Every flash sale crashed the monolith storefront. Marketing spent to acquire traffic the platform couldn't hold; engineering firefought instead of shipping.
Headless commerce architecture with edge-cached storefront, queue-protected checkout, and autoscaling infrastructure — plus an admin for merchandising that marketing runs without engineering tickets.
Performance is merchandising
The fastest-loading product page became the best-converting one. That is the whole argument for the replatform: not an infrastructure project that happened to help revenue, but a revenue project that happened to involve infrastructure. The 28% conversion lift came from speed alone — no change to price, range or creative accompanied it.
Edge-cache what does not change
Most of a storefront is identical for every visitor: layout, imagery, category pages, copy. Serving that from the edge and reserving origin capacity for the parts that genuinely differ — cart, stock, price — is what let the same architecture absorb ten times the previous peak while holding 99.98% uptime through festival trading.
The queue is what keeps checkout honest
A flash sale does not fail gracefully on its own. It fails at checkout, at the worst possible moment, on traffic that has already been paid for. Putting a queue in front of checkout means the platform degrades into a wait rather than into an error, which is the difference between a slow sale and a cancelled one.
Merchandising without a ticket
The admin exists so marketing can change a banner, a collection or a landing page without an engineering release. Replatforms are usually judged on load tests. This one is also fairly judged on how many tickets stopped being filed.