You can use the mesh’s AABB to approximate the size so that it always hits the mesh [...] But I think the better approach is to fiddle with the decals in a way that the end result is satisfying.
I've been trying this approach now for a while (at least if i understood you correctly). Though im approximating the size of the decal by finding the location where the raycast for the bullet exits the collider instead of using a bounding box. Currently struggling a bit with sizing and positioning the decal correctly for it to show up properly😅. But im also starting to think this approach isn't going to be satisfying when i do get it to work.
The problem is that i am not sure along which direction to project the decal. If the decal projects along the path of the bullet (which would make the most physical sense i think) it can end up incredibly stretched when it encounters the mesh at a very shallow angle. Eg:

So i'm back to the original problem of not being able to determine where/which face of the "visual" mesh the bullet would hit and what its normal is, at least not without making a lot of unnecessary colliders that match it more closely. And if i use the one normal i do have access to (the one from the collider) i cannot guarantee that it is similar enough to the normal of the mesh face the decal will be projected onto to not be stretched anyways.