What is hybrid search?
Hybrid search is a retrieval method that runs a keyword search and a vector search on the same query and merges the two result lists. Keyword search, usually a BM25 ranking over an inverted index, is exact. It finds the chunk containing part number “HX-4412-B” and nothing else. Vector search is approximate and semantic. It finds passages about the same idea in different words but can treat two part numbers that differ by one character as nearly identical. Running both and fusing the rankings, often with reciprocal rank fusion, gives the strengths of each.
Why hybrid search matters
Business documents are full of tokens that carry meaning only as exact strings: SKUs, policy numbers, error codes, statute citations, customer names. A retrieval-augmented generation system that relies on vectors alone will sometimes hand the model a passage about a similar but wrong item, and the model will answer confidently about the wrong thing. Hybrid search is the standard remedy and costs little to add when the document store already supports keyword indexing.
Hybrid search in practice
An electrical parts distributor’s quoting assistant must answer both “is the 2P 30A breaker in stock” and “what do you have for a detached garage subpanel”. The first needs exact matches on the rating codes. The second needs meaning. Hybrid search handles both in one query, and a reranking step orders the merged list before the model sees it.