Time Series with multiple entries/values for one date and feature


I’m trying to forecast the product demand.
My raw data looks (really simplified) like this:

	Date	    Customer_id	article_id	indicator1	indicator2	order_size
0	2020-03-01	D021     	104     	True    	213.6   	10
1	2020-03-02	D034    	243     	True    	325.2   	15
2	2020-03-02	D034    	311     	False    	65.3    	43
3	2020-03-02	D054    	104     	False    	853.2    	5
4	2020-03-03	D021    	554     	False    	125.8   	67
5	2020-03-03	D093    	219     	True    	34.2    	11

I want to predict the order_size (maybe separately for every article).
I already written something that combines Embedding for the categorical data with an LSTM.
I really would like to use a RNN, but I don’t have a proper time series since i have multiple rows for one date, because there are multiple orders every day from different customers for different products.

I could just “force” it into one row by replacing the customer_id-column with a "no. of customer that placed an order that day’-column, but then I would lose important information about which customer orders which product, etc.

Any idea how I could manage this without losing information?