Headless storefront rebuild that lifted conversion rate.

Northstar’s monolith handled Black Friday — technically — but PDP latency in the EU eroded conversion on high-AOV categories. Merchandising wanted editorial velocity; engineering feared every CMS field became a SQL join. Checkout abandonment sat stubbornly above benchmark despite discounts.

The CEO framed the trigger as “make premium feel premium online” before a partner channel launch. Twelve weeks to first composable release with no catalogue freeze longer than a weekend.

We kept Shopify as commerce truth and refused to fork pricing logic into a bespoke cart. Constraint: one CMS for storytelling; no second content swamp.

// TODO: Verify with client legal before publishing — commercial framing and benchmarks.

Framing. A storefront is a latency budget and a trust surface. We prioritised TTFB and checkout friction before visual novelty.

Architecture. Next.js storefront calling Shopify APIs, Sanity for editorial, Cloudflare for caching and WAF rules tuned with merchandising promos in mind. Inventory reads stayed near real-time; cache keys respected stock thresholds.

Phasing. Weeks 1–3: performance baseline + dead code removal. Weeks 4–8: PDP and PLP rebuild with component library. Weeks 9–12: checkout experiments and edge rollout.

Judgment. We killed a 3D product viewer that tanked LCP; marketing mourned briefly, finance did not.

// TODO: Verify with client legal before publishing — architecture and partner integrations.

Conversion rate on the rebuilt path rose 24% with SKU mix held constant in reporting. Checkout drop-off from cart to paid fell 39% after address validation and shipping logic were simplified. Mobile Lighthouse p75 on PDP templates reached 94; EU p95 TTFB improved 22% once edge caching rolled out.

Northstar’s team owns the design system now; we left a playbook for promo-week cache invalidation. Engagement continues with seasonal performance reviews rather than a standing dev army.

// TODO: Verify with client legal before publishing — experiment design and uplift claims.

We waited too long to involve store ops in shipping logic — their reality checks would have saved a week of rework. We were surprised how aggressively bot traffic skewed checkout tests; bot rules moved earlier in the plan.

Next time we would publish a single executive metric sheet (conversion, AOV, returns) before touching components — alignment tool more than a dashboard.

// TODO: Verify with client legal before publishing — organisational reflections.