Advice on making a "baby" self service dataset based on the large "golden"dataset

[BobKoenen]

Hi all,

I have a Large "golden"dataset with tons of tables, columns and measures on which multiple thin Reports run. Now I have the bussinesscase that some employees want to use our data to Analyse in Excel. It would be ideal to let them connect to a dataset via Excel. However the big existing dataset is to complex for them so I want to offer them a simple dataset containing only the key tabels, columns and measures they would need but using the same datamodel and businesslogic as the "big golden dataset". 

I was thinking of using the composite model feature for this where I create a new power bi file, Connect to the exisiting "golden" dataset and then setting it to Direct query.  Then Simplify the model and publish it as a Self Service dataset. 

I was wondering what your ideas where and if I am overlooking something with this option. 

[Greg_Deckler]

@BobKoenen I *think* you need to do the reverse. Create the baby data model and then use the composite model feature to extend the baby model to the full model. And yes, I realize that is not what you wanted to hear.

[BobKoenen]

You are right I do not want to hear that. Can you Explain why this option is beter?
It would result in a lot of duplicate tables and measures, thereby enlarging my already large "golden"dataset. Sow why, please tel me why?

[Greg_Deckler · Accepted solution]

@BobKoenen Shouldn't increase anything. Basically start with your small data model. Then connect to it via DirectQuery for Power BI Datasets, add in the rest of your tables and now you have your large golden dataset. This is called chaining: DirectQuery for Power BI datasets and Azure Analysis Services (preview) | Microsoft Power BI Blog | ...

The issue is that probably the best you can do if you try to go the other way is to hide the "extra" data tables in your golden dataset with the baby dataset. DirectQuery for Power BI Datasets allows you to extend a model but you can't really delete the tables because you'd be deleting them from the original dataset.