can someone explain segments vs dimensions, aren't both a way to drill down into a subset of users?

Yeah, it's a little confusing right now. Dimensions are user attributes (browser, country, account type, etc.) that could have multiple values. So

browser

might have values Safari, Chrome, Firefox. Segments on the other hand identify a single group of users. So "Firefox users in EU" for example.

ahh okay good to know

Can Segments be used with imported experiments? Would this be an option to get significance testing for a subset of users (as splitting by dimension does not allow it?)

We don't support adding a segment to an experiment right now, but I think it could make sense. Similar to an Activation Metric, it can help in those cases where the exposure logging includes people who did not actually view the experiment. One common example would be a change that only affects mobile browsers, but the exposure is logged from the server.

Another example would be the exposure logging a page view, but some of its features are only visible to currently paying customers

I'll write up a GitHub issue for it. Shouldn't be too difficult to add since we have all the pieces already

Sounds great, thanks!

There's a PR in progress to add this feature (https://github.com/growthbook/growthbook/pull/127). Here's a mock-up of the new analysis settings for an experiment. Let me know if you have any feedback or if the UI isn't clear!

seems good! the subqueries would cover every schema available under the configured data source connection, right? would be nice to have an example of use somewhere in the docs for it

Yeah, you can basically do

user_id IN (
  SELECT user_id FROM ...
)

with any arbitrary query to only include specific users. Or use the same with

NOT IN

to exclude specific users.