Fails in internet advertising

For all the scary power of computers to track data, I find that focused advertising is a little...off.

I've read a few articles on this subject. (Most memorable are the ones written by Megan Mcardle...she has a gift for bringing together data, story, and presentation. Though those articles are more about the trials and travails of journalists trying to support their business through advertising, than about the economics of advertising in the digital world.)

In theory, computer networks and user-tracking should allow powerful predictive algorithms to show me ads for things I'm interested in. Or may need soon.

However, I find it much more likely that I'll see targeted ads for things that I've recently bought already.

This month, it's the rear sprocket for my motorcycle. I spent an evening digging up prices and options at two different online retailers, and made a purchase. During that same evening, I purchased a new chain and a new front-sprocket.

Every evening since then, I've seen ads for the rear sprocket that I already purchased. Except on Amazon, which is recommending other things I've bought within the past six months...or "books you may like", which are already on my bookshelf.

I wonder what is driving this behavior. The algorithm knows enough about me to know what I have gone shopping for. But it doesn't know enough about me to know what else I might be purchasing soon.

Or maybe the algorithm isn't capable of--or isn't even attempting to--predict my future shopping. So it tries to advertise to me things that I've shopped for in the recent past.