<b>"I A/B tested both networks for a week each"</b>

<b>"I A/B tested both networks for a week each"</b>

Credit where due: alternating is smarter than most people bother to do.

But display revenue carries an attribution and reconciliation tail. Week-one's reported earnings include impressions still settling, and week-two's network gets credited with demand primed during week one. Sequential testing bleeds across the seam. You measured a handoff, not two clean states.

🚩 Hidden variable: the reconciliation lag overlapping your switch date. Most networks finalize 3 to 7 days out; some exchanges, longer. Your clean A/B has a smeared boundary you can't see in daily totals.

If you must sequence, throw away the first and last 5 days of each arm. Better: split by page-group simultaneously so both networks face the same week's demand.

Not saying A/B testing is bad — saying the comparison is rigged.

—
Чтобы быть в курсе рынка — подпишись на @affcareers_tbilisi