We shaved a second off load time and nothing happened

We shaved a second off load time and nothing happened

A subscription site spent six weeks chasing Core Web Vitals after reading that every 100ms costs conversions. They cut Largest Contentful Paint from 3.1s to 1.9s. Conversions: flat.

The clue came from segmenting by page speed bucket against checkout completion. The slow pages weren't where people dropped. Checkout already loaded in 1.2s. The real leak was a confusing two-step address form mid-funnel.

Speed advice is real, but it's a multiplier, not a fix — it can't rescue a funnel that fails for a reason other than waiting.

We rebuilt the address step into one screen. Checkout completion rose from 54% to 68% in three weeks, dwarfing anything the speed work delivered.

—
Lesson: page speed matters, but only fix it where the data shows people leaving while waiting. Diagnose the drop-off before you optimize the stopwatch.