The effect is real
Slow pages lose visitors before anything else on the page gets a chance. This is not controversial and it is worth fixing. The problem is the numbers attached to it.
Why the famous figures do not transfer
Most quoted speed-to-revenue statistics come from very large retailers, where a 0.1s improvement across hundreds of millions of sessions produces a measurable revenue line. That arithmetic does not survive being applied to a site with a few hundred visits a month.
It is also survivorship-shaped: the companies publishing these results had the traffic to detect the effect. Sites where speed work produced nothing measurable do not publish case studies about it.
Where speed work genuinely pays
- Going from bad to acceptable. A page taking six seconds to become usable is losing people outright, and fixing that is unambiguous.
- Mobile on poor connections, where the gap between a heavy and a light page is measured in whole seconds rather than milliseconds.
- Anywhere a visitor is comparing you against alternatives in adjacent tabs.
- Crawl efficiency on large sites, which is a separate benefit from conversion.
Where it stops paying
Past roughly the point where a page feels immediate, further optimisation competes with everything else you could spend that engineering time on. A site loading in 1.4 seconds with an unclear value proposition does not have a speed problem.
The honest sequence is: fix genuinely slow, then fix the message, then return to speed if the numbers justify it.