The challenge
The previous peak season had gone badly: checkout slowed to a crawl at exactly the moment traffic mattered most, and the team had no idea why, because nobody had ever measured where the ceiling was. Subscription stores are harder to scale than standard ones for a specific reason — renewal processing runs on a schedule that has no awareness of your traffic, so a renewal batch can fire in the middle of your busiest hour and compete for the same database connections as live shoppers. On top of that, checkout can't be page-cached, so every checkout request hits PHP and the database directly.
Measured on the existing setup before any changes: checkout degraded past acceptable response times at roughly 400 concurrent users, sessions stored in the database with wp_options contention under load, renewal batches running unthrottled via WP-Cron, no autoscaling, and no load test had ever been run.
The approach
Peak-season failures almost never happen at the front end — they happen at checkout, where caching can't help you and every request hits PHP and the database. So the first job was to break the store on purpose, in a test, to find the real number rather than guessing at capacity.
The results
The store sustained an order of magnitude more concurrent load than before, with checkout response staying under a second at p95 through the peak window. The renewal-scheduling change mattered more than the raw capacity increase: it removed an entire category of failure where a scheduled batch could degrade checkout at the worst possible moment. The dress rehearsal two weeks out caught a caching misconfiguration on the account page that would otherwise have surfaced live.
What this didn't cover
Infrastructure and application performance only. No merchandising, pricing, ad-spend or email-marketing work — all of which move peak-season revenue considerably more than uptime does. I'd be cautious about attributing revenue growth to an engagement like this for exactly that reason: a store that stays up during peak can capture demand, but it doesn't create it, and any year-on-year revenue comparison is confounded by ad spend, catalogue changes and market conditions. The subscription product catalogue was migrated as-is with no restructuring, and customer service tooling was untouched.
Services used on this project
Want the same for your site? Start here: