morning, I wanted to check whether with the latest version there is a way to do 'all exposures' experiments? I see from the docs that this feature seems to have gone away, however we've got users at our org who would like the option to do this. Is there a workaround, or is this functionality not possible to reproduce?

Hi Matthew, there is not currently a way to do that kind of analysis. It was becoming complex to maintain and the performance of the query it requires does not scale well at all. There isn't a way to directly reproduce it, but I wonder if the Experiment Duration model would suffice for those users? The main difference is that if there is a gap between a users exposures that is not covered by a conversion window, the Experiment Duration will count conversions in that gap.

Thanks Luke, useful to understand where we're at with this type of functionality as well as the rationale. Out of interest is this to do with the queries generating very large tables as a result of the joins in the generated code? Just to check I understand, under the old 'All Exposures' functionality with say conversion window of 2 hours, event data would only be included for the timestamp of each exposure + 2 hours, so you could focus in on 2 hour windows immediately after the exposure. Under the new 'Experiment Duration' model, does it basically include all event data between first exposure and end of experiment/phase? So its not possible to focus in on those 2 hours windows as before? Thanks again

Under the new 'Experiment Duration' model, does it basically include all event data between first exposure and end of experiment/phase? So its not possible to focus in on those 2 hours windows as before?

Yes, that's correct. You can focus in on the first 2 hours after the first exposure only, but not after each exposure.

OK good to understand. Yeah we'd certainly found that you could inadvertently produce some super big queries (if of interest we also saw that you could trip into large queries if you used an a high volume metric as the activation metric and also had some high volume metrics within your experiment)