Teo Lachev explains why you want a datamart (or a data warehouse) for BI solutions:
I sent a proposal for implementing a classic BI solution: Azure SQL-based datamart (not Power BI datamart please), ETL, semantic model, and reports. The client had a sticker shock. Return to sender … as other BI companies that quoted can do it for half! Upon digging, it turned out the other companies would build the semantic model (aka Power BI dataset) directly on top of the data source. On a T&M basis, of course, what else? By contrast, I give fixed-price milestone-driven proposals and I don’t get paid unless I deliver and meet written and agreed upon success criteria, but that’s a different story.
So, let me count the ways as the poet would say. It’s certainly technically possible to slap a dataset on top of the data source(s). That’s what self-service BI is all about right … until it doesn’t serve anymore
Read on for more detail.