Skip to content

Retail & E-commerce · UrbanKart

AI personalization lifts e-commerce conversion 31%

A fashion marketplace deployed real-time recommendations and search relevance tuning across 2.3M monthly shoppers.

31%

conversion rate lift

23%

average order value increase

44%

search exit rate reduction

+$11M

annualized incremental revenue

The Problem

UrbanKart's generic, batch-updated recommendations converted poorly. Search returned irrelevant results for long-tail queries, and 71% of sessions ended without a product page view.

The Solution

We built a real-time personalization engine: session-based recommendations, vector-powered semantic search, and dynamic ranking that blends popularity, availability and margin. An experimentation framework A/B tests every ranking change against revenue per session.

Architecture

  • Event streaming of clickstream into a real-time feature store
  • Two-tower recommendation model with session context
  • Semantic search with embeddings + traditional relevance signals
  • Experimentation platform with guardrail metrics
  • Edge-cached inference for sub-50ms recommendations

ROI

Incremental revenue of $11M annually against a program cost under $600k — the experimentation framework now pays for a permanent growth team.

Engagement

Client
UrbanKart
Industry
Retail & E-commerce
Service
AI & Software Development
Timeline
14 weeks to first A/B win; personalization platform complete in 6 months

Technologies

PythonPyTorchElasticsearchRedisKafkaNext.jsAWS

Every ranking change is now a measured experiment. RippleCode gave us a growth engine, not just a recommendations widget.

Sarah Lindqvist

Chief Digital Officer, UrbanKart

Facing a similar challenge?

We'll walk you through how this architecture would apply to your environment — free of charge.