Ask HN: Privacy Oriented Cohort Analysis?

To do cohort analysis or calculating churn (as a SaaS startup) one needs to collect some kind of id over time. There are clever ways to calculate daily (weekly) uniques by salting a hash with a daily (weekly) random salt. How could we do cohort analysis with the least amount of information? The only thing to slightly increase privacy and make de-hashing more difficult is with a secret salt. Is there some clever way with Bloom filters?