Demand forecast for many items

Question

Good afternoon, folks. I’d like to hear your opinion on the following: I am working in an e-commerce company, which has more than 1000 items for sale, and each item of this, has a series of daily sales history, from 2019 to today. The challenge is to make a sales forecast for each item, based on your past sales history. To that end, the doubts I have are as follows::

1- which model to use?

2- which software is best suited for this forecast? being that I have more than 1000 products, and being that for each product is a different historical series

3- I will have to make a projection for each item? I see no sense in it, for the effort would be tremendous, and it would take a long time, but if this is the case, tell me!!

Note: the data are contained in an excel spreadsheet in the following format: each column is a SKU (product) and each row is the total number of sales of that respective sku, and the respective weekly date.

Please if anyone who has ever worked with sales forecast in any company, save me!!! I will be grateful.

Answer 1

The process of time series analysis encompasses unit root tests, seasonality, trend, but as there are many time series I will give a suggestion that probably would not in a normal situation. I recommend using univariate time series models, they use the history of the series to predict the behavior of itself, the library forecast of R has a function auto.arima seeking the best ARIMA model minimizing an information criterion of your choice.

If you search for ARIMA models in R you can get some information on how to do this, but I believe studying univariate models and the library forecastare a good way and that the function auto.arima helps save time sometimes.