It depends on your routing.
For instance let’s say you have a piano instrument which has its own audio channel. The piano output you route to a group named KEYS. You use a send from the piano channel to send to a Reverb FX channel. The output of the Reverb FX channel you route to the group named KEYS. In this instance, when you export by groups you will get the piano and the reverb in one exported stem.
If however you route the output of the Reverb FX channel to a different group, REVERB, then what is sent to the Reverb FX channel will be split from the piano. So then if you do a batch export of groups KEYS and REVERB you will get one stem with the dry piano and one stem with just the reverb channel element, not the dry piano channel element.
Two rather simplistic examples, I know.
But I do this all the time and for me it is all about careful planning of the routing and what I want separate export control over, and it has always been a trade-off between flexibility and practicality. I have tended to keep things simple and create a one-to-one mapping between main groups and the stems I want to export. (Of course, sometimes it’s helpful to sub-group stuff before routing to a main group that maps to a stem you want to export.)
This has saved me MANY hours of laborious exporting.
Was that what you meant, or have I missed the point?