The Challenge: Generic Recommendations
The client’s existing recommendation widget served largely generic, popularity-based products, resulting in low engagement and minimal impact on revenue. They needed a smarter system that could understand user intent and context in real time.
DevoxLabs' AI Recommendation Engine
We delivered a personalized recommendation engine that combined behavioral data with AI-powered context modeling:
Unified User Profiles: Aggregated browsing, purchase, and in-session signals into a single behavioral profile per user.
Gemini-Powered Context: Used the Gemini API to interpret session context (queries, category navigation, dwell time) and bias recommendations toward highly relevant items.
Experimentation Framework: Integrated an A/B testing layer to continuously evaluate and iterate on recommendation strategies.
Measurable Outcomes
Business impact of personalized product recommendations.
| Metric | Before DevoxLabs | After DevoxLabs | Improvement |
|---|---|---|---|
| Average Order Value (AOV) | Baseline | +25% vs. control | 25% Increase |
| Session Engagement | Baseline click-through | +15% click-through | Higher Engagement |
| Recommendation Relevance | Generic popularity-based results | Context-aware personalized ranking | Improved Relevance |