Headless commerce without a data migration.
A D2C brand capped by a plugin-heavy legacy storefront: slow on mobile, no personalisation, and a checkout with too many places to fail. We re-platformed onto headless while keeping the existing system as order-of-record.
What was breaking
Revenue had been built on a legacy WooCommerce storefront carrying a stack of plugins that fought each other. Mobile — the primary channel — was slow. Checkout lost carts at every extra step. Pricing experiments needed a deploy, so they effectively did not happen. And peak-season traffic was a known risk with no headroom. The brief was an architecture that could take a large multiple of current traffic without a rewrite.
The technical bets we made
- 1
Replatformed onto a headless architecture: Next.js + GraphQL frontend, with the existing WooCommerce kept as the order-of-record (no data migration risk).
- 2
Moved product catalog onto Elasticsearch with ML-driven 'people also bought' and personalised recommendations based on browse + purchase history.
- 3
Rebuilt checkout as a single-page flow with a persistent cart, address autocomplete, and UPI, cards and EMI on one screen instead of behind separate steps.
- 4
Set up GCP infrastructure with Cloudflare edge caching for product pages, so catalogue traffic is served from the edge and origin load tracks orders rather than browsing.
- 5
Built a feature-flag service so the marketing team could ship dynamic pricing, hero variants, and seasonal banners without engineering involvement.
What powers this system
From kickoff to production
Performance audit, behavioural funnel analysis, competitor benchmark on the same product category.
Information architecture, design system, headless data model. Replicated their top 5 pages in clickable Figma prototypes.
Frontend rebuild + GraphQL layer + recommendation engine. Daily deploys to staging, weekly demos.
Phased traffic shift via Cloudflare workers. 10% → 50% → 100% across two weeks. Rollback path tested before each ramp.
What the architecture changed
The storefront stopped being the constraint on marketing. Pricing, hero variants and seasonal banners moved behind feature flags, so campaigns ship without an engineering queue. Catalogue traffic serves from the edge, which means peak season is a caching question rather than an origin-capacity one. And because the legacy platform stayed the order-of-record, none of this required migrating order history.
Need something similar?
Book a free 30-min architecture audit. We'll sketch the system you need.
Book a 30-min Architecture Audit