Then why does it take the sales person at Macy's 15 minutes to look up my order before giving up and just giving me a refund for a poorly fitting sweater I brought back?
This kind of dichotomy exists in just about any organization that gets into BI/data science/whatever you want to call it, and it can be absolutely maddening. The state of ETLs and analytics systems, for example, means that we have basically gone back to "overnight batch processing" and a lot of reporting data only becomes available the day after the transaction, since all the ETLs need to run and then whatever analytics system, data-cube assembly or etc. But since the analytics system exists, it is taken as good enough and no one wants to re-implement the functionality in an online, system-of-record fashion.
Or another way to put it: the GP says "The second you walk into the store based on your face they can pull up your entire purchase history and contact." That's being pretty generous... most of the time they don't know any of that until an hour or a day later, because it comes from multiple disparate systems (and outside vendors) and importing that data into a data warehouse (in batches) and then correlating it later is far easier to implement, and the dominant method used in this kind of analytics.
(route-optimizing here likely has the intention of boosting sales, not necessarily reducing customer shopping time)
I suspect the real reason that the panopticon is kept separate from the customer is to avoid creeping people out.
I get that its considered a feature to get them into a store, but surely you don't want to optimise for associates not selling things?
That would be true if the customer service area was isolated inside an empty boring room, but at Macy's, it's out in the open on the store floor and surrounded by merchandise. Waiting in a boring line means eyes that wander to adjacent displays and stands.