At the end of the article he elaborates the problem in a way he seems to feel is too complex to mention in the interview itself: what if there's a third day?
If there are three server days of logs and a loyal customer continues to be defined as one who visits on two different days, the timezone problem essentially goes away. More fully:
- if a loyal customer is one who visits regularly;
- and we sample several days of visits;
- then loyal customers will be detected regardless of their timezone.
If you want the concept of a "loyal customer" to match the detection threshold, the weirdness related to timezones will still exist, but if you think of the detection threshold as a tool that's good enough to detect loyal customers, then it won't.
