Case Study: Houzz achieves 50% higher search accuracy with Provectus semantic query understanding

Houzz boosts search accuracy 50% with Provectus

Houzz, an online platform connecting homeowners with home improvement professionals, faced a challenge with its search engine accurately processing long-tail queries. Less than 40% of these specific user searches were routed to the correct product page, which risked lost sales and user dissatisfaction. Provectus implemented a solution using Amazon Titan embeddings and a Flair NER model trained on synthetic data generated by Claude 3 Sonnet.

The solution built by Provectus significantly improved search accuracy. Category and attribute identification accuracy rose from 52.94% to 78%, a nearly 50% improvement, while recall increased to 85%. This allowed Houzz's search engine to correctly understand a much wider range of customer queries without increasing latency, leading to more qualified traffic and longer user engagement on the platform.

Houzz