Lazy loading ad units sounds like an obvious win — load ads only when they're about to enter the viewport, red

Lazy loading ad units sounds like an obvious win — load ads only when they're about to enter the viewport, reduce initial page load time. But the revenue math is more complicated.

The problem: when you lazy load, SSPs often can't prefetch bids. By the time the unit enters the viewport, you're running an on-demand auction that may have worse demand than a pre-fetched auction would.

Also: some DSPs apply viewability-based bidding. If you lazy load properly, your viewability scores improve — which can attract higher CPMs from buyers who weight for viewability. The effect can offset the latency penalty.

The correct approach: test lazy loading with a holdout. Run 50% of sessions with lazy loading enabled, 50% without. Measure RPM (not CPM in isolation) across the full session. CPM might go up but fill rate might drop if demand is thinner in on-demand auctions.

Don't optimize in one direction without measuring the other.