|
YOU ARE SITTING ON A GOLD MINE. Really! It is almost certain that there are "nuggets of gold" hidden in the customer and sales data that you already have. The challenge, of course, is to find them and take them to the bank. "Predictive modeling" is a sophisticated statistical technique that can help you do this. |
WHAT IS A PREDICTIVE MODEL?
A model is a kind of map that shows you the "big picture" of relationships. A descriptive model describes your customers as they are today and shows you "where you are." A predictive model goes further. It shows you how your customers are likely to behavior in the future. This tells you how to get where you want to go from where you are today.
Predictive modeling tools presume that the best way to predict future behavior is to study past behavior. This works because customers generally tend to be very habitual in their behaviors. Customers who spend a lot with you will tend to continue to spend a lot with you. Customers who buy from your frequently will tend to continue to buy from you frequently. Customers who spend a certain average amount for each of their purchases from you will tend to continue to spend about that amount for each purchase. This kind of list can continue indefinitely, of course.
Based upon past behavior, predictive modeling tools anticipate how each customer segment is likely to behave in the future. This "before it happens" knowledge gives you the abilitiy to focus your marketing efforts upon those customers who are most likely to respond. Higher campaign success, sales and profits are the direct result.
Because predictive modeling tools are based upon past behavior, it is essential to start with accurate customer, response, sales, and margin data. Predictive modeling tools are seriously disrupted by "dirty data." (See Data Hygiene - Database Bullets to Dodge for more about this.)
WHAT IS A PREDICTIVE MODEL GOOD FOR?
Predictive modeling tools answer these tough marketing questions:
- Who are the best targets for my promotions?
- How many prospects exist that “look like” my best customers?
- Which of my existing “smaller” buyers are most likely to become “bigger” buyers?
- Which prospects are most likely to respond?
- Which customers are most likely to buy?
HOW DOES PREDICTIVE MODELING WORK?
Predictive modeling uses already-existing data about your customers' past behavior. Opinions, polls, surveys and other subjective information is not a part of this kind of analysis.
We start by analyzing the characteristics of your "best" customers to define a “model”. These characteristics include purchase behavior, geographic location and demographic characteristics among others. Then, the characteristics of this model are used to identify and select existing customers who are candidates for becoming "best" customers. These same characteristics can also be used to find and qualify prospective customers from names on an outside list.
Predictive Analysis can use a variety of statistical tools. These include regression analysis, CHAID (Chi-Square Automatic Interaction Detector) and C&RT (Classification & Regression Tree) among others. A regression analysis applies a score reflecting the predicted likelihood (or degree of statistical conformance) to each record. CHAID and C&RT segment each record into mutually exclusive segments. Past this point, predictive modeling starts to get complicated!
WHAT'S THE BOTTOM LINE?
Predictive modeling tools give marketers the ability to predict which customer segments will respond to offers, promotions and sales incentives. These insights guide the design of narrowly-targeted campaigns that enjoy high success with minimal invstments. However, predictive modeling tools are complex. They require lots of clean data and are more expensive to prepare than less sophisticated techniques.
Do you need to use predictive modeling to achieve your marketing goals? Perhaps not. Here is how to know. Start with a simple “cross-tab” profile analysis. Even such simple analysis provides substantial benefits compared to doing no analysis at all. Then, proceed to more sophisticated tools as justified by your situation. The larger your marketing budget, the more you should expect to invest in making the best possible decisions about your offer, your media and your targets. In short, start simply and stop when the costs outweigh the benefits.
Finally, predictive modeling is both a science and an art. Experience is required to prepare, use and interpret these sophisticated tools. You will probably discover that an experienced database and modeling expert is needed to set up and interpret predictive modeling tools. Check out Consultants: Getting Your Money's Worth for how to do that.
Questions? Just ask! [Back to How-to Resources] [Top] |