Question

The credit card company, VISCARD, with 2 million customers, has designed a promotional offer and wants to communicate this to customers via direct mail so that they can enroll into the program. It costs $1 to mail a flyer, and there is only $30,000 budgeted to reach out to customers. You are assigned the task of identifying a cohort of 30,000 customers that are most likely to respond to the marketing promotion. You are given historic enrollment data with 100 variables. This is a classic binary classification problem: A customer enrolls or ignores the offer and does not enroll. After a few weeks, you have built a complicated model that turns out to be very accurate but does not give you any insight into which variables are associated with high enrollment probability. Should this model be used for this purpose? Are there any benefits in understanding how the model works, or is it just important to know that it works? Explain and justify your answer.

Answer 1

The problem is to identify the most likely 30,000 customers that will respond to the marketing promotion by the credit card company, VISCARD. The company has 2 million customers in total. A historical dataset consisting of 100 variables is provided which shows enrollment of customers. It is a simple binary classification with the classes 'ignore' or 'enroll'. If the customer enrolls in the promotion it is 'enroll' else 'ignore'.

The pros and cons of using complex model for modelling. When you make a complex model, it losses its explainability but accuracy of predictions might increase. Then there are simple models which are very explainable and the sometimes many have good prediction accuracies as well depending on the problem statement. Suppose a problem statement is to identify the voice i.e., voice recognition, then we explanation of model is not needed. When we are dealing with the above problem of credit cards it is important to have understanding of how model works. This helps in understanding which variables are important and how they are linked with each other. This will help business to take decisions easily on this basis. As it is always said 'simple is better', the same approach here is needed for the problem statement we are dealing with.