Re:Facebook will count one person as two on its platform
by hardtke on Monday October 18, 2021 @12:56PM
(#28907708)
I'm curious how they would handle this with regards to lift tests. You can ask Facebook to not show your ads to faction of the addressable audience, and then calculate sales for the treated and control groups. If they don't make the randomization binding across platforms, it can several reduce the incrementality measurement (effectiveness of Facebook ads).
[ Parent
| on HN ]
Re:Facebook will count one person as two on its platform
by jsnell on Monday October 18, 2021 @01:07PM
(#28907830)
Just key the experiment diversion on the email address rather than an internal user ID. That way the experiment and control groups will be consistent across the apps, with the downside that a user changing their email might migrate between groups. But changing your email has to be a very rare operation.