Aurora Commerce Orchestrator
Lifted blended conversion 32% and cut experimentation cycle from weeks to hours.
The challenge
A retail alliance serving 27 markets needed a storefront architecture that could simultaneously reflect each market's cultural narrative and pricing context while remaining composable for the engineering team. Their existing monolithic Shopify theme was limiting: A/B tests took two weeks to instrument and deploy, personalisation was limited to Shopify's built-in segmentation, and regional teams could not localise the storefront experience without a developer deployment. Every experiment required a full QA cycle because changes to one market affected all markets.
Architecture
The new architecture decouples presentation from commerce logic using Shopify Hydrogen as the headless commerce layer and Next.js 14 with RSC as the presentation layer. Market-specific experiences are managed through a configuration-as-code system — each market has a config object that controls layout variants, component visibility, pricing display rules, and localisation strings, all resolved at the edge via Cloudflare Workers without a round-trip to origin. The personalisation engine runs as a Cloudflare Worker that evaluates user segment assignments (based on browsing behaviour, cart history, and geolocation) and selects the appropriate layout variant from a pre-rendered variant pool. BigQuery powers the analytics layer — every component render event, interaction, and conversion is streamed via a lightweight edge event system, giving the analytics team real-time experiment data without sampling.
How we shipped it
The migration ran in parallel with the existing storefront over 14 weeks. We used a gradual traffic shifting strategy: new architecture served 5% of traffic from week 8, scaling to 100% by week 14. The component library was built market-by-market starting with the two highest-revenue markets. Each market configuration was reviewed and approved by the regional team before go-live. The analytics pipeline launched ahead of the storefront so that the baseline conversion rates were already measured before the new experience went live.
Results
After 30 days at 100% traffic: blended conversion rate increased 32% across all 27 markets. Time to instrument and deploy a new A/B test dropped from 12 days to 4 hours. Regional teams can now localise storefront content without a developer deployment — a capability that had been requested for 18 months. Core Web Vitals improved across all markets: average LCP dropped from 3.8s to 1.2s.
What we would do differently
The configuration-as-code approach was slower to build than a CMS-based approach but paid off significantly in runtime performance and testing reliability. Pre-rendering variant pools at build time — rather than personalising at render time — was the key architectural decision that kept LCP under 1.5s even with complex personalisation logic.
Written by Mudassir Khan
Agentic AI Consultant & AI Systems Architect · CEO of Cube A Cloud · Islamabad, Pakistan
Free tools
Related service
Next.js Development
See scope & engagement →Related case studies
Want to build something like this?
Book a 30-minute strategy call and let us map out what is possible for your situation.