The problem
The search engine worked. Relevance tuning was not the bottleneck. Discovery still felt uneven: some brands and categories surfaced reliably; others barely appeared even when shoppers were looking for them.
The root cause sat upstream. Across 1.8 million SKUs, metadata quality varied by vendor. Some feeds arrived rich and consistent. Others shipped thin attributes, mismatched categories, or incomplete fashion detail. Search could only rank what the catalog actually described—so metadata inequality became discovery inequality.
The approach
Instead of replacing the search stack, the retailer invested in upstream product intelligence. Perspiq enriched and validated catalog data before it reached index and merchandising workflows.
- Structured enrichment — product records expanded into a consistent, shopper-ready attribute model
- Attribute depth alignment — thinner vendor feeds brought up toward the depth of stronger brands
- Confidence-aware validation — low-confidence fields flagged for review instead of silently polluting the index
- Zero stack disruption — existing search, PIM, and commerce systems stayed in place
The outcome
Search concentration improved because more of the assortment finally had the signals the engine needed. New SKUs ramped into discovery faster, and vendor feed quality stopped setting a hard ceiling on what shoppers could find.
The lesson was simple: when discovery looks broken, the engine is not always the problem. Catalog inequality often is.