Hello! Me again 😆 And again thanks for a great tool! We have had some internal discussions on the conversion window(CV), and that for many teams it does not make sense to actually have a limit. We have solved this by setting CV = 9999999 and adding the experiment’s start and end date to the metrics, but it is not bullet proof. Better might be to add more possibilities to choose another type of CV. I.e. when should, in theory, a conversion window actually start and end? Some of the answers are: 1. From the first

experiment_viewed

to first

experiment_viewed + window_length

2. From the last

experiment_viewed

to last

experiment_viewed + window_length

3. From the first

experiment_viewed

to last

experiment_viewed + window_length

4. From any

experiment_viewed

event to

experiment_viewed + window_length

(multiple conversion windows) In GB it is currently 1, but in many use cases we would prefer 4 or at least 3. Is this something you have considered, or have thoughts about?

That makes sense. Supporting #3 is fairly easy (just need to change some

MIN

to

MAX

in the query). #4 might be trickier since we'll need to remove the group by and add a later aggregation step.

Hi again! 😊 Just realized from this that the end date of the metric is actually not the end date of the experiment as I thought. So our current work-around doesn’t actually work (unless we never update the queries after experiment ended, but we often need to… ). So if we could get the #3 (see👆🏻 ) up and working soon, that would be a huge help! That way first experiment viewed event is the start, and the last experiment viewed + window_length is the end of the metric count. Right? Or do you know how others solve this? Is it only us that want to count conversions in, and only in, the whole time frame?

We have a PR that we're hoping to launch this week - https://github.com/growthbook/growthbook/pull/479

Nice! Thanks for the info and efforts! 😊

Thanks for working on this, Jeremy!

@full-island-88199 @stocky-energy-64916 After thinking this through some more, moving our default behavior from #1 to #4 (multiple conversion windows) might make the most sense. So every exposure event for a user will have its own conversion window. Any metric that happens within any of the windows will be included. Question for you: Are there any situations where you would want to keep the existing behavior? (first exposure event only)

Hello! Yes, I also think that makes the most sense. I can’t think of any situations where we’d need to keep the exiting behavior, but I’ll quickly run it by the rest of our team too 🙂 Thanks for looking into this!

Hi - also actively following this PR/issue. We would also have no need to keep the first option.

Just merged this! I kept the default behavior the same for now, but you can switch it on a per-experiment basis. I'm using the term "*Attribution Models*" to describe the different ways to include/exclude metric conversions in an experiment. • First Exposure - Single conversion window based on the first exposure event (default, current behavior) • All Exposures - Multiple conversion windows, one for each exposure event You can change the attribution model in the Analysis Settings section on experiment pages. Let me know how its working. I want to make this the new default behavior in a future release.