Back to work
AI Commerce

AI Product Recommendation Integrations

Behavior-driven recommendations across five global DTC storefronts.

Client
GUESS Australia, rag & bone Australia, Beach Riot, Bombshell Sportswear, Miss Me
Industry
Fashion & Apparel — DTC
Role
Lead Shopify Developer
Duration
6 months across rollouts
+18%
AOV uplift
+24%
PDP engagement
+11%
Conversion
-32%
Bounce on PDP
The Challenge

What we needed to solve.

  • Static merchandising rules were under-performing on PDP and cart pages.
  • Five storefronts on different theme architectures needed one consistent recommendation layer.
  • Recommendation widgets had to load without harming Core Web Vitals.
The Approach

How it was built.

01

Unified data layer

Standardized product, collection, and customer events across all five themes so the AI engine received clean, comparable signals.

02

Behavioral signals

Tracked view, dwell, add-to-cart, and post-purchase patterns to power 'frequently bought together', 'recently viewed', and personalized PDP rails.

03

Performance-first widgets

Lazy-loaded recommendation sections with skeleton states, deferred JS, and edge-cached responses to protect LCP.

Tech Stack

The toolkit.

Shopify
Shopify PlusLiquidTheme SectionsMetafieldsCustomer Events
AI & Data
Recommendation APIsBehavioral trackingEvent pipelines
Frontend
JavaScriptWeb ComponentsIntersection ObserverTailwind utilities
Features Shipped

What's live.

Personalized PDP recommendation rails
Cart drawer cross-sells
Recently viewed with persistence
Complete-the-look bundles
Editor-friendly section controls per storefront
Results

The outcome.

+18%
Average order value

Driven by complete-the-look and cart cross-sells.

+24%
PDP engagement

Higher click-through into related products and bundles.

+11%
Conversion rate

Site-wide uplift across the five storefronts.

A
Lighthouse performance

Maintained after rollout with deferred loading.

In summary

Five storefronts now ship with a shared, performance-safe recommendation layer that brand teams can tune per collection — translating directly into AOV and conversion gains.