A new framework called TRACE is designed to automatically enrich e-commerce product catalogs with missing or buried attributes. It uses a combination of large language models and multimodal evidence from various sources to propose candidate attribute values. In offline testing, the proposed attribute values were found to be accurate 98.2% of the time, covering 74.7% of all attributes. When deployed in production on an industry-scale catalog, TRACE increased enrichment coverage by 90.4%. Further online experiments showed that surfacing enriched attributes on product detail pages led to a 0.48% increase in checkout conversion rates. --- Why it matters: This matters because e-commerce product catalogs are often incomplete or hard to navigate, making it difficult for customers and downstream systems to find what they're looking for. TRACE's ability to automatically enrich these catalogs with accurate attribute values could improve customer experience and drive business growth. Source: https://arxiv.org/abs/2608.20844