Case Study: Fortune 500 Fashion Retailer achieves >15% CTR lift and increased revenue per visitor with Grid Dynamics' real-time STAMP session-based recommendations

Personalized sessionbased recommendations for a fortune 500 retailer

Fortune 500 Retailer, a global fashion goods retailer with a catalog of 100,000+ products, engaged Grid Dynamics to explore session‑based recommendations that work with implicit feedback (views and purchases) and a cloud‑native stack. The customer needed low‑latency (<400 ms) real‑time inference across many ecommerce recommendation zones, while addressing data sparsity and maintaining or improving conversion and revenue per customer.

Grid Dynamics implemented a STAMP (Short‑Term Attention/Memory Priority) session‑based recommender with a Kubeflow training pipeline and a real‑time serving application on Google Cloud Platform, using catalog clustering to combat sparsity. In A/B tests versus the incumbent, Grid Dynamics’ solution lifted mobile CTR by >15% and desktop CTR by >10%, increased revenue per desktop visitor by >3% and per mobile visitor by >1%, and delivered production performance at scale (single pods >100 rps, auto‑scaled peak >1000 rps, serving 100% of traffic in one product zone).