SQL (Snowflake) – 30 day look ahead / rolling average - sql

Question

I have two tables – purchases and activity.
The purchase table is structured like so:
|----------|----------------|----------|
| user_id | purchase_date | status |
|----------|----------------|----------|
| 1234 | 2020-01-01 | active |
|----------|----------------|----------|
| 2345 | 2020-01-10 | cancelled|
The activity table is structured like so:
|----------|----------------|-----------------|
| user_id | date | videos_viewed |
|----------|----------------|-----------------|
| 1234 | 2020-01-02 | 4 |
|----------|----------------|-----------------|
| 2345 | 2020-01-03 | 3 |
|----------|----------------|-----------------|
| 2345 | 2020-01-10 | 10 |
|----------|----------------|-----------------|
| 2345 | 2020-01-11 | 7 |
I am looking to query out a first 30 day activity average for each users' first 30 days based on a set purchase period.
The query I have written so far is this:
SELECT avg(t3.viewsperday)
FROM
(SELECT
date
,sum(t1.videos_viewed)/count(t1.user_id) as viewsperday
FROM activity t1
INNER JOIN (SELECT * FROM purchase c
WHERE status = 'active'
AND purchase_date BETWEEN '2020-01-01' and '2020-02-01') t2
ON t1.user_id = t2.user_id
where date between '2020-01-01' and '2020-02-01'
group by 1
order by 1 asc) as t3;
However, the problem here is that if a user purchased on 2020-01-31 I only get the first day of activity. I need help to figure out how to get the rolling average / look ahead 30 days from each purchase date – and get the average activity from those 30 days.
I suspect a window function would be appropriate here, but I am not sure how to formulate it as it is a bit outside of my knowledge. Any help would be greatly appreciated.

Answer 1

the following should work. I'm assuming that you want the average over 30 days even when there may have been zero views on some of those days? You may also need to adjust it slightly depending on exactly how you are defining the 30 day date range i.e. is the 30th day included, is the purchase date included, etc.
I've written it as an outer join so that even users with no views will be included
SELECT
P.USER_ID,
SUM(A.VIDEOS_VIEWED)/30
FROM PURCHASE P
LEFT OUTER JOIN ACTIVITY A ON P.USER_ID = A.USER_ID AND
A.DATE >= P.PURCHASE_DATE AND A.DATE <= dateadd(DAY, 30, P.PURCHASE_DATE)
GROUP BY P.USER_ID;
Update...
To get daily averages, try this (views on purchase date show as day 0, add 1 to the Day_after_Purchase formula if this should be day 1):
SELECT
(a.date - p.purchase_date) as Day_after_Purchase,
avg(A.VIDEOS_VIEWED)
FROM PURCHASE P
LEFT OUTER JOIN ACTIVITY A ON P.USER_ID = A.USER_ID AND
A.DATE >= P.PURCHASE_DATE AND A.DATE <= dateadd(DAY, 30, P.PURCHASE_DATE)
GROUP BY 1;