I am working in a industry. Most of the time I do statistical reports on sales. I'm new to time series analysis, so please be patient: the question might be obvious.
I would like to monitor the sales of every product in every territory of the country across time. Since sales are affected by seasonality and variability, I think that decomposing time series with procedures like STL, would be useful to extract underlying trends in the data. Percentage changes are not as useful because they are affected by outliers, variability of data and they miss trends too.
I would like to understand for every product where sales are going down and where they are going well. The aim is not to forecast sales to plan production, I just want to understand how sales are going for every product and territory. The only way I know to tackle this problem is to decompose each time series separately, one for each combination of product and territory. This means that if the industry makes $m$ distinct products and the nation is composed of $k$ territories, I need to decompose $m$ x $k$ univariate time series. This process would take some time because for every time series I need to: understand if the series follows an additive or multiplicative model, calibrate the parameters of the decomposition method used, assess the quality of the decomposition.
I would like to know if my approach is common or if there are other ways to solve the problem, perhaps using multilevel models.