The Automated Valuation Model (AVM) in WDSuite provides instant, data-driven property valuations for multifamily assets. Built using advanced machine learning techniques and trained on tens of thousands of commercial real estate transactions, our AVM offers a fast and scalable way to estimate market value. However, as with any model-based estimate, results can vary — and we want to help you understand why.
What is the AVM — and What It’s Not
The AVM is an estimate of market value, powered by a machine learning model trained with public transaction data and enriched with contextual market factors.
It is not an appraisal, broker opinion of value (BOV), or a replacement for professional judgment. Use it as a starting point to estimate your property's value — not the final word.
For a professional appraisal, get in touch with one of our appraisers here.
Who Is It For?
The WDSuite AVM is built for stabilized multifamily commercial real estate assets. Specialty property types — including student housing, manufactured housing, and assisted living facilities — have been excluded from model training and are not supported by the AVM.
Accuracy: How Close Is the Estimate?
We benchmark our model’s performance against real transactions using median absolute percentage error (MdAPE) — a standard in AVM accuracy. A MdAPE of 6% means that half of the AVM's predictions are within 6% of the actual transaction price.
Here’s what you can expect for the various model inputs:
5.95% MdAPE when net operating income (NOI) is provided
8.02% MdAPE when asking rent is provided
11.05% MdAPE when only average unit size is provided
For context, Zillow’s Zestimate for single-family homes has a median error rate of 7.06% for off-market properties (Zillow Zestimate Accuracy).
WDSuite's multifamily AVM performs competitively in comparison, especially when detailed inputs like NOI are included. Modeling multifamily property value presents unique challenges not found in single-family residential valuations due to the lower frequency of transactional data — making the AVM’s performance especially noteworthy in this context.
How It Works: Under the Hood
The WDSuite AVM uses a supervised machine learning model based on gradient boosting decision trees (GBDT) — a technique known for high performance in predictive modeling.
What it uses:
Public transaction data (e.g., tax records, CMBS filings, REIT disclosures)
Operational data (e.g., NOI, asking rent)
Contextual & hyperlocal data: crime, retail density, schools, green space, cultural amenity proximity, etc.
Market trends: demographics, employment statistics, interest rates, and more
These inputs feed into a model trained on tens of thousands of property transactions to identify patterns and relationships that impact value.
Why the Estimate Might Seem Off
There are several reasons an AVM estimate might deviate from your expectations:
Missing property inputs: The accuracy varies depending on which input is provided. The most accurate results occur when NOI is provided, followed by asking rent, and then average unit size.
Market volatility: The AVM reflects historical data trends. Rapidly changing conditions may not be fully priced in.
Property-specific factors: Unique characteristics (e.g., recent renovations, deferred maintenance) may not be captured in available data.
Geographic nuances: While the model leverages hyperlocal data, niche market quirks can still lead to discrepancies.