The Knowledge Exchange / Business Analytics / Tips for solving a new analytical problem

Tips for solving a new analytical problem

Blog series shows how to develop an advanced analytics strategy

Speed to decision makingIn the beloved classic The Hitchhiker’s Guide to the Galaxy, we learned that the answer of life, the universe and everything is 42. Of course, without knowing the question, the answer is useless. Some dedicated readers have suggested the question could’ve been: How many roads must a man walk down? How many good deeds must a person do? We’ll never know. These suggestions highlight the importance of framing the right question and picking the right problem, because without them, the answers are meaningless.

When you approach a new analytic problem, keep this in mind as you try to determine the phrasing of your problem. Often, the question will need to be rephrased multiple times before the team arrives at the exact problem that they are trying to solve.

For example, entertaining a series of different questions can help tease out the purpose of the project:

  • Do I want to go after cases that had the highest propensity of fraud?
  • Or cases where I have the highest likelihood to recover monies?
  • Or where I can maximize my hourly return rate on fraudulent cases investigate?

Each of these problems will require a different target variable and would likely give you a very different predictive model.

Another pitfall is trying to frame the project based on a particular modeling technique. I have heard “I want to cluster my customers” far too many times. When I ask, “Why do you want to cluster your customers?” I sometimes I start to get answers like: “to find ones that I can cross sell to,” “to find high value ones,” or “to look for different shopping behaviors”! Now these are problems that I can start to solve!

Having a very clear and precise problem can make your approach more focused. I once worked with a small donor organization that wanted to reduce the costs of their mailings. Rather than mailing to everyone, they wanted to drop the bottom 20 percent of potential donors. Of course, this gave me a very different model than if I only wanted to mail to the top 20 percent of donors. I needed to construct my model evaluation criteria so that I found the best result for the problem at hand.

If you’re struggling to refine a problem of your own, one guideline that’s always helped me is Occam’s razor. It states “when faced with competing hypotheses, select the one that makes the fewest assumptions.” In layman’s terms, it’s saying that the simplest answer is usually the best. So looking back at our Hitchhiker’s problem, we might be forced to abandon some of the provided queries, and come to the conclusion that the best question for our answer of 42 is simply: What is 6 x 7?

The next post on the Myths and Realities of Successful Analytics series will give a checklist of questions to ask before building your analytical model. Until then, download this research from MIT Sloan Management Review to see how several companies including eBay, AstraZeneca and Caesars Entertainment approached innovative analytics efforts.

  • Facebook
  • Twitter
  • Digg
  • LinkedIn
  • email

Post a Comment

Your email is never published nor shared. Required fields are marked *


You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>