But clearly if you'd get the mail the inference would have a statistically higher chance of being incorrect and they are optimizing to minimize wrong inferences? For me checking if a child's fare has been bought in the past to target mailing is more problematic than wrong assumption on sender's behalf. I don't mean to offend anybody, but however poor the assumption in the subject is, it likely is statistically significant enough to target mailing. I'm quite certain it is possible to figure out based on the data about your online actions that you and your wife would feel offended by the assumption you have kids and you should not get these kinds of emails, but are you sure you prefer companies to dig that deep into your life in order to avoid that?