The Store That Learns: How AI Personalization Lifted Revenue Per Visitor by 38% Without a Single New Visitor
eBizTrait built a real-time AI recommendation and merchandising engine that turned a one-size-fits-all storefront into a different experience for every shopper.
Results At A Glance
-
+38%
Revenue Per Visitor
AI-personalized sessions
-
+21%
Add-to-Cart Rate
personalized homepage
-
3.1×
Email CTR
AI-curated product picks
-
61%
Sessions Personalized
within 4 weeks
The Challenge
CircuitHub sold everything from smart-home gadgets to gaming peripherals to office electronics – a product catalog so broad that no single homepage layout could be relevant to more than a fraction of visitors. A gamer landing on a homepage built around office accessories bounced. So did the parent looking for smart-home basics who was shown gaming headsets.
Their existing approach was manual: the merchandising team hand-curated homepage collections weekly, guessing at what different segments might want based on gut feel and last quarter’s bestsellers. It was slow, it didn’t scale, and it treated every visitor from every channel identically – a TikTok ad click and a returning VIP customer saw the exact same homepage.
They didn’t need more traffic. They needed the traffic they already had to see a store that felt like it understood them.
Our Approach
We built a real-time personalization layer that sat between Shopify’s storefront and a machine learning recommendation model, using first-party behavioural signals (browsing history, purchase history, cart contents, referral source, and on-site engagement) to reshape the experience per visitor – not per segment.
Rather than relying on a black-box third-party personalization app with limited control, we built the recommendation engine on top of a customer data pipeline we owned, feeding a hybrid collaborative-filtering and content-based model that updates in near real time as a session progresses.
Every personalized surface – homepage collections, product page “you might also like,” cart upsells, and post-purchase email – pulled from the same underlying model, so recommendations stayed consistent across the entire customer journey instead of feeling disjointed.
What We Built
- First-party customer data pipeline aggregating browsing, purchase, and engagement events into a unified customer profile store
- Hybrid recommendation engine (collaborative filtering + content-based) served via a low-latency API integrated into Liquid and app-block sections
- Dynamic homepage that reassembles its collection order and hero content per-visitor based on inferred intent and referral source
- Real-time “complete the look” and “frequently bought together” modules on product pages, replacing a static, manually-curated app
- Personalized cart drawer upsells weighted by margin and predicted purchase likelihood, not just “related products”
- AI-curated product blocks embedded into Klaviyo email flows via API, replacing static best-seller blocks
- A/B testing framework to continuously validate personalized experiences against control groups
- Privacy-first implementation: first-party data only, transparent consent management, no third-party data resale
Technologies Used
Results – Before & After
| Metric | Before | After |
|---|---|---|
| Revenue Per Visitor | $2.14 | $2.95 |
| Homepage Add-to-Cart Rate | 3.2% | 3.9% |
| Email Click-Through Rate | 1.8% | 5.6% |
| Returning Customer AOV | $86 | $121 |
| Merchandising Hours (weekly) | 12 hours manual | 1.5 hours (model oversight) |
| Sessions Receiving Personalized Content | 0% | 61% |
“We used to guess what people wanted and hope. Now the store figures it out in real time, and honestly it’s better at it than we ever were manually. The recommendation quality alone changed how customers talk about us – people comment that the site ‘gets’ what they’re into.”
– Director of Ecommerce, CircuitHub
Key Takeaway
AI personalization delivers its biggest wins not by finding new customers, but by making every existing visit more relevant. For catalogs too broad for one-size-fits-all merchandising, a real-time recommendation layer turns traffic you’re already paying for into meaningfully more revenue – without touching the media budget.
Ready to achieve similar results?
