How to define the signal box for signal significance / FOM?

This seems to be a basic question but I cannot get a good explanation anywhere and I’d really appreciate it if you can help.

After doing a TMVA classification training, I need to decide on a cut that maximizes FOM = S/sqrt(S+B). How should I define the range of signal fitting region (eg fitting on mass) in order to calculate expected number of signal and background that falls within it? Is there any theoretical reason that I should use, say, 3 sigma width as oppose to 2 or 4 sigma around the signal peak? Wouldn’t it yield different optimal FOM cuts if different signal regions are used?