Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

How can AB testing work without storing personally identifiable information?


AB testing doesn't require personally identifiable information.


How do you bucket users into a control and enabled group without having some way of identifying them? (usually ip address)


You could for example bucket them based on the last digit of their IP address.

So, everyone whose IP address ends in 0, 1, 2, 3 or 4 gets the new version, everyone else gets the old version.

You don't need to store the IP address for that, you just need a rule that decides which version to serve as the user requests it.


Why are you bucketing users? If you have some sort of current session for their presence on the site, flip a coin and assign them your A or B. Store nonpersonalized info about their experience on the site associated with the given A or B.


Generate stand-alone, opaque identifiers for whatever sessions you want to analyse. Make sure these tags are ephemeral and decoupled from anything else.

Then, store only the flow data, discard everything else. Or if you need to keep some data around for the test duration, delete all of it once your A/B test has concluded.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: