WHAT ARE WE TRYING TO ACHIEVE
We are using Google Analytics Cohort Analysis to AB test a feature. We generate a parameter on the client side and set 50% of users to "Yes" and 50% to "No". This parameter is then sent to Google Analytics, and used to create two custom segments:
- Feature Enabled: Feature "Yes"
- Feature Disabled: Feature "No"
We then wait for a week and generate two retention cohort analysis, one per segment, to determine which version of the feature (A or B) is better.
THE ISSUE
We noticed that Google Analytics returns significantly different version of data, apparently without reason. More specifically, it seems to have TWO different version of data, different enough to lead us to completely opposite conclusions, and to show one or the other depending on other random factors, such as whether I add a third segment to the analysis.
"Version A is clearly better. <\page refresh> Oh no wait, now version B is better!?"
Anybody else experiencing the same issue? Is this a bug on Google Analytics?