>There also can be hybrid implementations if data preparation steps involved are too time consuming for total data pool. For instance a monthly time series analysis might
>calculate current partial monthly data
>read already calculated monthly pool data
>calculate time series data back from current start position - which might be needed every time, since drop of terminal value is not distributed even or current events will create different factors if trying to estimate base values of empirical data.
>
That's true, you can have hybrid approaches. Maybe 95% of the data is something you refresh daily, but 5% changes so frequently that you want to query it in real time.

You have have a Power BI report with 5 datasets, where four are based on a published model that's updated daily (or twice a day, or whatever), and one dataset that fires a query back to the original source. Now, to do that, you need to install a gateway back to the data source. This can *usually* work, but there are some nuances and oddities with it that some people experience.

For that reason, most places using Power BI will only take the hybrid approach if there is a compelling reason. There's nothing wrong with the idea and I'm glad you brought it up - but truthfully, knowing what I know about Power BI, I will try to do everything to steer people away from that - more because of the implementation than the idea itself.