Tracking purchasing behavior online has been a hot topic for a long time. People have legitimate concerns about how websites track usage and behaviors and then use the data to display ads and suggestions tailored to a particular person’s interest. This has never bothered me as much as some people for one main reason. It seems that even the biggest sites aren’t particularly good at this yet.
Take Amazon, for example. I get a weekly email from Amazon titled “Book recommendations for Jamie Rubin” and they never seem to be current with their recommendations. Take a recent example. In the section of the email titled “New based on your author interests” there are two books suggested:
Amazon got at least one thing right: both of these authors interest me, and indeed, I am looking forward to reading both of these books. But that is where the system breaks down. Neither book has been released yet, and that is okay, too, since it is alerting me to books coming soon by authors I like. The problem is that I have already pre-ordered both of these book through Audible.
Audible is owned by Amazon. You’d think that with Amazon’s vast data resources, they’d reach into Audible’s database of purchases and see that I have already purchased these books, and perhaps, leave them off the list, knowing that I’ve already committed money and there is no need to convince me further. That they don’t do this surprises me and makes me skeptical that Amazon’s algorithms are all-knowing, or even particularly useful for that matter. Someone might argue that Amazon is offering me the hardcover, not the audio book, but from a content perspective, they are the same thing as far as I am concerned. Amazon should recognize this and find something else to suggest. It is almost as if their left hand doesn’t know what their right is doing.
Later in the same email is a section titled “Based on your reading” and there, Amazon recommends the following books:
In this case, I have read Nightmare’s & Draemscapes. It is book 836 on the list of books I’ve read since 1996. It is also listed on my Goodreads “read” list. As it turns out, Amazon also owns Goodreads. So why aren’t they tapping that as a resource to help improve recommendations? If they’d use that data, they would know that I have already read Nightmare’s & Dreamscapes and could have recommended something else instead.
I have not read The Closers by Michael Connelly and so at first glance, this seems like a good recommendation. But it isn’t. The most recent Michael Connelly I have read is Angel’s Flight, which is the 6th book in the Harry Bosch series. The Closers is #11 in the Bosch series. Given that Amazon has meta-data on series both on its store and in Goodreads, why are they skipping ahead 5 books in the series to make a recommendation? They should be using my data to determine (relatively easily) that the better recommendation is A Darkness More Than Night which is the 7th book in the Bosch series. Of course, I have already purchased that book, so Amazon would have to keep checking sources to see which one I haven’t yet read for a useful recommendation.
In the next section of the email, titled, “Books by authors you follow”, Amazon appears to give up entirely. They recommend the following books:
- The Perfectionists: How Precision Engineers Created the Modern World by Simon Winchester
- Basin and Range: Annals of the Former World, Book 1 by John McPhee
I read both of these books. The Perfectionists is book 759 on my list. And Basin and Range is 807 on my list. Both books have been purchased in Audible and both books have been marked as “read” in Goodreads. Amazon is just plain guessing at this point; they are not really making use of their vast resources of data.
I’d love to get great recommendations for possible reads from Amazon, recommendations that make use of everything they know about me. But I just don’t believe they have the ability to pull it off; not when they can’t even tap their own data resources to make a better set of choices. This is why I don’t worry much about sites that track my behavior to suggest purchases. If Amazon can’t get it right, I’m not worried about every other site out there.