Furniture.com Transforms Online Search with Databricks

Overview

Furniture.com unifies over 60 retail partners and 1.5 million SKUs on the Databricks Data + AI Platform, using Delta Lake, MLflow and Unity Catalog to run its ML lifecycle. Its Find It AI product-discovery tool uses generative AI to create a synthetic image representing shopper intent, then matches it against the product catalog for image-based search. A Collections model uses LLMs to automatically group related products, finding more than 16,000 collections across 50+ partners with no human intervention. Users who interact with Find It AI show a click-through rate 8x higher than baseline and a return rate 3.2x higher than baseline.

The challenge

The fragmented nature of the furniture industry posed significant challenges for Furniture.com in unifying over 60 retail partners and 1.5 million SKUs, including inconsistent image sizes, missing attributes and unstructured data with no industry standards. Manual processes for cleaning and standardizing product information, such as reconciling different color names across partners, quickly became impractical and could not be reused across the catalog.

The solution

Built on the Databricks Data + AI Platform, Furniture.com runs its entire ML lifecycle on a single platform, using Delta Lake for data processing at scale, MLflow for model experimentation and deployment, and Unity Catalog for granular access control. Its Find It AI product-discovery tool uses generative AI to generate a synthetic image representing shopper intent, then matches that image against the product catalog to deliver personalized, relevant search results. The team also built computer vision models to classify and crop product images into a clean, standardized format, and a Collections model that uses large language models to automatically group related products, such as bedroom or dining sets, even when partners don't explicitly label them. Databricks Apps is also used to share interactive QA tools across the company.

Reported business value

Users who see the Find It AI experience have a click-through rate of 10%, and those who actively interact with it show a CTR of 23.5% — more than an 8x increase over the 3% baseline. The baseline return rate of 7.17% climbs to 23.2% for those who interact with Find It AI, a 3.2x improvement. The Collections model found more than 16,000 collections across 50+ partners with no human intervention, running automatically as part of the ETL process. The merchandising team now operates with just four or five people supporting a catalog of over 1.5 million SKUs, which the company estimates would otherwise require 10+ people.