Get a rolling order count into session data - google-bigquery

Question

I have the following table
One client has two purchases in one session.
My goal is to assign a order counter to each row of the table.
To reach this goal I am using the lag function to call the last order_id and the last order_timestamp:
SELECT
lag(event_timestamp) over (partition by session_id order by
ecom_data.order_id) as prev_order_timestamp,
lag(ecom_data.order_id)
over (partition by session_id order by event_timestamp) as
prev_order_number
From table
My desired output is this:
Problem, I do not get the previous order time. Instead I get the event_timestamp from the previous event.
My second challenge is that I do not know how to assign a order_count. My desired output is this:
Ideally, this order count should be rolling as in the real data set I dont know how many orders in total each session had. There can be 0 - infinite orders.
Can you help ?
Thank you!

Answer 1

### create sample table (helps to introduce these in your questions)
WITH
base AS (
SELECT
'A' AS client_id,
1 AS session_id,
'pv' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:17:41") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'ts' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:17:42") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'pv' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:27:14") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'atc' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:27:15") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'p' AS event_name,
123 AS order_id,
DATETIME("2022-05-12 10:30:47") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'pv' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:30:50") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'pv' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:31:01") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'atc' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:31:20") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'ts' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:31:22") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'rv' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:31:32") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'pv' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:31:35") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'pv' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:32:49") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'p' AS event_name,
456 AS order_id,
DATETIME("2022-05-12 10:33:35") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'pv' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:33:48") AS event_timestamp
UNION ALL
SELECT
'A' AS client_id,
1 AS session_id,
'tv' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:33:50") AS event_timestamp
UNION ALL
SELECT
'B' AS client_id,
1 AS session_id,
'pv' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 10:31:50") AS event_timestame
UNION ALL
SELECT
'B' AS client_id,
1 AS session_id,
'p' AS event_name,
123 AS order_id,
DATETIME("2022-05-12 10:33:50") AS event_timestame
UNION ALL
SELECT
'C' AS client_id,
1 AS session_id,
'pv' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 11:13:50") AS event_timestame
UNION ALL
SELECT
'C' AS client_id,
1 AS session_id,
'pv' AS event_name,
NULL AS order_id,
DATETIME("2022-05-12 11:33:50") AS event_timestame),
prev_order1 AS (
SELECT
*,
LAG(order_id) OVER (PARTITION BY client_id ORDER BY event_timestamp) AS prev_order_number1
FROM
base),
### filling in order number using your requested output
prev_order2 AS (
SELECT
*,
MAX(prev_order_number1) OVER(partition by client_id ORDER BY event_timestamp ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS prev_order_number2
FROM
prev_order1
ORDER BY
event_timestamp )
### inserting order_counter logic
SELECT
*,
DENSE_RANK() OVER(partition by client_id ORDER BY prev_order_number2) - 1 AS order_counter
FROM
prev_order2
Think about edge cases and perhaps if you want to partition by other dimensions such as client_id vs total table (as you have it now). I included client_id = B as an example.