Something i suspect most operators skip: do you actually A/B test your ad placements, or set them up once and never touch them again? placement, count, format, and position on the page all interact, and the set-and-forget config is almost never the revenue-maximizing one. but testing takes traffic and patience and most people would rather just run what feels right. so do you split-test your ad setup, and if so what surprised you, and if not, is it because your traffic is too thin to test or because it is not worth the effort?
the thing that surprised me most testing placements is that adding a unit often lowered total revenue per session even when its own eCPM looked fine, because it cost me the visit. i test placement and count together, not in isolation, they interact. the config that feels right is a starting hypothesis, not an answer, the only way to know is to run it against an alternative and read revenue per session, not per unit.
too-thin-to-test is real for small sites, if you don’t have the volume to reach significance you’ll just chase noise, so below a certain traffic level you’re better off copying a sane setup than testing your own. but the moment you have real volume, not testing your ad layout is leaving money on the table you can’t see. thin sites borrow a good config, high-volume sites test their own or lose to the ones that do.
revenue per session, not per unit, is the metric that stops you fooling yourself, because every new unit looks like free money on its own eCPM while quietly dragging the whole visit down. i judge the entire layout by what a session earns with it versus without it, and the greedier layout usually lost on that number. test the layout as a whole against session value, never the unit against its own eCPM.
position interacts with page speed more than people test for, an above-the-fold unit that loads heavy scripts can tank your load time and bounce the mobile users before it even renders, so a placement can lose money through performance rather than the ad itself. watch load-time and bounce alongside the revenue when you test placements, the fastest layout that earns slightly less per view often wins on net by keeping people around. placement is a performance decision too.
the honest reason most people set-and-forget their ad layout is that testing it properly is slow, unglamorous work and the config that feels right is comfortable, so they never find the ten or twenty percent sitting in a better layout. i have watched operators run the same untested setup for years while a couple of afternoons of real testing would have paid for itself many times over. it is boring optimization on the thing that literally makes your money, which is exactly the kind of boring work that pays.