This video is very informative, it explains perfectly why marketing videos should not be published without the permission of an engineer.
First, 'analog' is completely irrelevant to the consumer, because the switches only have two positions: on and off. If you want to use an ADC to demux all the keys you put on one input pin, that's your business, but it has zero inherent advantage over a digital read, because the signal is discrete.
Second, The datasheet for those switches says they bounce for up to 5ms at 16in/s actuation speed. Essentially, after you first register a keypress, you have to wait 5ms before you know if it's a press and hold or a press and release. Now, you might have several bounces within that 5ms, so if you take instantaneous reads you have a maximum latency of 5ms, but if you attach an interrupt and listen continuously, you might only have to wait for tens to hundreds of microseconds(depends on the switch) to know for sure the switch isn't bouncing. If you do hardware debouncing, you can make the pin's value persist through however long the bounce takes, cut the maximum latency down by whatever the longest bounce is for that particular switch, and eliminate the need for interrupts. As for NKRO, you can do that with a digital controller as well, it's just a question of the board being designed for it, and the microcontroller having enough IO pins for your chosen method.
tl;dr: you can engineer the latency under 1ms with NKRO without an analog to digital converter. Marketing is just trying to talk up their product by comparing it to garbage keyboards you can buy for $5.
Unfortunately I don't know any reviewers that actually test the latency of keyboards. Not too complicated but it does require getting access to the keyboard's PCB in order to snoop or simulate a key press. As for required test hardware, an arduino or raspberry pi would work, so it's not particularly expensive.