Your search bar is losing you sales. Modern AI powered search — semantic matching, vector search, dynamic product ranking — turns product discovery into a conversion engine....
AI-Powered Search Transformation
Enhancing eCommerce discoverability with AI-driven search.
- Period
- 2023–2025
- Role
- Product and technical leadership
- Context
- Large-scale retail eCommerce
- Outcome
- Produced a clearer route for improving product discovery through search
This record is anonymised because the underlying work was completed within a confidential professional environment.
Search quality in a large retail environment depends on catalogue data, ranking rules and customer intent signals — but these were being treated as independent technical concerns rather than interconnected parts of the same product problem.
Situation
The retail catalogue contained a broad product range, and search was the primary discovery mechanism across multiple customer journeys. Search quality was measured, but the measurements belonged to separate teams with different priorities: catalogue accuracy, ranking performance and customer behaviour analytics.
Problem
Improving search relevance required connecting those perspectives. Treating search as an infrastructure or algorithmic concern alone would miss the product decisions — what to rank for, how to weigh catalogue quality signals, and whether relevance improvements actually changed customer behaviour.
Constraints
The work had to fit within existing catalogue operations, established retail processes and the confidentiality boundaries of a large commercial environment. No wholesale platform replacement was feasible.
My remit
I framed the product problem, established evaluation criteria linking search quality to customer behaviour, and aligned engineering, product and operational teams around a shared relevance model.
Key decisions
We treated relevance as a product outcome rather than a search-engine tuning parameter. This meant connecting catalogue quality work to ranking improvements and validating changes against customer journey metrics rather than search-engine-internal scores. Deliberately not done: building a custom ranking model before the product and data foundations were in place.
Outcome
The work produced a clearer framework for improving product discovery through search — connecting catalogue quality, ranking rules and customer behaviour within a single evaluation model. The approach reduced the gap between search-engine measurements and observed customer behaviour.
Lessons
Search relevance improvements are most effective when catalogue quality, ranking logic and customer behaviour are treated as a connected system rather than separate technical concerns.
