In our companion post “Understanding the Basics of Predictive Modeling,” we reviewed the basics of modeling, and the principles of how you can “load the dice” for the most favorable response to every marketing campaign. If you haven’t read it yet, you can review it here. Now let’s look at what “loading the dice” looks like in a real marketing campaign:
Each box in the diagrams above represent the same mailing universe, just ordered differently. That’s important because ordering is what can be accomplished using a predictive model (how we “load” the dice). There are fifty people (or dots) and 10 responders (red) in each. The blue box represents the portion of the universe you’re going to mail to. In the example on the left, you don’t have any knowledge to “order” the universe, so the responders are scattered randomly. If you mail to 50% without sorting, you’ll find an equal split of responders and non-responders, or five red dots inside the blue box. In the example on the right, we’ve developed a model that “predicts” a response and allows us to order the universe, placing more likely responders towards the left. Notice the model isn’t perfect, but it does allow the same size mailing to produce 7 responders instead of 5. The box on the left is what you would get using a standard set of dice, and the box on the right is the lift you would see if you used the “special” set of dice and shows a lift of 4%.
When constructing a direct marketing tactic, you and your customer work through audience criteria (the list), the offer and the creative. None of this needs to change in order to use a model. Even if your audience selection criteria are as good as they can be, unless you’re planning to mail to everyone meeting that criteria, a model can help you order the resulting universe and increase your response rate.
However, in most cases you can improve on your basic audience selection criteria. Building a model does this because it identifies characteristics that are similar to previous buyers or responders and dissimilar to non-buyers/responders. It can discover characteristics you might never have considered and places several of them together in complicated ways that result in an uncomplicated ordering of your audience, placing the most likely responders first.
For example, you may have noticed that people who are responding to your mailings are older, have children living at home, and have a college education – so, you decide to mail to everyone who is of a certain age, has children, and a college degree. But you can’t mail to everyone meeting that criteria, so you pick the first 10,000 and go. When you build a model, you can employ statistics to make the same kind of criteria selection but do so based on mathematical principles, and more importantly, using over 1,000 variables to come up with the right mix and formula. That formula calculates a predicted probability of response for every customer in your marketplace and that predicted probability is used to order that marketplace (loading the dice) so you can select the most likely responders.