Why does the subject knowing about it make it any less useful for genocidal purposes?

You pretty much need a state ID or drivers license, which lists your skin color. Political contributions are a matter of public record for anticurruption purposes. Vital records offices are more than sufficient for tracing ancestry. Welfare offices know who is poor and medically needy. Telecoms and the post office knew who you communicated with.

There was more than enough information to mount a holocaust on any of those axes long before the current crop of tech companies.

Different, though: Now its all in one place, stored forever, and in much more detail: your cell phone leaks your location continuously and in real-time, your photos and videos provide detail we are only just learning to harvest, your formerly secret or at least hard to track written communication is now trivial to record and analyze, and with the coming of IoT all this is just getting a bit more minuscule. Future governments seeking for unwanted persons will have an unprecedented dataset to profile whomever they might not agree with.

Again, this worst case scenario may not happen, but the fact that it can, even if only with a small chance, makes this massive data collection totally irresponsible.

Skin colour, religious and sexual preferences ... One can only hope the next dictator shares your lack of imagination.

Also it's a bit of a bizarre argument that tech companies are doing data collection on such a massively intrusive scale, we should really look at the government databases first because they're already more than enough to screw us over with? It's not even a "but they're doing it more!" ...