The A/B test that "proved" lazy-loading killed revenue

The A/B test that "proved" lazy-loading killed revenue

The claim: a publisher turned on aggressive lazy-loading, saw revenue drop 8% in week one, reverted, and declared it a mistake.

The asterisk: week one is inside the demand-partner reconciliation lag. The drop they saw was estimated revenue, which under-reports recently-served viewable impressions while header-bidding partners are still settling.

🚩 Hidden variable: the measurement window versus the settlement window. Viewability-optimized impressions pay more but get confirmed slower, so the early estimate looks worse than the final number.

When the same month settled 50 days later, lazy-load was up 5% on RPM because viewability rose.

Not saying lazy-loading always wins — saying you cannot judge a viewability change inside the estimation window. You measured the lag, not the lift.