This week we are very excited to share a guest article shared by our Jigsaw-Bocconi EPBA (Executive Programme in Business Analytics) student Abhiskeh Mittal. In his article, Abhishek speaks about how to create a successful marketing analytics model. Go ahead read the article here..
Originally posted on LinkedIn pulse
So you have come up with your marketing analytics response model (Response models are used to improve response rates by identifying prospects who are more likely to respond to a direct solicitation) using different data mining techniques and validated it with various techniques such as ROC, Confusion Matric etc. and you are ready to go to production. Fair enough, you can go ahead and deploy your model in production. Let’s suppose after you have deployed your model in production, your Boss comes to you and asks – “Tell me how effective your Model is”. You better have an answer to the most logical question.
The value of measurement and continuous improvement is widely acknowledged, and yet most likely to be overlooked. Marketing tests are an important part of analytic marketing, as is data mining. The two often complement each other, and marketing tests are an important part of understanding whether data mining efforts are working. Typically, two things should be tested when using data mining for a marketing treatment. First, is the marketing message working? Second, is the data mining modeling working? So how do you do it? The key is to use holdout groups intelligently factors. In practice, four potential groups exist:
Target Group: Receives the treatment and has model scores indicating response.
Control Group: Receives the treatment and is chosen either at random or based on lower model scores.
Holdout Group: Does not receive the treatment and is chosen either at random or based on lower model scores.
Modeled Holdout Group: Does not receive the treatment and has model scores indicating response.
These four groups are indicated in the following figure:
The responses from these four groups then provide useful information. Using these groups for modeling is called incremental response modelling. The goal of incremental response modeling is to reach prospects that are more likely to make purchases because of having been contacted. Comparing the response rates of above mentioned four different groups will clearly tell you how effective your model is.
Abhishek Mittal has over 15 years of experience in designing and developing enterprise applications using technologies such as Big Data, Analytics, Solr & Elastic Search, Cloud Azure and AWS, .Net, IOT and Mobile. Bitten by the analytics bug, he has enrolled in the Jigsaw-Bocconi EPBA (Executive Programme in Business Analytics) course and is looking to use analytics to solve real-world problems