Accidentally inferring. They were using basic machine learning to send coupons for predicted future purchases based on past purchases and general trends. And as far as I’m aware, it only happened once (or was only publicized once).

It almost certainly happened regularly since the entire point of the program was to make accurate predictions about future purchases. The practical impact of acting on those predictions only caused mild controversy and became publicized the one time (at least that I'm aware of) but we have no reason to expect that the program was unsuccessful and every reason to expect that it was.