What is faceted search?

Faceted search is the way of navigating results that lets someone narrow them down by combining several independent filters, each built on an attribute of the items in the catalog: brand, price, availability, material, size. Each facet is a dimension of its own, so selections add up instead of excluding one another, and next to every value there is usually the number of results that choice would leave, so the searcher sees the outcome before clicking. The technique predates the web: it descends from the faceted classification proposed by the Indian librarian S. R. Ranganathan in his colon classification, first published in 1933, and reached online retail through interface research between the 1990s and the early 2000s. It should be told apart from hierarchical categories, which place each item in a single branch of a tree: facets describe the same item along several dimensions at once and impose no order of choice.

See also: Hybrid search and reranking · The ranking stage of RAG: it fuses lexical and vector search, then a reranker reorders the top candidates.

A data problem disguised as an interface problem

A facet works only if the attribute behind it exists on every item, is normalized, and has manageable cardinality. If the "color" field holds "red", "Red" and "RED", the filter shows three distinct values and each one hides the products of the other two. If the attribute is filled in on sixty percent of the catalog, every selection throws away a good forty percent of the items, which were not incompatible: they simply had no declared value. And if a facet carries hundreds of values, the list becomes unreadable and has to be bucketed into ranges. The defect shows up in the interface, while the cause almost always sits in the PIM or the ERP, where attributes are keyed in by hand by different suppliers under different rules. That is why faceted search is first of all an exercise in data quality: completeness and consistency of values, not the choice of a UI component. It stays distinct from retrieval, which is the job of hybrid search with reranking: facets do not decide which items enter the result list, they decide how the searcher walks through it.

An enterprise example

An industrial spare parts distributor publishes eighty thousand codes: bearings, seals, fittings. The useful facets are not commercial but technical: inner diameter, seal material, working pressure, reference standard, lead time. These are exactly the attributes that supplier catalogs deliver as PDFs, with mixed units of measure and unaligned abbreviations. Until those values are extracted and normalized into a shared model, the maintenance engineer looking for a 42 millimeter EPDM seal gets zero results and concludes the part does not exist, while it sits in the warehouse. The remedy for an empty combination is a technical decision to take up front: disable the values that would lead to zero, or show the empty page while explaining which filter to drop and offering the closest items.

Why it matters for decision makers

The cost of faceted search is not the engine, which ships it as standard today, but the catalog enrichment work, and that work is recurring rather than a project that ends. Before funding a new interface it pays to measure the coverage of the attributes proposed as facets: below a certain threshold the filter hides products instead of surfacing them, and search gets worse than it was. The same investment then pays a second time: when a conversational agent queries the catalog, facets are the structured parameters it can filter on. A catalog with clean attributes is also the catalog an agent can actually use.