Valnow
AVM

How a commercial real estate AVM values a property

Automated valuation models are common in housing, but commercial property is harder to model. Here is what goes into a commercial AVM, how to judge one, and where it needs a person.

Illustration of a network of data points feeding one value

An automated valuation model (AVM) estimates a property's market value from data, without a person forming the opinion. Residential AVMs have been used for decades, helped by millions of similar homes and frequent sales. Commercial real estate has fewer sales, more varied buildings and values driven by income. A commercial AVM has to be built differently.

1. It starts with verified transactions

A model is only as good as the sales it learns from. Commercial sales data is full of traps: portfolio sales where one price covers several buildings, partial interests, sale-leasebacks, related-party transfers, and the same sale recorded twice. A serious commercial AVM researches and verifies each sale before it enters the training data, and removes duplicates and non-market transactions.

Valnow's models are trained on more than a decade of researched and verified US commercial sales across office, retail, industrial and multifamily.

2. It describes each property with features

The model needs a consistent description of every property. Typical features include:

  • location: coordinates, MSA, county and ZIP-level market indicators;
  • physical characteristics: property type and subtype, building size or units, site size, year built, condition;
  • income: net operating income when it is known;
  • timing: the market conditions at the date of value.

Feature choice also matters for fairness. Variables that act as proxies for protected characteristics, such as the demographic composition of a neighborhood, should not be inputs to a model used in lending.

3. Machine learning finds the relationships

Instead of a fixed formula, modern AVMs use machine-learning models that learn from thousands of sales how size, age, location and income combine into price. Commercial AVMs often use more than one model: for example, one for properties with a known NOI, where income explains much of the value, and one for properties without it.

Markets move, so the models must be retrained regularly on new sales. Valnow retrains its models every month.

4. It is tested on sales it has never seen

The honest measure of an AVM is how close its estimates come to actual sale prices that were held out of training. Useful measures include:

  • the share of estimates within 10%, 15% or 25% of the sale price (the "hit rate");
  • the median error, and whether the model tends to over- or under-value (bias);
  • the same measures broken down by property type and by market, since accuracy varies;
  • a fairness test comparing errors across neighborhoods grouped by demographics and income.

A vendor should be able to show these figures for the market and property type you lend in, not only a national average.

5. It tells you how much to trust each estimate

Every estimate should come with a measure of confidence based on how the model performed on similar properties. We explain how to read one in AVM confidence scores.

Where a commercial AVM needs a person

  • Condition and quality: the model does not see a leaking roof or a fresh renovation unless someone tells it.
  • Leases: a long lease to a strong tenant, or a building about to go vacant, changes value in ways the data may not capture.
  • Unique properties: special-purpose buildings and very large assets have few comparables.
  • Thin markets: in small markets with few sales, confidence is lower and the model should say so.

This is why the Interagency Guidelines accept AVMs in evaluations only with supplemental information on the property's condition and a validated tool, and why Valnow uses the model value as one approach and a cross-check inside an evaluation signed by a valuer.

A note on the 2024 AVM rule

In 2024, federal agencies adopted quality control standards for AVMs, in force since October 1, 2025. The rule covers AVMs used for mortgages secured by a consumer's principal dwelling, so it does not apply to commercial property. Its principles are still a good benchmark for any lender: confidence in the estimates, protection against data manipulation, no conflicts of interest, random sample testing, and compliance with nondiscrimination laws.

This article is general information, not legal or regulatory advice. Check your institution's policies and your regulator's current guidance before applying it to a loan.