Improving Measurement With Big Data

The data being used by managers is becoming increasing messy. Unstructured data lacks the nice organization of traditional data. Of course, the profusion of such unstructured data (text, videos, music) makes analysis complex but also brings considerable opportunities. Big data brings big headaches and big possibilities. We have some advice on improving measurement with big data.

Data On Radio Listening An Churn

Alina Nastasoiu, a former Ivey PhD student now at Booking.com, Mark Vandenbosch, then Ivey’s Acting Dean, and I looked at some data from Sirius the internet radio company. We wanted to predict churn, i.e. a consumer ending their subscription. What variables could help us with that goal?

Perhaps not too surprisingly, if you listen to the service a lot you are less likely to churn. We had data on how long people listened so we could check this. It is simply a better value for some consumers so they keep up with the service.

The amount of variety in listening is also associated with churn (as shown in prior studies). This is a bit trickier to operationalize. (Operationalize means turn the idea into a usable measure). Previous research used a simple measure of variety — just number of stations listened to.

Improving Measurement With Big Data: Creating A Measure Of Variety

We thought that we could do better than that. There are vast amounts of data on musical genres in the Million Song Dataset. (This is available at http://static.echonest.com/MusicMatrix/matrix.html). The dataset contains information on how different genres of music relate to each other. To do this we tracked how often descriptions were used together by listeners tagging the music. For example, we looked at how often the Heavy Metal tag was used at the same time as the Rock tag. This gave a measure of the genre’s similarity. Similarly, we looked at how often Heavy metal and Classical music co-occured. We used this as a measure of how similar these genres are etc….

Using this data we can then see how much variety a consumer experiences in their listening. The research can create a model of consumer variety seeking much more cheaply than using properitary sources (given the data is freely available). The research can create the model more quickly (it can be scraped from the internet). What is more the model can have less bias (no one is telling us what they think we want to hear) than asking paid subjects to assess the similarity of musical styles.

Predicting Variety In Models of Churn Using Big Data

The model we constructed, using the variety of music that a consumer listened to, led to better predictions of who would drop the service than models without. Armed with this knowledge a manager can test taking actions before consumers churn. They can test ways to re-engage the consumers who seem to be gaining less utility from the service.

The bigger point is that Big Data isn’t just something that can be studied itself. It can be a useful tool in studying the traditional questions marketers have always asked.

The positive point is that the same qualities that make Big Data hard to use (the sheer size and the way the data reflect numerous divergent opinions), make it an ideal tool to help analyse other data.

Nastasoiu, Bendle and Vandenbosch, (2018)

For more on Big Data see here, here and here.

Read: Alina Nastasoiu, Neil Bendle and Mark Vandenbosch (2018) Improving measurement with Big Data: Variety-seeking and survival, Applied Marketing Analytics, 4(3)