There are many things that I enjoy in life, but cooking and analytics are two of the things that excite me the most. In my free time, one of my favorite things to do is experiment in the kitchen to try to come up with something new and different to eat.
One of my favorite ways to innovate in the kitchen is to bring culinary techniques from different cultures to bear on my traditional, southern style of cooking. What happens when I season my pot roast with some Italian flavors or spice up my macaroni and cheese with a little Mexican flair? My cooking technique may not change, but by bringing together flavors from different cultures, I can create lots of innovative new dishes in my kitchen.
Believe it or not, innovation in the analytics space often happens in much the same way. Innovation is not always about coming up with a brand new analytical technique. Sometimes, it’s about taking a technique that’s been around for a while and applying it in an unexpected way.
For example, survival analysis provides methods for dealing with “time to event” data – also referred to as censored data. This technique has been used for decades in the analysis of clinical trials data where the time until a particular outcome occurs is of critical interest. With SAS 9.3, we’ve taken traditional predictive modeling techniques, sprinkled in a little bit of survival analysis, and provided our users with a new way of analyzing customer data that we call “survival data mining.”
You might be asking yourself, “How is survival data mining different from traditional survival analysis?” The first difference can be found in the data. Historically, survival analysis methods have been applied to relatively small data sets containing a well-defined set of variables that were collected as part of a designed study. In contrast, data miners typically have data on hundreds of thousands of customers with a wealth of potential explanatory variables available for modeling.
Survival data mining also differs from survival analysis in that it focuses only on discrete time intervals (e.g., day, week, month,…). And, in survival data mining applications, the data is often both left- and right-censored. That is, the data may be truncated on either end of the time scale as customers are continuously being added to and deleted from the data base. However, the two techniques are similar in that survival data mining allows you to ask questions like “What is the average survival (lifetime with the company) of customers during a particular time period?” - questions which are typical of traditional survival analysis. Both techniques also enable you to determine how other variables affect that outcome.
One of the most common applications of survival data mining is to the problem of customer churn. Historically, predictive modeling techniques have enabled you to predict which customers are likely to churn. However, those techniques don’t provide you with any insight into when the customer is likely to churn. Will it happen next month, or is it more likely to be a year from now? Having that information is critical in being able to determine how to treat the customer in order to try to prevent the churn from happening. You need that information about the time until the event happens in order to get a full picture of what’s happening in your customer base.
Survival data mining brings techniques together that enable you to determine not only which customers are likely to churn but also when they are likely to churn and which variables are influencing that outcome. And, the great news is that we’ve taken this functionality and packaged it up in a single node in SAS Enterprise Miner 7.1, making it easier than ever to try these techniques on your own customer data!
So, the next time you’re faced with a difficult problem, take a few minutes to think of analytical techniques that are used outside of your industry. Think about how the characteristics of those problems are similar to the characteristics of the problem you’re trying to solve. Then, mix those techniques up and see what happens. You might be surprised at what you can cook up!
What are some of the innovative ideas that you are using to analyze your customer data?