is a response to a structural constraint. When recent transactions no longer provide a sufficient guide, the industry is compelled to look elsewhere for clarity. The competitive edge is no longer in access to data alone, but in how effectively it is modelled and applied within decision systems.
