When recommendation engines go terribly wrong

When recommendation engines go terribly wrong

One of the most popular forms of personalization used on eCommerce sites is the “people who bought this, also bought that” recommendation. The assumption is that people have similar tastes and so if other people are buying the same things, so should you. Sometimes, the intention is to ensure that someone who buys an item requiring additional accessories be reminded of all that is needed before checking out their shopping cart.

Some sites use an algorithm to predict which products should be paired with each other. Others simply tag products and bucket them into groups. Now the latter can same time and money, but sometimes leads to some strange recommendations as seen below on the Tiger Direct website.

I was shopping for some Bluetooth wireless speakers for my wife to use with my iPad. My initial search came up with a two-page list of possibilities, ranked by relevance. I clicked on the first search result and immediately found what I was looking for. Good news! But then I noticed something very strange, something very wrong. All the cross-sell recommendations were for speaker wire were for speaker wire! Oh wait! Maybe this isn’t really wireless? Check the description: Yup… wireless! See for yourself.

Obviously someone decided that speaker wire was an automatic cross-sell recommendation for anything in the speakers category. And when wireless speakers were introduced into the category, nobody thought to make the change or create an exception. This example just goes to show you that unless you do personalization and recommendations the right way, it can backfire badly. Oh, by the way, they also offer an upgrade from the D5 to the… D5 for $0.00. Now that’s a bargain. LOL

About the Author