03/14/2023, 4:07 PM
Hello, I wanted to ask a question about conversion windows and ratio metrics meaning data from outside of the experiment is including in calculations for the experiment. If I have a ratio metric where the denominator can occur substantially (in time) before the numerator, from what I can see I need to include a conversion window value. For my use case, this can be the length of the experiment, so might be high 10s of hours up to anything around 100 or 200. Setting a conversion window to 100 for example means that the endDate values in the generated query end up getting timeshifted by +100 hours, so if I’m doing retrospective analysis (or I have phases set up), I end up including data from outside of the time period of interest. I’m aware of the startDate and endDate jinja keywords that I can include in the definition of the metrics, but these end up getting affected by the conversion delay/window values too. The only workaround at the moment seems to be hardcoding date filters into the queries retrieving the data for the metrics which isn’t viable for anything other than playing around to work out how Growthbook functions. Am I missing something or am I doing something incorrect in my setup? Thanks in advance.


03/14/2023, 5:02 PM
I think your understanding the current situation perfectly well. This is indeed a limitation and we are (right now, actually!) adding a feature where you can use all data from (A) experiment exposure + any conversion delay to (B) the current time/end of the phase/experiment. Thus effectively ignoring the conversion window.
Also, just to confirm, our proposed new analysis type would solve your issue, right?


03/14/2023, 7:05 PM
hey Luke, thanks for coming back to me. I think that would absolutely solve the issue. I'll keep an eye out for that update!