They’re right of course. Currently tags are low range and readers are expensive, meaning that it can be incredibly costly to get accurate coverage over a large space. That combined with the amount of information that each customer would generate and you’re starting to hit major data mining problems as well.
Having said all that, when has anything remained costly or difficult for more than a few years in the world of chips and data analysis? The minute you take the long view pretty much all of these arguments become meaningless. Nice try though.