E-commerce recommendation engine and personalization
Head of Data · GOZEM
Delivered a production recommendation engine. In the first month after launch, recommendations were associated with approximately 12% of delivered orders.
- Challenge
- As Gozem expanded its e-commerce offering, Product needed a scalable way to improve product discovery, engagement, and conversion. The team needed to choose a recommendation strategy, identify useful behavioral features, put machine learning pipelines into production, and test where to show recommendations in the customer journey.
- Approach
Led the recommendation engine initiative from concept to deployment. Partnered with Data Science, ML Engineering, and Product to evaluate approaches used by leading technology platforms and choose one suited to Gozem’s customer behavior, available data, and business objectives.
Guided the design of behavioral and transactional features based on customer interactions, purchases, browsing activity, and product relationships. Oversaw the team’s implementation of feature engineering workflows, Vertex AI training pipelines, model evaluation, deployment pipelines, and serverless inference services integrated into the mobile app.
Worked with Product to define tracking events, success metrics, and the experiment design. Product used the results to select the homepage placement for rollout.
The A/B test involved approximately 600 users across three groups: no recommendations, homepage recommendations, and recommendations during search. The team observed the strongest engagement and conversion on the homepage. The 12% figure measures the share of delivered orders associated with recommendations in the first month after deployment.
- Tools
Vertex AI
BigQuery
Python
Cloud Run
Cloud Functions
Google Cloud Platform