Get in touch
team@kovazu.comHigh-Performance Storefront & AI Personalisation Engine
Direct-to-Consumer Lifestyle Brand
Key Outcome
41% conversion rate lift — 99 Lighthouse score — sub-1.8s LCP globally
The Challenge
The brand's Shopify theme was built on a starter template that had been modified by three different freelancers over four years. Page load times were averaging 4.2 seconds on mobile. The add-to-cart conversion rate had stagnated at 2.1% despite significant paid media spend. Product discovery was purely category-based — no cross-selling, no personalisation, no intelligent bundling. The brand was leaving significant revenue on the table.
Our Approach
We started with a full performance and UX audit before writing a line of code. The decision to go headless was driven by the requirement to own the full frontend performance stack without being constrained by Shopify's theme limitations. The personalisation engine was scoped to three specific high-ROI touchpoints: the product detail page, the cart, and the post-purchase page — rather than attempting site-wide personalisation from day one.
Tech Stack
System Architecture
How the system flows
Build Pipeline
How We Built It
Headless Architecture Design
Decoupled the frontend entirely from Shopify's theme layer, using the Shopify Storefront API for product data, cart management, and checkout. All product data is cached at the edge via Vercel's KV store with a 5-minute TTL, reducing Shopify API calls by 94% and eliminating the most common cause of slow page loads.
Performance-First Frontend Build
Built with Next.js App Router, React Server Components, and aggressive image optimisation via next/image with AVIF format serving. Critical CSS is inlined; non-critical scripts are deferred. Product images are pre-transformed and served from Vercel's CDN at the exact dimensions required by each device viewport.
Product Embedding & Similarity Index
Generated semantic embeddings for all 340 SKUs using product titles, descriptions, materials, and tags. Embeddings are stored in a Redis vector index. Product similarity is computed at build time for static recommendations and at request time for dynamic personalisation, with results cached per product at the edge.
Real-Time Personalisation Engine
An anonymous session fingerprint tracks in-session browsing signals (products viewed, time spent, scroll depth, category affinity). On the product detail page, the recommendation widget fires a lightweight Edge Function that combines session signals with product similarity scores to rank the recommendation set — no external API call, sub-20ms response time.
AI-Powered Bundle Suggestions
A GPT-5.5 call at cart-open time generates a contextual bundle suggestion based on the cart contents and the brand's product catalogue. The prompt is given the brand's tone of voice guidelines and instructed to suggest one complementary product with a one-sentence rationale. Conversion on AI-suggested bundles was 3.2× higher than static cross-sell rules.
A/B Testing & Conversion Instrumentation
Implemented Vercel's Edge middleware-based A/B testing framework, enabling flag-controlled experiments on recommendation placement, bundle messaging, and cart UI variants. All events are instrumented with server-side analytics (no client-side trackers that slow the page) feeding into a custom Metabase dashboard for the brand's growth team.
Results
What We Delivered
41%
Conversion Rate Lift
Add-to-cart conversion rate increased from 2.1% to 2.96% within 60 days of launch, measured against a 90-day pre-launch baseline.
99
Lighthouse Performance Score
Achieved perfect Lighthouse performance score on mobile and desktop, with sub-1.8s LCP measured from servers across 6 global regions.
3.2×
AI Bundle Conversion vs Static Rules
AI-generated bundle suggestions outperformed the previous static cross-sell rules by 3.2× on a controlled A/B test across 40,000 cart sessions.
LET'S WORK
TOGETHER
Work with us if average isn't your thing. Drop it, we'll build it!
SAY HELLO