Modern marketers often frame their role as developing customer-centric marketing strategies to maximize some metric of importance, such as response rates, ROI, or cost per acquisition (CPA). Predictive models are analytical tools that provide us a formal way for maximizing (or minimizing) those metrics. This post is Part I in a series offering a deep dive into how predictive models are developed and can be used in new customer acquisition.
What is a model? In the most general sense, a model is a simplified description of reality that can be used to improve understanding or make a prediction. A model car, for example, might not be full size, or even have an engine, but it can help a design team decide if it is worth building. A road map is also a model. Maps are simplifications that do not include every possible detail, but they have enough information to be useful.
A marketing model works similarly. Predictive models for acquisition marketing are simplified representations of a person’s decision-making process. In effect, we are trying to learn how attributes that describe a person relate to a purchase decision. Of course we cannot include every attribute that goes into their decision-making process, but we might have enough to be useful in making a prediction.
The predictions of acquisition models are represented by a score – a numeric value measuring the likelihood of an action happening. This score serves as a ranking or prioritization for acquisition marketing campaigns. We know we could maximize the number of new customers by targeting everyone, but often that is not realistic because the population of potential prospects is simply too large.
Using the model score, we can select only those people most likely to respond to marketing and dramatically reduce the cost of targeting.
The two most common types of models used by marketers for new customer acquisition are:
- Response models; and
- Cloning (“look-alike”) models
Response models (and variants like conversion models) are designed to rank prospects by a propensity to take an action, such as responding to a direct mail advertisement or clicking on an email link. They are useful because they are directly predicting the event of interest – namely, will the person respond to our marketing efforts.
A higher model score is meant to represent a person more likely to respond to the messaging. Before we can build a response model, however, we need to collect results from a past campaign.
A cloning model is designed to rank prospects by their similarity to existing customers. They are popular in marketing because they are relatively simple to create and use; you only need a sample of existing customers.
A cloning model is built by learning how the attributes describing customers are similar or different from a random sample of the prospect population. The model score, therefore, is interpreted as “the degree to which this person ‘looks like’ an existing customer.
Once you have determined the type of model that best suits your needs, the next step will be to source the data and build the model—we will explore how to do both of these things in Part II of this series.
See how a regional furniture retailer utilized predictive models to improve their customer acquisition campaigns by identifying and targeting the most responsive consumers among premover and new mover audiences. Click here to discover the results.