I want to use 'analyse in Excel' for a simple Power BI Model with a star schema in it (3 dimensions, 1 fact table, all dimensions only linked to the fact table) but some users of us don't understand that when they choose 2 dimensions without choosing any measure you get a 'crossjoin'; every possible combination is displayed untill you choose a measure.
Let's see we have 200 clients and 500 products; you get every client for every product, while in fact, it's not the reality. The reality is that 1 client buys only a few products. When you choose measure 'profit' everything works fine. But when you don't the crossjoins comes.
In my opinion it's completely explainable; region and manager have NOTHING to do with each other when you don't use the connecting facttable. So you'll get all the possible combinations. For me not a big deal; you never ask 'which client product combination has been there', but you ask 'which profit (or number of products or whatever) is made for client X and on which products did it happen'. But some user find this is very difficult and to be honoust; we're migrating from some old multidimensional OLAP cubes and there it went good without selecting a measure; when you ask this question to a multidimensional cube it gives you only the combination of products and cliënt that are part of the fact table; even if you don't select anything from fact table.....
Is this just the difference between multidimensional (prepared) and tabular (on the fly) technology or can we do something to make our Analyse in Excel to only present the common combinations?
Power BI internally host a temporary Tabular database for each opened .pbix file. So if you create a model within Power BI Desktop, you can't avoid "direct cross join" for multiple dimensions entries without putting measure into visual.
Currently, for your requirement, I think you can only "connect live" to your Multidimensional OLAP cube in Power BI Desktop.
So my feeling that it's a tabular behaviour is right? And I can't change this behaviour in Tabular?
Your cube idea is good but not the way we want to use; we're full in Azure and don't want to create Multidimensional cubes.