<b>Deep dive: the conversion cost of latency is front-loaded</b>
The stat everyone quotes — "a 1-second delay cuts conversions by 7%" (Akamai/Aberdeen lineage) — is true but flattens a curve that's actually steepest at the start.
Google's mobile speed research found that as page load goes from 1s to 3s, bounce probability rises ~32%; 1s to 5s, ~90%; 1s to 6s, ~106%; 1s to 10s, ~123%. Read the shape: the damage per added second is brutal early and the curve bends. Deloitte's "Milliseconds Make Millions" study found a 0.1s improvement in mobile load lifted retail conversions ~8% and lead-gen ~10%. The marginal value of the <i>first</i> 100ms dwarfs the marginal value of optimizing from 4s to 3.9s.
The mechanism is expectation violation, not raw waiting. Users don't experience load time on an absolute scale; they experience it against a forming expectation. A blank screen past ~1s reads as "broken," and abandonment is a rational exit from a page presumed dead. This is why perceived performance (skeleton screens, progressive rendering, above-fold-first loading) often beats actual performance — you're managing the expectation, not just the milliseconds.
The affiliate-specific trap: paid traffic is your most expensive, least patient visitor. They didn't seek you out; they were interrupted by your ad. Their tolerance for a slow first paint is near zero, and you paid for the click whether it renders or not. Largest Contentful Paint on the hero is therefore your highest-leverage technical metric — get the first meaningful screen painted fast and defer everything below the fold.
<b>TL;DR</b>
— Bounce probability climbs ~32% from 1s to 3s and ~90% to 5s (Google) — latency damage is front-loaded, so the first seconds matter most.
— Deloitte found a 0.1s mobile improvement lifted conversions ~8-10%; the first 100ms beats shaving 4s to 3.9s.
— Optimize perceived performance and hero LCP first — paid traffic is the least patient and you've already paid for the click.
Above Fold Lab
@AboveFoldLab
<b>Deep dive: the conversion cost of latency is front-loaded</b>
Этот пост опубликован в Telegram-канале Above Fold Lab. Подписаться можно по ссылке: @AboveFoldLab.