Hi! Could you please explain how Segments work? The query requires a

user_id

and a `date`: could you explain what the date is for? I could not find any documentation on it. From the FAQ I understand that this is mostly used to correct/filter experiments and limit them to a subset of

user_id

, but I cannot understand the role of the

date

column.

We use the date column to determine the date when the user was exposed to your experiment. This is important as GrowthBook supports conversion windows, and also phases, where if you change something in your experiment, and want to only look at the experiment results from time x.

Thanks, after exploring it some more I understand that the

date

must occur before the experiment

timestamp

for users in this query to be included in the experiment results for the segment. Is that correct? If we go back to the premium users example from the FAQ. What would the date be then? Would it be the moment that they joined premium?

SELECT
  user_id
  joined_premium_date AS date
FROM
  my_user_table
WHERE
  account_type = 'premium'

If so, what if they turned off premium since then? Shouldn’t this type of condition require a start and end date (user X had a premium account between date Y and Z)?

yes, generally - though you can set negative windows

OK I see, thanks that’s helpful.

Thanks for posting this question. I had the same one and this resolved some confusion. I’m still wondering though about what Olivier wrote “I understand that the

date

must occur before the experiment

timestamp

for users in this query to be included in the experiment results for the segment.” Which

timestamp

is the

date

being compared to/how are they being compared? How does the conversion window (and maybe attribution model) play in?

Hi @ambitious-apartment-58735! I was referring to the

date

in the definition of the segment (a segment query must provide a

user_id

and a

date

) and to the timestamp in the definition of the experiment (an experiment must provide a

user_id

, a

timestamp

, an

experiment_id

and a

variation_id

). According to my test, the

date

from the segment must occur before the

timestamp

for a

user_id

to be included in the experiment with the segment.

What happens if there are multiple

date

and

timestamp

per user_id?