Could just be the date of install in UTC as a prefix, the other part randomized on the first of the month... they could still calculate relative drop off, and still get better stats more anonymously.