Page speed affects revenue through a mechanism simpler than the technical discussion around it suggests. Visitors leave before the page appears, and they are never counted as visitors at all.
Abandonment happens before measurement
Analytics scripts typically load with the page. A visitor who leaves during loading may never fire the tag that records the session.
That means the reported bounce rate excludes the worst outcomes, and the slowest experiences are the least represented in the data.
The gap between clicks recorded by an ad platform and sessions recorded on the site is the closest available measure of this loss.
Paid traffic is the most fragile
Someone who typed the address or followed a bookmark has already decided to visit and will tolerate a wait.
Someone who tapped an ad has no such commitment, and returning to the feed is trivially easy.
Because the advertiser has already paid for that click, every abandoned load is a purchased visit that produced nothing.
The delay is usually caused by tags
Most marketing pages are not slow because of images or hosting. They are slow because of the accumulated collection of measurement and personalisation scripts.
Each tag was added for a defensible reason, by a different person, and almost none are removed when the reason expires.
Scripts that load synchronously block rendering entirely, so a single slow third party can hold the whole page while the visitor waits.
Personalisation tools are the worst case, because they deliberately hide the page until they have decided what to show, and a failure there leaves a blank screen.
Mobile networks amplify everything
A page tested on office broadband behaves very differently on a congested mobile connection, and mobile is where most ad clicks originate.
Processing power matters as much as connection speed, since scripts have to execute after they arrive and older devices execute slowly.
Testing on the fastest available conditions therefore systematically understates the problem for the audience that matters most.
What to check before rebuilding anything
Compare platform clicks against site sessions per campaign. A large and consistent discrepancy points at load abandonment rather than at tracking configuration.
Audit the tag list and remove anything whose owner cannot be named, which usually accounts for a meaningful share of load time.
Then measure the time until the first meaningful content appears rather than until the page finishes, because the visitor is judging the former.