Dumb Search: 30% of Sales Lost in Retail

TL;DR Classic keyword search (Elastic/Solr) understands text, not intent. In 5 audited e-commerces, 30% of searches end with no click — not for lack of product, but due to semantic illiteracy. Vertex AI Search resolves this via vector search + multimodal. Average conversion gain: +16%.

Friday night. Your customer has a wedding on Saturday morning. She opens your site and types desperately: "long dress for daytime countryside wedding".

Your search engine, which costs R$ 15,000/month, responds:

Search Result No products found for "countryside".
Did you mean "country side"?

The customer closes the tab and buys on Amazon. You lost R$ 800 not for lack of product, but because of your software's semantic illiteracy.

Keyword vs Vector: the million-dollar difference

Traditional search looks for words. Vertex AI Search looks for meanings. See the architectural difference:

Criterion	Traditional search (Elastic/Solr)	Vertex AI Search (vector)
Query: "Black running shoes"	Exact match: "shoes" AND "running"	Understands: "dark performance athletic footwear"
Typo	Fails ("no results")	Automatic contextual correction
Multimodality	Text only	Text + image (photo search)
Vague intent ("gift for sister turning 30")	Returns noise	Returns relevant curation
Average conversion (5 e-commerces audited)	Baseline	+16%

The code: how the machine "thinks"

When the customer types "light dress", Vertex doesn't search for the string "light". It converts her intent into a mathematical vector (embeddings) and searches for neighboring products in that vector space.

// Embeddings Response (Simplified) { "query": "dress for daytime wedding", "intent_vector": [0.82, -0.45, 0.12, ...], "nearest_neighbors": [ { "id": "SKU-992", "name": "Floral Midi Dress", "score": 0.98 // High semantic relevance }, { "id": "SKU-551", "name": "Wedge Sandal", "score": 0.85 // Visual cross-selling } ] }

⚠️ Vector doesn't replace a well-built catalog Embeddings pull from the text you have. If the SKU-992 description says only "floral M dress", that's what the agent works with. Catalog enrichment (attributes, usage occasion, visual context) is a prerequisite for 100% of the gain. Multimodal helps when the image compensates for poor text, but doesn't fix an empty catalog.

The customer finally finds what they couldn't even describe properly. That's the difference between 30% of searches with no click and 96% with relevant results in the top-3.

Search audit

Is your e-commerce losing 30% of sales in search?

Autenticare audit analyzes 30 days of search logs, identifies queries with no clicks, estimates lost sales and projects gains with Vertex AI Search. Leaves with business case and 60-day plan.

Request audit → Intelligent catalog

Keyword vs Vector: the million-dollar difference

The code: how the machine "thinks"

Is your e-commerce losing 30% of sales in search?

Also read