Datamilk AI
Our Challenge
As customer expectations for personalization grew, our product experience needed to evolve beyond static recommendations. Our goal: deliver meaningful, dynamic product suggestions and content throughout the user journey — engaging customers in real time, aligning with their interests, and ultimately driving conversion.
The challenge wasn’t only technical. We needed to integrate live AI personalization into an established e-commerce ecosystem while maintaining site stability, speed, and user trust.
The Approach
We integrated Datamilk AI by embedding a single line of JavaScript across Zenni’s site, allowing real-time learning from live customer behavior.
Over two weeks, the AI trained on actual user interactions, continuously optimizing product recommendations and content modules within each session.
Working cross-functionally with merchandising, QA, and product partners, we adopted a launch-and-learn model — focusing on iterative progress, adaptability, and data-driven optimization.
To empower business teams, we partnered with Datamilk to create a Merchandising Dashboard, enabling non-technical stakeholders to surface trending or new products dynamically.
We also introduced a Quick Search component to further test and refine how AI recommendations could enhance discovery and engagement.
Despite initial technical dependencies that delayed launch, the team stayed focused on maximizing impact and learning through real-world data rather than theoretical optimization.
We launched initially at 30% traffic to test performance and minimize risk.
Phase 1: After 34 days, “Good” components generated $34,602 in incremental revenue. Confident in the AI’s ability to deactivate underperforming modules, we scaled traffic to 50%.
Optimizations
We enlisted Datamilk to create a dashboard empowering our merch team to curate new or trending content, expanding user exposure to a broader range of products. Additionally, we integrated a Quick Search component for testing and review purposes.
We increased the traffic to 50%
Phase 2: At 50% traffic, after 30 days, “Good” components produced $73,704 in revenue. Continued validation pushed that number to $179K, leading to an 80% rollout.
Final Results: Post-launch analysis projected $2.08M in annual incremental revenue, far exceeding the original estimate of $500K.
Final Results
Key Learnings
The Datamilk integration showed that AI-driven personalization isn’t just a technology shift — it’s an organizational mindset shift.
We learned that one-on-one stakeholder alignment early in the process clarified objectives and reduced friction during rollout. Balancing bold experimentation with risk-aware governance created a healthy rhythm of progress and accountability. And most importantly, growth happens when curiosity, data, and design converge — enabling teams to deliver personalized, human-centered experiences at scale.