Hello Steve,
Masking from model can be helpful in the following cases:
1. In projects, where the object is scanned in several sessions, for example: top and bottom separately, after rotating the object itself. Then each sub-set of images can be processed individually in separate chunks up to the model (for example, in Medium quality), then everything not related to the object of interest (including the background, on which the object is standing) should be removed from the mesh in each chunk and the masks from the model are generated in each chunk. After that the chunks are merged (not caring about relative orientation, as it's only needed to bring the images and masks to the single chunk) and the processing is started from scratch for the complete dataset with the masks applied.
2. For projects, where some features are detected in the background or around the model, which are present even after mesh reconstruction. The mesh model for such project can be manually cleaned up (or with the help of Gradual Selection -> Connected Components filter), masks are generated from the clean model and the processing is re-started either from the very beginning or from the depth maps generation stage (to make them more clean and focused on the object of interest only).
3. Also this approach can be used even in common cases, when after building the mesh model and getting the acceptable quality you would like to try the processing with the masks applied in order to check, if it would give a better result, when the key and tie points are detected only on the surface of the object of interest.