Replacing 3–5 day appraisal processes with instant, 97.2% accurate automated valuations—covering 100 million US properties with price forecasts 50% more accurate than the industry standard.
Valuation Accuracy
Valuation Speed
Properties Covered
Key Outcomes
97.2% accuracy achieved by combining traditional AVM with computer vision condition detection
Calibrated confidence intervals enable risk-appropriate AVM use without false precision
12-month price forecast at 6.4% MAPE is 50% better than industry standard
100M property training base provides statistical robustness across all US market types
Daily transaction ingestion keeps the model current in volatile markets
HouseCanary's Automated Valuation Model (AVM) achieves 97.2% accuracy by combining a gradient boosting model trained on 100 million US property records with a computer vision layer that analyzes property imagery to detect unreported renovations and condition changes. The model ingests 100+ property signals—MLS data, permit records, satellite imagery, neighborhood trend vectors, and local school ratings—and produces valuations with 80% confidence intervals that outperform human appraisers on portfolio-level accuracy metrics.
HouseCanary is a real estate data analytics company serving mortgage lenders, institutional investors, insurance carriers, and real estate operators who need fast, accurate property valuations at scale. Their AVM powers instant mortgage decisioning for thousands of lenders and portfolio valuation for institutional investors managing billions in real estate assets.
A traditional real estate appraisal requires a licensed appraiser to physically visit a property, manually collect data, and produce a report—a process that takes 3–5 days and costs $400–800 per property. For mortgage lenders processing thousands of loans per month, for investors evaluating portfolios, and for insurance carriers managing property risk, this friction is an operational and competitive constraint.
3-5 days
Traditional Appraisal Time
Days required for a traditional physical appraisal—blocking mortgage closings, investor due diligence, and portfolio risk management decisions.
$400-800
Cost Per Appraisal
Cost of a licensed human appraisal, multiplied by thousands of properties in a lender or investor portfolio makes comprehensive valuation prohibitively expensive.
89%
Industry AVM Accuracy
Industry baseline accuracy for automated valuation models before HouseCanary—HouseCanary achieves 97.2%, a 9.2 percentage point improvement.
AGIX Technologies built a property valuation system that combines traditional hedonic pricing models with a computer vision layer for physical condition assessment and a neighborhood trend model that detects market direction changes before they appear in comparable sales data.
100M Property Data Graph
A continuously maintained database of every US residential property, combining MLS listing history, tax records, permit data, and transaction history into a rich property-level feature set.
Computer Vision Condition Assessment
Satellite and street-level imagery analysis detects property condition changes—renovations, additions, deterioration—that aren't captured in public records, improving accuracy for modified properties.
Neighborhood Trend Vectorization
Micro-neighborhood trend signals (school rating changes, new development, demographic shifts, walkability improvements) are quantified as vectors and incorporated into valuations before they appear in comparable sales.
Comparable Selection Engine
ML-powered comparable selection identifies the most relevant recent sales for each property, weighted by similarity across 40+ dimensions rather than simple geographic proximity.
Confidence Interval Generation
Every valuation includes a calibrated 80% confidence interval that communicates uncertainty explicitly—allowing lenders to set appropriate risk thresholds for different use cases.
Price Forecast Model
12-month forward-looking price forecasts for every property, with MAPE of 6.4% vs industry average of 12.8%—enabling lenders and investors to assess risk over the loan or investment horizon.
AVM Accuracy
vs 89% industry baseline—verified on independent holdout samples of properties with known sale prices
Valuation Speed
API response time vs 3–5 days for traditional appraisal—enabling instant mortgage decisioning
Forecast MAPE
12-month price forecast error vs 12.8% industry average—50% improvement in forward accuracy
Risk Detection
Accuracy in flagging properties with elevated value risk vs 78% with rule-based methods
"We've underwritten $4 billion in loans using HouseCanary as our primary valuation source for eligible properties. The accuracy consistently meets or beats our traditional appraisal accuracy benchmarks, at 1% of the time and cost."
Chief Risk Officer
Regional Mortgage Lender
Match the address to the property graph
The API receives a property address or APN (Assessor's Parcel Number) and matches it to the HouseCanary property database. The system retrieves the full property record: physical characteristics (square footage, bedrooms, bathrooms, lot size, year built), renovation and permit history, and ownership and transaction history.
Computer Vision Bridges the Data Gap
Public records often miss renovations completed without permits. Computer vision detection of physical condition changes was the single largest accuracy improvement over records-only models.
Micro-Neighborhood Trend Signals
Detecting school rating improvements, new transit stops, and rezoning activity before they appear in comparable sales creates a lead indicator advantage over models that only look at recent sales.
Calibrated Uncertainty
Providing accurate confidence intervals rather than false precision allows lenders to make risk-appropriate decisions—using the AVM where confidence is high and ordering physical appraisals where it's low.
100 Million Property Training Base
Training on the entire US residential property universe rather than a sample provides statistically robust models even for unusual property types and thin market areas.
Continuous Model Freshness
Daily ingestion of new MLS transactions and monthly model retraining ensures the AVM reflects current market conditions, not 6-month-old data that would skew valuations in volatile markets.
Every AI system has constraints. Here's what to know before building something similar.
Limited Accuracy for Unique Properties
Luxury properties, historic homes, farms, and highly customized properties with few comparable sales have inherently lower AVM accuracy. Physical appraisal remains essential for these segments.
Rural Coverage Is Thinner
The comparable pool is thinner in rural markets with few recent transactions. Confidence intervals are appropriately wider in these areas, flagging them for additional review.
Cannot Assess Interior Condition
Computer vision from satellite and street-level imagery cannot assess interior condition: flooring, kitchen quality, bathroom finishes. Interior condition differences can be material for older properties.
Regulatory Constraints on AVM Use
Federal regulations limit AVM use for higher-balance purchase mortgages. The AVM is most broadly applicable for refinance, home equity, and portfolio valuation use cases.
Explore the services, industry solutions, and intelligence types that power this system.
Common questions about building automated valuation ai systems like the one deployed at HouseCanary.