Home/Property Data/Home Valuations (AVM)

Property data, United States

Home Valuations (AVM)

Machine-learning property value estimates with confidence ranges, plus the comparable sales the estimate rests on.

AVMValuationCompsAPI

An automated valuation for every US residential property, retrained weekly on 200 million historical transactions and current listings. Each estimate ships with a confidence score, a range and the comparable sales used, so an analyst can defend the number.

Available as a bulk file for portfolio marks or as a real-time API for lending, insurance and consumer products.

Sample rows

Illustrative and synthetic. Real samples for your geography are a reply away.

property_idaddressestimatelowhighconfidenceas_ofcomps_usedlast_sale_pricelast_sale_date
P0148••••••• W Oakwood Ave, Chicago IL$9••,000$8••,000$9••,0000.8809/01/20266$8••,00006/14/2018
P5522•••2•• Vine St, Fresno CA$4••,000$3••,000$4••,0000.8109/01/20268$3••,00005/14/2021

Schema

FieldTypeNotes
property_ididentifiernon-personal
addresstextpersonal data
estimatecurrencynon-personal
lowcurrencynon-personal
highcurrencynon-personal
confidencenumbernon-personal
as_ofdatenon-personal
comps_usedtextnon-personal
last_sale_pricedatenon-personal
last_sale_datedatenon-personal

Abridged. The full data dictionary ships with every sample and lists every field, its fill rate and its source.

Used for

  • Portfolio and collateral marks
  • Equity-based marketing audiences
  • Consumer home-value products
  • Model features for underwriting

How it is built

Gradient-boosted models on transaction, assessor, listing and permit features, retrained weekly with holdout validation by metro. Comps are selected by distance, similarity and recency.

Questions we get

How does it compare to the big AVMs?

Competitive on median error in most metros and better in markets where we have permit and listing signal the others lack. We are happy to run a blind test on your portfolio.

How fast is the API?

Built for single-record lookups inside a user-facing flow. We will benchmark it against your expected volume before you commit.

Related