Almost everyone is subject to advanced analytics whether they know it or not.  For avid music listeners this is especially true.  Applications such as Pandora, Spotify and ITunes use machine-learning algorithms, creatively named recommender systems, to predict which artists or songs that you may enjoy. The two main types of recommender systems in use today are collaborative filtering and content-based filtering.


Collaborative filtering relies on deciphering your similarity to other music consumers.  By understanding your preferences and the preferences of users similar to you, recommendations can be made which are most likely to appeal to you.  If you and another user have similar taste, expect to see suggestions based on what they are listening to. Most people have laughed at recommendations made by an application because they may be completely inconsistent with someone’s taste.  No model is a perfect reflection of reality. 

Content-based filtering does not leverage other consumers when making recommendations.  Instead, content-based filtering systems rely on comprehensive descriptions of products as well as a comprehensive profile of you.  This data is then used to output an offering that is likely to be appealing.

Hybrid recommender systems also exist.  They combine both collaborative and content-based filtering to output a prediction.

Whichever system is being used, one fact is true: the longer you use the application, the better the recommendations will be.  This, of course, is true of any machine learning or statistical technique.  If the method and model are valid, more data will make predictions more precise.  This is not only intuitive but mathematically verifiable (subject to some assumptions). Prolonged use will lead to much better recommendations.

How does this cross over into marketing? Retailers are using this all the time. Think about the last time you visited the Amazon or Best Buy websites. Your purchase history and the purchase history of others greatly contribute to the next most likely product. The same ideals apply to banks and insurance affinity marketers, as well.  Applying sophisticated algorithms or simply the studying the use habits of an individual user in comparison to other users makes the marketing experience seamless and much more personal.

So if you download a new music application, pay attention to the recommendations.  It is sometimes fun to try to trace the recent activity that may have triggered a recommendation.  Furthermore, watch the recommendations get better over time.  If you are asked to rate something or give feedback, do so knowing that you are actually building a profile that will feed an algorithm for recommendations.  Be honest if you want useful suggestions…or lie and see what kind of wacky suggestions you receive.  And keep in mind, these techniques are not just limited to music consumption.  Any time you spend online is most likely feeding a recommender system.  Some dislike this idea while others (like myself) enjoy the empowerment of controlling the content being consumed.