Zillow

Accurate, instant home valuations had never been available outside of formal appraisals costing $300-500 and taking days. The opacity of real estate pricing frustrated buyers, sellers, and the market. Zillow had data on 100M+ homes but lacked the models to utilize it effectively.

Zillow's Zestimate uses neural networks and gradient boosted models trained on public records, MLS data, tax assessments, user-submitted data, and proprietary data sources. The latest models incorporate natural language processing on listing descriptions and computer vision on listing photos.

Zestimate achieves a median error rate of 1.9% for on-market homes, making it as accurate as many professional appraisals. Off-market accuracy is ~6.9%, still valuable for ballpark understanding. The system covers 104M+ US homes.