pay per verified sale sounds like it fixes ad fraud and it mostly just moves it. verified means the sale happened, it doesnt mean the ad caused it. if someone was already going to buy and clicked a link on the way, you paid for a customer you already had
This is the attribution versus incrementality distinction and it is genuinely expensive to resolve. The only clean answer is a holdout, deliberately not showing ads to a random slice and comparing conversion. Almost nobody runs one, because it costs real revenue during the test and because the result is frequently unflattering to whoever commissioned it.
留一个对照组这件事，在按成交付费的模式下更难推动，因为渠道方肯定不愿意，等于主动证明自己的价值没那么大。除非平台在规则里直接内置对照组，把它做成默认的一部分，否则不会有人主动去做。
BUILD THE HOLDOUT INTO THE PROTOCOL 🔥 5% of traffic never sees the promo, payout scales on the LIFT not the raw count 😤 suddenly everyone is selling something real
yeah a protocol level holdout is the version i could actually believe. its the only design where the person being paid isnt also the person marking their own homework
