You will find that the most probable (the mode of the distribution) thing to buy in transaction n+1 is either a substitute for, a complement of, or identical to transaction n.
i think no one would disagree that there are things they buy very frequently, almost on a schedule. (then again, with these items people are probably very accustomed to buying the same brand of milk every week, so ads don't seem reasonable here as well).
Ads work at human psychology; they're not fully logical. Though I'm sure there's inefficiency in marketing but if post-purchase ads weren't ROI positive, I doubt the market would be paying for them.
In a sibling comment I replied to someone with an objection that it doesn’t apply to cars… with a personal anecdote that it does, at least sometimes, apply to even cars.
Even governments do it: Having previous purchased a nuclear submarine is highly predictive of future nuclear submarine purchases. Likewise, you probably have not purchased a nuclear submarine, which makes me think you probably won’t buy one this year. ;-)