lemon-library-48160
10/04/2023, 1:11 PMmiddleware
and, if not already set, they set a unique_id
cookie, that we later use as the id
attribute when initializing the Growthbook SDK
• when the application is mounted client-side, we initialize the Growthbook SDK with the id
specified above
• in our trackingCallback
, we push an experiment_joined
event with the experiment_id
and variant_id
to our GTM data layer, which, in turn, sends a GA event
We set up an A/B Experiment inside our signup flow, that evaluates the value of our experiment (via growthbook.getFeatureValue
) right before we want to display the tested UI component (there are some extra pre-conditions to the test that are evaluated first). The experiment is running for 100% of traffic, split in equal 33.3% for the Control and 2 Variants we are testing.
Everything seems to be normal, but after we deployed this to production, it seems like a lot of the sessions were not part of the experiment. I checked both our sign_up
and experiment_joined
events via Big Query and it seems like more than 50% of `sign_up`s did not have a corresponding experiment_joined
event, even though all conditions seem to have been met.
Some extra details in case they help:
• I am fairly confident the id
is already set when the experiment is evaluated, so the user should not be excluded from the experiment - we have Sentry events that should let us know otherwise
• We have no other targeting rules set up for the feature flag or the experiment
• I am confident in the queries we are running for sign_up
events and they include the same pre-conditions of the A/B experiment
• We are using Big Query as a data source in Growthbook
• We are querying both daily and intraday tables of the events in BigQuery with this filter:
FROM
analytics_<redacted>.`events_*`
WHERE
(
(_TABLE_SUFFIX BETWEEN '<START_DATE>' AND '<END_DATE>')
OR (
DATE(CURRENT_DATE()) BETWEEN DATE('<START_DATE>') AND DATE('<END_DATE>')
AND REGEXP_CONTAINS(_TABLE_SUFFIX, '^intraday')
)
)
• We are using the Enterprise version of Growthbook
• We are running version ^0.17.0
of @growthbook/growthbook-react
Do you have any suggestions for things I should try out next, to get this issue solved?
Thanks!helpful-application-7107
10/04/2023, 11:58 PMI checked both ourAs in when you join or group by these events onandsign_up
events via Big Query and it seems like more than 50% of `sign_up`s did not have a correspondingexperiment_joined
eventexperiment_joined
id
there are 50% of id
with at least one sign_up
event who do not have a corresponding experiment_joined
event? Just ruling out that there's some issue with sign_up
events firing too often.
• I am confident in the queries we are running forTo confirm, they are querying the same table?events and they include the same pre-conditions of the A/B experimentsign_up
• in ourAre there any differences (besides the fact that the tracking callback is getting called within the GB SDK) between this and how you track sign up events? Is it the case that you expect every single, we push antrackingCallback
event with theexperiment_joined
andexperiment_id
to our GTM data layer, which, in turn, sends a GA eventvariant_id
id
that fires a sign_up
event to also fire an experiment_joined
event?lemon-library-48160
10/05/2023, 8:48 AMexperiment_joined
events. For the 0.5% that is missing, we are assuming it's people with cookies disabled, or in-app browsers that do not handle our cookie.
To quickly answer your questions, yes, we have all events in the same table(s) in BigQuery, the events were fired in the same way, and we were expecting everyone that had a sign_up
event through a specific flow, to have a corresponding experiment_joined
event.
I appreciate you for jumping on my questions and offering help, thanks again! 🙌