A better way to think of it: if the system only has a 0.1% chance of an incorrect match and at least 700 people regularly shop there, each of them is more likely than not to be misidentified as somebody else.