Article
Visual-level measures versus model measures
A visual-level measure lives on one chart. A model measure lives in the semantic model and can be reused. I keep time intelligence and shared calculations in the model so a review can find them and a second page does not rewrite them.
What goes wrong
A visual-level measure is easy to add in a hurry. The next page then has a slightly different version, and nobody can search the model for the name.
Handover assumes operations can explain a number. That is harder when the DAX is hidden on a visual.
What I do instead
The base measure and the time-intelligence set go in the model. Names stay consistent. BI Workbench formats the DAX and builds the usual YTD set so the paste is readable.
A visual-level measure is only acceptable when it is truly local to that chart and will never be asked for again.
Questions
Should DAX live in the model or on the visual?
Shared calculations belong in the semantic model so other pages can reuse them and a review can find them. Visual-level measures stay local to one chart.
How should model measures be written?
He keeps names consistent with the base measure and uses BI Workbench to format DAX and generate time-intelligence measures for paste into the model.