Machine Learning for Marketing explained

24. November 2015 Big Data 0
Machine Learning for Marketing explained

Machine Learning ThumbnailMachine Learning as a term can sound alien and slightly difficult to grasp. The first thing actually coming to my mind talking of machine learning is something like Skynet from the terminator movies. The concept migh also sound like covering something very complecated, while it does not necessarily have to be.

I have bumped into a Harvard Business Review article recently which takes a great, and visual, approach to explaining machine learning. They break it down to be a decision about to people in a pretty usual business case study.

In the situation set up we are representing a cable TV service provider and we are to decide who is to be targetted with a proactive retention campaign and who does not. Basically deciding who has the higher probability of becoming a “cord-cutter”, Karl or Cathy.

It is also important to note that in the case study they bring in the perspective what I believe is the real reason behind deploying machine learning for marketing, and that is saving money. It’s not purely better targeting, but rather the fact that you have limitations of how much discount you can give out to people and besides budget control you also want to maximize the impact.

The decision tree methodology used I think is very common, plus it makes it visible that machine learning does not have to be a “black box”. It is nothing magical or complicated, but a lot of the statistical results can be explained by common sense. So here we go with the specific case

Machine Learning 1






the question is, do we really need then machine learning based on this example? Well, try to decide which attributes to use, in which order and from where do you consider differences significant. Also try making the decision on how many segments, how many attributes to have and go through on the manual experimentation process of combining all these possible outcomes. See you in 10 years from now 🙂

The speed, simplicity and scalability of such methods is undeniable. Plus you really have a chance to combine intuition with data and cross validate your hypothesis. Oh, and I did not yet mention a process of gradually improving modells and methods. Anyways, I think I made my point.

Care to add more? Let me know in the comments! (many thanks for them!)

Leave a Reply