The brief
Their best marketing kept crashing their own store.
BluePeak had built a genuine following around limited gear drops: new packs, new shells, gone in a weekend. The marketing worked almost too well. The old store, a monolith on a single overworked server, would slow to a crawl within minutes of a drop email and fall over at the peak. Support spent launch mornings apologising.
There was a quieter problem underneath. Serious buyers choose gear by the numbers (weights, denier, temperature ratings), and all of that lived in manufacturer PDFs. Shoppers left the site to research, and plenty never came back.
The build
Static where it can be, dynamic only where it must be.
The architecture followed one principle: a product page under drop-day load should cost the server nothing. Every page is statically generated and served from a CDN edge, so ten thousand simultaneous visitors read cached files, not a struggling database. Only the things that genuinely change (stock, cart, checkout) hit live services, each isolated so a spike in one can't drown the others.
The PDF problem became a data problem. We turned every spec sheet into structured data, and wired Algolia to understand it: shoppers can now filter jackets by warmth-to-weight, or search "sub-2lb 2-person tent" and get exactly that. The specs stopped being homework and became the shopping experience.
Before the first real drop we rehearsed with load tests at five times the worst historical traffic, then sat on the launch call with the BluePeak team, watching dashboards stay green while the biggest drop of the year sold through.
“First drop in two years where I watched sales instead of server graphs.”
The result
The drop that finally stayed up, and sold double.
The first launch on the new store did twice the sales of any previous drop, with zero downtime and checkout latency flat through the peak. The team's launch-morning ritual changed from watching error logs to watching the sales counter.
The spec-aware search turned out to be more than a defensive fix: filter-driven sessions convert at nearly twice the store average, because the people using them arrive at exactly the gear they meant to find.