Split-test two networks on the same offer cleanly

Split-test two networks on the same offer cleanly
Run them side by side or you're comparing weather, not networks.

— Send the same source split 50/50, same hours, same creatives
— Match on geo and device — un-matched splits bias EPC ~25%
— Gate the read at n=400 conversions per arm, not per click
— Compare net EPC after chargebacks, not gross approvals
— Hold the test ≥2 weeks to average out day-of-week effects

Arm A EPC $0.44 ▏▏▏▏ vs Arm B $0.38 ▏▏▏
Difference significant only at n≥400/arm.

Read: a network split-test is only honest when every variable except the network is frozen.