Vector (Hybrid) Search Live A/B Test Results in E-commerce (DIY) Sector
Istvan Simon • Location: Theater 5 • Back to Haystack 2024
“We have statistically significant A/B test results for our Vector (Hybrid) Search for 2 of our large DIY retail clients E-commerce website.
This was a 50-50 user split A/B test. The 2 variants were:
- Prefixbox’s existing, manually-optimized Search Engine
- Prefixbox’s Vector (Hybrid) Search
Results from 1 of the tests are the following: 21% revenue increase for search users 12% increase in transactions for search users 17% increase revenue for overall site visitors 11% increase in transactions overall site visitors 5.7% increase in SERP clicks
We are using ElasticSearch both for keyword and vector search. We use OpenAI’s embeddings.
In this talk, we are going to explain:
- how we create the embeddings for the products and the search query
- how we blend the keyword and vector search results
- and the results of the live A/B tests
- show live search examples”
Istvan Simon
PrefixboxIstvan Simon is the CEO and founder of Prefixbox, an AI-powered E-commerce site search solution. While working on Bing at Microsoft London, he was inspired to build a service that would help retailers optimize their online stores, and thus Prefixbox was born 10 years ago. With more than 20 years of technical experience, he brings the ground-breaking ideas that make Prefixbox so powerful.