Semantic search across a 240,000-SKU catalogue
Replacing keyword search with hybrid semantic retrieval lifted search conversion 23% and cut zero-result searches by four fifths.
Where they started
A marketplace with 240,000 SKUs from thousands of sellers had wildly inconsistent product data. 14% of searches returned nothing, and shoppers describing what they wanted rather than naming it rarely found it.
What we did
- 1
Enriched sparse listings by generating structured attributes from product images and seller descriptions.
- 2
Built hybrid retrieval so exact SKU and brand lookups stayed precise while descriptive queries matched semantically.
- 3
Added query understanding to separate attributes (colour, size, price band) from the descriptive remainder.
- 4
Ran an eight-week A/B test against the incumbent search across 2.4 million sessions.
The outcome
Search conversion rose 23% and zero-result searches fell from 14% to 2.6%. Catalogue enrichment improved listing quality across the marketplace as a side effect.
Other case studies
Cutting support volume by half with grounded retrieval
A national operator's support assistant answered 19% of queries correctly. Rebuilding the retrieval layer took it to 54% without changing the model.
HealthcareClinical documentation support that clinicians actually trust
Structured note drafting deployed entirely inside the hospital VPC, cutting documentation time 41% with every output clinician-reviewed.
FintechAutomating KYC review without losing the audit trail
Document extraction and verification cut manual KYC review 62% while making every decision reproducible for regulators.
Let's talk about what you're building
Tell us the problem. We'll tell you honestly whether AI is the right tool, and what it would take.