Daily consumption delta based on purchase dates - mysql

Question

I need to make a (Tableau) daily graph depicting consumption dynamics against previous day grouped by those clients who increased consumption, decreased consumption, and net change overall. Sample is below.
Calculation logic for sample: for every day for every client calculate difference vs previous day for that client, sum those above 0, sum those below 0, sum total.
The sample was made manually from a relatively small data set.
The real table has over 2 mil rows, and is not very consistent in that clients start buying at different days, may skip various periods buying nothing.
Initial table structure is like that:
client_id date sales
1 2018-09-01 4
1 2018-09-02 5
1 2018-09-04 3
2 2018-09-1 2
2 2018-09-2 2
While calculating table difference per date is simple, calculating pure growth and pure churn is hard, because the date row is not continuous for all clients.
I thought of adding the delta_to_previous column to each row when loading the initial dataset from the data storage, like:
WITH orders AS (
SELECT client_id,
date,
SUM(sales) as sales
FROM dwh_orders
GROUP BY client_id, date
)
SELECT
client_id,
date,
sales,
LAG(sales, 1) OVER (
PARTITION BY client_id
ORDER BY date
) as prev_date_order_value,
sales - LAG(sales, 1) OVER (
PARTITION BY client_id
ORDER BY date
) as prev_date_order_delta
FROM
orders;
Then for each date I can just show sum of positive values, negative values, total.
Problem, this approach will show consumption change at the next date of purchase, and if client buys 5 items on March 1 and then 5 on May 1, there will be no change for him at all. What it should do is show -5 for March 2 and +5 for May 1.
I am a bit puzzled at the optimal approach to this. The general solution could also use some review probably.
If someone dealt with a similar problem, I could really use your advice.
If you are experienced with sql, I could use your advice on how to convert the initial dataset (see sample above) into something like
client_id date sales delta
1 2018-09-01 4 0
1 2018-09-02 5 1
1 2018-09-03 0 -5
1 2018-09-04 3 3
2 2018-09-1 2 0
2 2018-09-2 2 0
If you know a bit about Tableau, I could use help on building graphs like this using its tools.

Answer 1

with cdates as (
select client_id, min(date) as dte, max(date) as maxd
from dwh_orders
group by client_id
union all
select client_id, dateadd(day, 1, dte), maxd
from cdates
where dte < maxd
),
cd as (
select client_id, date, sum(sales) as sales
from dwh_orders
group by client_id, date
)
select cdates.client_id, cdates.date,
coalesce(sales, 0) as sales,
(coalesce(sales, 0) -
lag(coalesce(sales, 0)) over (partition by cdates.client_id order by cdates.date
) as delta
from cdates left join
cd
on cdates.client_id = cd.client_id and
cdates.date = cd.date
option (maxrecursion 0);