#1Where AVMs fit in modern real-estate workflows
AVMs touch three workflows that matter: consumer search (instant valuations), lender underwriting (decision support and audit), and broker advisory (pricing strategy). Each demands a different latency, accuracy, and explainability profile.
#2The data ingredients of a serious AVM
- MLS records and tax assessor data, reconciled and version-controlled.
- Hedonic features (school zone, walkability, view) computed from third-party sources.
- Imagery features extracted from aerial and street-view sources.
- Market microstructure features (days on market, list-to-sale ratio, inventory).
#3Explainability is now a feature, not a constraint
Consumers and underwriters both want to know why the model returned a number. We invest in SHAP-style explainers and policy-aware rationales, surfaced in the UI alongside the price. A model that cannot explain itself is a model that will eventually be replaced.
#4Regulation is catching up — get ahead
AVM governance is moving from voluntary to mandatory in lender markets. We build governance scaffolding into the model lifecycle from day one: bias testing, geographic accuracy reviews, and documented overrides.

