Automated valuation model (AVM)

An automated valuation model estimates what a property is worth using data, recent comparable sales, property characteristics, tax records, and location factors, run through a statistical or machine-learning model, rather than sending a human appraiser to inspect the property. Online home-value estimates that update automatically as market data changes are the most familiar consumer-facing example.

Most AVMs rely on some form of regression analysis or, increasingly, other machine learning techniques, trained on historical sales to learn how features like square footage, location, and condition relate to sale price, then applied to a target property using its known characteristics and nearby comparable sales. This differs from a formal appraisal, which involves a licensed appraiser physically inspecting the property and exercising professional judgment about condition and unique features a model cannot see.

AVMs matter because they are fast and cheap compared to appraisals, and are widely used for initial estimates, portfolio monitoring, and some lending decisions, though regulators generally still require a full appraisal for many mortgage transactions. For income-producing property, an AVM's output is often cross-checked against an income-based estimate derived from capitalization rate and occupancy rate. A well-documented pitfall is degraded accuracy for unusual properties, those with atypical features, renovations, or in areas with too few recent comparable sales for the model to learn from reliably.