Type "warm jacket for rainy commute" into most storefront search bars and you get products containing the word jacket, ranked by whatever was indexed first. The shopper told you the season, the weather and the use case, and the engine kept only one token.
The gap between matching and understanding
Keyword search matches strings. Intent-aware search maps the query and the catalog into the same semantic space, so "something for my first 10k" can surface running shoes, anti-chafe gear and hydration belts even though none of them contain the words in the query.
The measurable difference shows up in three numbers: search-led conversion, zero-result rate and average order value. In a recent build we saw search-led conversion rise 18% with no change to page latency, because the heavy lifting happens at index time, not at query time.
Where to start
You do not need to replatform. A semantic layer can sit beside an existing search engine, handle the queries it classifies as intentful, and fall back to keywords for exact SKU lookups. Start with your zero-result queries; they are a free list of demand your current search cannot see.