The challenge
The client, an e-commerce retailer, relied on a merchandising team manually maintaining "you might also like" rules for every product. The rules were static, went stale as inventory changed, and had no way to reflect what a given customer actually browsed or bought — every visitor saw the same generic suggestions.
Cross-sell revenue was underperforming, and the merchandising team was spending hours each week updating rule lists by hand instead of working on higher-value merchandising decisions.
The solution
Vizontra built a hybrid recommendation model combining collaborative filtering on purchase and browsing behavior with content-based similarity on product attributes, so new products and new customers still get relevant suggestions from day one instead of defaulting to generic bestsellers.
The model is deployed as an API integrated directly into the storefront's product and cart pages, and retrains on a scheduled cadence as new interaction data accumulates — so recommendations improve automatically rather than requiring manual upkeep.
Results
- Personalized recommendations live across every product and cart page, replacing static manual rules
- Measurable lift in average order value from cross-sell recommendations after rollout
- Model retrains automatically as new customer interaction data comes in
- Cold-start handling for new products and new customers via content-based fallback
- A/B tested against the old manual rules before full rollout to confirm the lift
- Merchandising team freed from maintaining "related products" lists by hand