CAD software knows everything about your parts...(assuming you tell the things it doesn't know...as mentioned, material properties) After doing this, calculating multibody C.O.G or even inertias of complex bodies is 'trivial' (of course, it wasn't for the programmers).
These things can be very accurate, or can be made very accurate, however they can and usually will be off by a certain tolerance. How much one is willing to accept for an error is up to the user.
They can also be wrong simply because something as trivial as the electronics and wires within your body weren't modelled correctly, or given the correct initial data, on top of that, no material is perfectly homogenous, which CAD software always assumes. This may not even account for even 1% error and not matter for something small, but is extremely important when dealing with large heavy loads.
Because of this, it's almost always a good idea to experimentally verify such data yourself, using trifilar pendulums as an example or other equipment and methods
I generally use CAD data as a good lower boundary for initial guesses with empirical verification.
Of course, this again will only be as accurate as your measurements and is an entire branch on it's own.