https://www.growthbook.io/ logo
p

polite-balloon-1280

07/14/2023, 1:11 PM
Hi all! [Help with Namespaces] We are using the Namespace's Feature to separate the population of two A/B test that we are running. We set 49% for "login-social" and 50% for "Onboarding - Security Info" but instead, we have a lot of recording of hotjar demonstrating users that see both tests. Someone can help us find what we are making wrong?
đź‘€ 1
f

fresh-football-47124

07/14/2023, 8:29 PM
What are you using for the user attribute for assignment?
p

polite-balloon-1280

07/14/2023, 8:43 PM
We use the "customerIdHash" that is a unique ID for each user in our base of signup users
f

fresh-football-47124

07/14/2023, 8:48 PM
is it possible that this ID is less persistent than the ID hotjar is using?
p

polite-balloon-1280

07/14/2023, 9:11 PM
Apparently, not! Because we can see this same pattern analyzing the event "Experiment" that is set when a user gets in a test. In the image, you can see that the conversion between the both tests are high
f

fresh-football-47124

07/14/2023, 9:57 PM
ya, that’s not great. Do you have any other identifier you can use?
is that a hotjar metric?
p

polite-balloon-1280

07/14/2023, 10:08 PM
The "customerIdHash" is an ID that we use for all of our customers. It's a hash created with the CPF, a unique document number that identify a person in Brazil. We use this hash for management of all journey events, A/B tests and information about our signup users. 🫤 In Hotjar, we don't use the "customerIdHash". We don't integrate the Hotjar associating this hash, and we can't identify a user in Hotjar. In Hotjar we just can filter with the test ID and the variation name.
f

fresh-football-47124

07/14/2023, 10:09 PM
so there are only a few ways a user can get double exposed…
GrowthBook uses deterministic hashing to do the assignment, so the same inputs result in the same output. In our case, we take the experiment key, and the hash ID, to give us a value from 0 to 1. The ways a user could see both are: • The hash ID changed • The experiment key changed • The variation split changed (ie, if you started at 90/10, and then made it 50/50, you’d risk double exposing up to 40% of your users.)
🤔 1
p

polite-balloon-1280

07/14/2023, 10:23 PM
ok! I will analyze a possible problem that I think that would cause a change in hash. Thank you, for your help.