[dangrossman]

Campaign A: Show 10,000 random people ads for espresso grinders.

Campaign B: Show 10,000 people who recently looked at espresso grinders some ads for them.

95% of the first group didn't know espresso grinders were a thing people bought. The remainder probably aren't actively shopping for one.

In the second group, some portion of the audience has already bought an espresso grinder, or decided they don't want one right now after all. But it's also guaranteed to include thousands of people actively shopping for one right now.

Which campaign do you think is going to result in more espresso grinders sold? The one that doesn't include you, or the one that does?

[oneeyedpigeon]

What about campaign c: show 9,900 people who recently looked at espresso grinders an ad for one, and 100 people who actually bought one an ad for something related that they might need. If you're harvesting that kind of data anyway, at least use it!

As a matter of interest, is someone who has previously purchased an espresso grinder more or less likely to buy one than someone who hasn't?

[ryandvm]

Of course you're correct, but the problem is Campaign C requires a far higher invasion of privacy. Now, not only do you want the advertisers to track the sites you visit - you're also proposing they somehow get information about everything you purchase?! No thanks.

[semi-extrinsic]

It wouldn't be that hard to handle without any further privacy invasion: online shop waits O(12) hours before reporting to AdSense, and if visitor bought the product within that time, reports other, related product instead.

(If the ads in the first 12 hours are important, it makes little sense that they keep advertising for a whole month.)