Music is inherently and intrinsically personal, I have a very eclectic taste in music, I’m a butterfly floating across genres depending on my mood, the weather, music in a film I just watched etc.
The other day I listened to some classic country music (yes I know, my bad) on Amazon Music. The next day I listened to some alternative rock but Amazon was still recommending country. My county music fad had passed, that was yesterday and if I’m honest I was a little bit disappointed that Amazon was out of kilter with what I wanted.
No doubt Amazon were using some form of machine learning affinity analytics to work out what music I liked and then predict what else I might like. The thing is that affinity analytics doesn’t always get it right especially for people with varied tastes. Hyper personalisation will not always deliver relevance. No matter how clever the data algorithms get one of the problems with affinity analytics is that it is often using a rear-view mirror. Looking into the past is not always a good prediction of the future.
Big data analysis like this will get it right most of the time but the occasions that it gets it wrong can create a really big disappointment. Customers now are used to having recommendations and are almost not noticing when a brand gets it right but they will definitely notice when it gets it wrong.
Using a combination of data sources will help you understand your customers better. You need to develop a full, rich picture based not just on your brand behavioural and transaction data but also what your customers do when they are not interacting with your brand, what other brands they like, what interests them, and what type of person they are.
If Amazon had worked out that when it comes to music I’m a butterfy, they may have looked at the data differently.