I certainly do not suffer from "privacy blind spots". As a matter of fact, I've worked for one company in the telco space doing industry-leading work on setting standards about how to properly anonymize call detail records. And then, in this German telco that I mentioned, we were doing groundbreaking work setting up our data engineering in such a way as to comply with Germany's stringent privacy protections where consumers are concerned. I now work in another privacy-focused company.

But this one story I was telling:

Management: Please make a mathematical model that will predict if a radio unit has repeated and prolonged outages, using well over a hundred independent variables, ranging from technical information, to network topology, to information from the ticket system.

Me: Done. Mathematical model says: Radio units have problems if the people servicing them are lazy bums. Here is a system, based on natural language processing, that will predict people being lazy bums with huge accuracy.

Management: So who are the lazy bums?

Me: Can't tell you that. Because privacy.

I mean. Come on...

"Here is a system, based on natural language processing, that will predict people being lazy bums with huge accuracy."

What could go wrong... I was quite impressed until you got to this point too and thought I had been too harsh.

The patterns my system noticed were literally things like this: There's an error entered into the system by the NOC operator as an intermittent error on the station. Field technician drives there. Enters something like "Station on air upon arrival" into the ticket system. Does nothing else. (Like nothing to diagnose or fix the actual problem). Closes the incident. Drives home. Pockets the money for the time spent driving. A few days later (who would have thought that?) a new incident is opened for the same station. Repeat once a week for months and months.

I'm definitely not someone who is too naive to understand that technology like NLP has limits. But this was a perfectly appropriate usecase for NLP and it worked perfectly fine.

And people acting in this manner should not be immune from repercussions for reasons of data protection.

...also, it seems weird to me that workers representatives knew, before the project even started, that it was in their best interest to ask for this data to be anonymized. It was as if workers, collectively, knew the whole time that this sort of thing was systematically happening, and were actively working towards preventing management from finding out. That's what I mean by "not pulling on the same string".