Calculate weekly retention in google big query - sql

Question

I have a big table in google big query and there are two columns on which I want to perform retention:-
Date user
2021-02-03 08:35:07 UTC foo#abc.com
2021-02-03 08:35:07 UTC foo1#abc.com
2021-02-04 08:35:07 UTC foo2#abc.com
2021-02-05 08:35:07 UTC foo#abc.com
2021-02-03 08:35:07 UTC foo1#abc.com
2021-02-10 08:35:07 UTC foo#abc.com
2021-02-13 08:35:07 UTC foo1#abc.com
2021-02-18 08:35:07 UTC foo3#abc.com
2021-02-21 08:35:07 UTC foo2#abc.com
2021-02-23 08:35:07 UTC foo2#abc.com
2021-02-24 08:35:07 UTC foo5#abc.com
2021-02-24 08:35:07 UTC foo2#abc.com
I want to calculate retention on the below condition:-
percentage of unique users for week1 present in week2
percentage of unique users from week2 present in week3 and so on.
The desired out format will be:-
week2 week3 week4
23% 56% 33%
I want to perform this on a time frame like one month or 6 months and whatever timeframe I choose the output should be in the above format.
I want a solution for Big Query but even a MySQL solution will help me.

Answer 1

Here is a possible solution:
WITH leads AS (
SELECT
user,
EXTRACT(ISOWEEK
FROM
`Date`) AS visit_week,
EXTRACT(ISOWEEK
FROM
LEAD(`Date`) OVER (PARTITION BY user ORDER BY `Date`)) AS next_visit_week
-- here you look the user's next visit and take the week. If the user is there the following week, next_visit_week = visit_week + 1
FROM
`your_project`.`your_dataset`.`your_table`)
SELECT
visit_week+1 AS `week`,
SUM(CASE
WHEN visit_week= next_visit_week-1
THEN 1
ELSE 0
END
)/COUNT(DISTINCT user)*100 AS retention_pct
FROM
leads
GROUP BY
`visit_week`
For each week, you count the number of times the next visit of a user occurs the week following the current week (NB: it can only occur once for each user). You divide the total by the number of distinct users.
You therefore obtain the retention rate for the following week (hence the '+1' in the "visit_week+1 AS week").