Read on Twitter
1/ Been meaning to follow up with some extra thoughts on the piece @kenarchersf and I got published in @adexchanger on Monday.
As we were fleshing out the piece, we started to ponder what measurement Cohorts even ARE. https://www.adexchanger.com/data-driven-thinking/as-user-level-tracking-falls-apart-cohorts-offer-an-exciting-measurement-alternative/
As we were fleshing out the piece, we started to ponder what measurement Cohorts even ARE. https://www.adexchanger.com/data-driven-thinking/as-user-level-tracking-falls-apart-cohorts-offer-an-exciting-measurement-alternative/
2/ If memory serves, we broke Cohorts down into 4 types:
(and a little disclaimer: this is focused on MEASUREMENT, not targeting...whether a Cohort is addressable / targetable is a whole other ball of wax)
(and a little disclaimer: this is focused on MEASUREMENT, not targeting...whether a Cohort is addressable / targetable is a whole other ball of wax)
3/ FIRST-PARTY DETERMINISTIC COHORTS: Cohorts comprised of user-level 1P data. I.e., you have full access to the user-level data & you've grouped it into Cohorts, perhaps to be shared externally (thus protecting user privacy) or to normalize your data against external Cohorts.
4/ SECOND-PARTY DETERMINISTIC COHORTS: This would be like what you get out of Ads Data Hub. The underlying data is entirely deterministic, but advertisers can only interact with data at the cohort level. This protects user privacy & the intellectual prop. of the data owner.
5/ AGGREGATED COHORTS: E.g., observed site analytics UTM data; it's comprised of users, but it's aggregated. Insights would have to be probabilistic, because you're wanting to know how some campaign, event, or behavior affected various characteristics of these aggregate Cohorts.
6/ PROBABILISTIC COHORTS: These Cohorts would be defined by the PROBABILITY that its members meet some criteria. So the cohort exists as an ephemeral probability inside some larger data set. E.g., out of 500 users, we estimate 50% were exposed to an ad.
7/ Naturally, the Aggregated and Probabilistic Cohorts could originate on a 1P / 2P / or 3P basis, so there are some additional sub-categories here.
8/ If you accept that Cohorts are somewhat inevitable, the next fascinating part to me is the tech + organizational + math problem of making sense of these Cohorts.
9/ THE TECH PROBLEM 
: different Cohorts are going to be sourced and transported through different backend, advertiser & consumer tech. This is going to affect the composition of the data.
: different Cohorts are going to be sourced and transported through different backend, advertiser & consumer tech. This is going to affect the composition of the data.
10/ THE ORGANIZATIONAL PROBLEM 
: Cohorts beg the need for standardized taxonomies that allow Cohorts of varying sizes & characteristics to be normalized against one another. You can't "cheat" on taxonomy any more by normalizing at the user level.
: Cohorts beg the need for standardized taxonomies that allow Cohorts of varying sizes & characteristics to be normalized against one another. You can't "cheat" on taxonomy any more by normalizing at the user level.
11/ E.g., if you have multiple campaigns, agencies, technologies, etc collecting data and building Cohorts under different taxonomies, you're going to have a tough time normalizing Cohorts. E.g., if one Cohort is age 18-35 and another is age 21-40, how do you normalize them?
12/ How do you herd all those cats? 
13/ THE MATH PROBLEM 
: IANAM (I am not a mathematician) but:
A.) it seems solvable
B.) the bigger challenge seems to be getting marketers, tech, agencies, etc to adapt what they do to the mathematical realities of Cohorts
: IANAM (I am not a mathematician) but:A.) it seems solvable
B.) the bigger challenge seems to be getting marketers, tech, agencies, etc to adapt what they do to the mathematical realities of Cohorts
14/ E.g., if the marketer or their vendor/agency continues to run lots of personalizations or optimizations at a 1:1 (or sub-Cohort) level, the outcomes of user-level tweaks may end up completely inaccessible inside Cohorts! This reality may be anathema to much of #adtech.
