Predictive modeling can help businesses better target opportunities with prospects/customers and implement marketing and sales programs in a more efficient way to improve return on investment (ROI). Modeling allows you to select targets that are more likely to act in a certain way. Examples include:
- Prospects who are most likely to buy a new service
- Customers who are the most at risk for leaving you
Once you have modeled and scored your prospects/customers, you can select your best targets (those with the highest scores) for specific programs. Since you are selecting targets with a higher likelihood to act, your ROI for these programs will be much higher than selecting targets at random.
The table below illustrates how targeting as a result of modeling can reduce cost-per-acquisition and drive better ROI. In both cases of this hypothetical example, we have the same number of prospects, the same overall expected acquisition rate, the same goal for adding new customers and the same cost per participant. The only difference is that with the modeling case, a target group of best prospects was identified and, as a result, the program participant pool required to reach to the same results was half the size of its no-modeling counterpart. Since the cost per participant was the same for both cases, the net costs for the modeling case were cut in half, contributing to a better program ROI.
|No Modeling||With Modeling/Targeting|
|Expected acquisition rate of new customers||2%||2% in aggregate
4% for best prospects (top 20,000)
|Goal for new customers||800||800|
|Required program participants to reach goal||40,000
(40,000 x 2% = 800 goal)
(best 20,000 x 4% = 800 goal)
|Cost of program||$10/participant x 40,000 =
|$10/participant x 20,000=
Before embarking down the road of a predictive modeling project, you might want to read through the section below to better understand how the process works. If you’re familiar with the process then you’re ready for the next step, selecting and defining a modeling project. Read about that in our next marketing analytics blog.
Typical Predictive Modeling Process
A Predictive Model allows you to score (rank) a universe of prospects/customers based on their likelihood to "look like" or "act like" a selected target. In order to predict who will take a specific action in the future, we use historical data to determine who has taken this action before. We then use historical data to determine which characteristics are most correlated to taking that action, i.e., which characteristics are able to best distinguish those who took action from those who did not. The diagram below shows a timeline of how a model is constructed.
As an example, let’s apply a target of “identifying non-users most likely to buy our services” to the diagram shown here. Although we specify time periods below, they can be modified for specific projects. In some cases, a much shorter timeframe is more relevant, in other cases – especially when we don’t have a lot of targets - we may increase timeframes.
- We start by pulling together a universe of prospects from one year ago. Since we are modeling a non-user universe, we will likely be limited to using firmagraphic data for our model. Characteristics typically include size, industry, location, organization type, and family hierarchy. This becomes our History dataset.
- We then use the most recent year of data as our Target Window. We tag each prospect that became a user as a Target and all other prospects as a Non-Target. We add a flag to our History dataset that represents the Target = yes or no.
- The modeling process helps us determine which characteristics are most correlated with the Targets. For example, if you sell services that are highly attractive to the education industry, then the education industry might end up as one of the predictive characteristics in your model. Those in the education industry will then get scored higher than those in other industries.
The strength of a model is in its ability to take not just one but a broad range of characteristics and identify the relevant importance of each. Your output might include 5 to 20 characteristics, with each having a specific weight towards the final score.
About the Author:
This article was written by Vida Tamoshunas, Director of Marketing Analytics at SIGMA Marketing Group.