You have to find the correct substitution. But I am quite sure that the result will be not very reliable. For testing I in you position would generate a theoretical pattern. Then load this again as "experimental" and try to solve it. However, since you don't know preferred orientation, profile function, change of the profile function with 2theta, vacancies (changing occupation), noise and so on, you should set free all other parameters (or at least change them slightly). I am sure you will realize that you will never end up in the substitution you entered. Theoretically it should work but i don't guess it will be. Nevertheless, do it since only then you will realize how speculative Rietveld analysis can be, and it perhaps will give you n impression how much you can trust data extracted from powder pattern. There is a reason why cystal structures are preferably solved with the help of single crystals.      

As Gert points out, this is a difficult proposition.   Your best chance would be if you actually got clear domains as opposed to  a random distribution.  You might be able to get a better answer if you have access to various profiling forms of surface spectroscopy such as microprobe analysis.  Then, you can sample various domains in the sample and see what their elemental composition is.

Thanks Dr. Gert Nolze and Dr. Joseph S Merola for suggesting me