In a response model, just because something is positively correlated doesn’t necessarily mean that people with those attributes will even be mailed.  Because of the complexity of the interactions, other variables on that same individual could cause them to drop down as an ideal candidate.  Also, there is a chance that all the records with a slight positive correlation on one particular variable could all fall below the score threshold for the mailing.  Pre-selections are actually much more discriminatory as they result in an absolute; no matter what the person or household’s other attributes indicate, they can be excluded based on just the single variable.  Because of this, always use pre-selects (or parameters) with great care and consideration.

As mentioned in yesterday’s blog, we are using insurance policy mailings as our example. For example, a hospital accident policy is a product that is designed to have the broadest appeal with people of average financial means during their active earning years.  More affluent households typically have stronger insurance coverage, more savings and disposable income to deal with emergency expenses, and more reserves to handle emergency medical care and loss of income.  It is a product that was developed to meet the needs of people with an income of modest to average financial health often living paycheck to paycheck (which is projected to be a significant portion of the U.S. population).

Because of the very nature and purpose of the product, income will be a factor in targeting and marketing this product whether modeling is used or not.  The model is entrusted with finding the group with slightly below average but not rock bottom income and financial health.  It is also designed to meet the needs of wage earners, which often fall in certain age groups.  Younger consumers are not likely to want or need the product.

So, as you can see, utilizing a single variable exclusively is not likely to improve your targeting sufficiently to produce your desired results. One should take into account a myriad of factors such as age, household and disposable income, mortgage rates, credit scores, households per square mile, population 25+ with a college degree… and the list goes on. The more variables you include in your modeling process, the more likely you are to ultimately be able to predict the desired behavior.  When you have refined your model to include just the most predictive of those variables, the elements work in synchronicity with one another, allowing you to select the optimal targets from within your prospect universe. It’s not easy, but when you are cultivating millions of opportunities into the thousands, the more fine-tuned your model is, the greater your ROI at the end of the day.