Its a good question you have. The main issue is there are so many data sources available and the way the data is structured and the purpose of what you need impacts what data sets you need to build.
My advice would be to go a training course where you can sit in front of computer with people in a similar scenario. You can only learn so much from Videos and Data modelling theory is important enough to invest the money to learn. I am going to attend Marco Russo & Alberto Ferrari 3 day workshop in August this year.
In saying that, Power BI is an analysis tool ONLY and holding vasts amounts of data which is not required is not recommended. Power BI likes tall data tables …and not wide ones if possible…its a columnar data base so it can really calculate super fast…
The main tip is to workout your FACT and DIMENSION tables and try and set them with only required columns for visuals, every other column can be excluded from your query. Try and make a star schema in most cases is best approach with your Fact and Dimension tables.
Sam has some good content on this part of the approach.
Design also takes into account what kind of data you are accessing and how you need to ensure it reconciles to the master data set. (Source)
For example, reporting on accounting data like in a profit and loss or balance sheet requires the General Journal Fact Tables from the source data. You need to ensure that your figures balance to the source and have some kind of system to agree Power BI to the source accounting system holding the fact data.
You may also want to get Dimension Tables for things like Accounts & Jobs and those Dimension tables have supporting tables too which you might need to break up and group items.
It takes a long time to develop a technique that works well and fits a purpose. I have built this experience up over time and doing workshops and readings. I am an accountant and worked a lot with software so Power BI is a natural extension of what I already knew.
A lot of people can show off nice reports and dashboards but the theory of how the data is obtained and reconciled is paramount for a proper solution you can manage or walk away from so your client can control.
Its a worthwhile topic , “How to setup your data set so you can reconcile what I call control records”.
I am happy to develop some future content, if Sam see’s the need.
Cheers and Good Luck!