Semantic Model Advice

[Newcolator]

Hi. 

I'm looking for some general advice around creating semantic models in the Power BI service for report builders to connect to in desktop to build reports. I work for a large company with lots of different departments and areas and the data I work with comes in many different grains.

I don't think I should be building one giant semantic model that covers everything as that would be confusing and inefficient. I want to create a small number of models that service as many reports as possibe.

I had an idea to several create models that don't duplicate columns and all have different grains, that can be loaded into desktop and joined together. For example, I could create one model at an orders levels, detailing the customer order, dates and total sales. Then I could create a second model which contains the detail product order line information. A report may just require the first model, but another report may need both. So the report builders could connect to just the models they need.

Having tried this, I'm not sure it's a very good idea! Connecting to more than one model makes the combined model quite messy and complex.

I guess my question is, are there any guidelines on how to decide which models to make and how much information should be in each, and what to do about reports that require a lot of information from potentially different models.

I hope that makes sense. 

Thanks! 

[lbendlin · Accepted solution]

There are guidelines, but they are all very fluffy and sometimes contradictory.  The answer is a resounding "It depends".  Company size, number of data subjects (tens to thousands), maturity of your company's Chief Data Officer's team, availability of resources for maintenance etc etc. 

You need to figure that out for your specific situation and boundaries.  In general keep in mind that your job is to facilitate insights and decisions, not to prevent people from using tools or accessing data.