I once saw a Softimage video where they took a fully animated human model and placed it overlapping a completely different model that was not animated at all. Then, using some special tools within the app, they baked the animations from the fully animated model into the non-animated model. From what I remember, the non-animated model did not have a skeleton set up or any information for which vertices were attached to which bones, etc. So the non-animated model got all of its information from the fully animated model and, despite the differences in the models, it worked out fairly well. This meant that new models could have animations baked into them and then tweaked if needed, thus speeding up the process.

I am not an animator, so I might have described this incorrectly.