Cohort analysis with Amazon Redshift / PostgreSQL

Question

I\'m trying analyze user retention using a cohort analysis based on event data stored in Redshift.

For example, in Redshift I have:

timestamp                 


        

Answer 1

Eventually I found the query below to satisfy my requirements.

WITH 

users AS (
  SELECT
    user_id,
    date_trunc('day', min(timestamp)) as activated_at
    from table
    group by 1
  )
,

events AS (
  SELECT user_id,
         action,
         timestamp AS occurred_at
    FROM table
)

SELECT DATE_TRUNC('day',u.activated_at) AS signup_date,


       TRUNC(EXTRACT('EPOCH' FROM e.occurred_at - u.activated_At)/(3600*24)) AS user_period,


       COUNT(DISTINCT e.user_id) AS retained_users
  FROM users u
  JOIN events e
    ON e.user_id = u.user_id
   AND e.occurred_at >= u.activated_at
 WHERE u.activated_at >= getdate() - INTERVAL '11 day'
 GROUP BY 1,2
 ORDER BY 1,2

It produces a slightly different table than I described above (but is better for my needs):

signup_date  user_period  retained_users
-----------  -----------  --------------
2015-05-05   0            80
2015-05-05   1            60
2015-05-05   2            40
2015-05-05   3            20
2015-05-06   0            100
2015-05-06   1            80
2015-05-06   2            40
2015-05-06   3            20