Six Things to Consider Before Building a Predictive Data Model

Over the last two years, we have seen the use of Predictive Data Models rise and become one of our most utilized solutions for both offline and online marketing campaigns. Industry wide, these models have become a standard across companies of all sizes and industries.

While there are virtually unlimited applications for these intelligent targeting tools, the success of any model is largely determined by the data used to create them, and how that data is prepared. In order to streamline the back and forth with your modeling experts, here are six things that should be considered in advance of building a predictive model.

1. What Is the Goal of the Model?

Predictive Models come in many different forms and go by many different names, which can make it difficult to evaluate which one is best for you. However, by focusing on your business goals and how a predictive model can help achieve those goals, it becomes much easier to choose the model that will best meet your needs:

  • Acquisition — Is your business goal to acquire new customers, and therefore find like prospects that have a similar profile and buying habits as your existing customers?
  • Campaign Optimization — Do you have an upcoming marketing campaign - whether retention, engagement-driven, or other - that you would like optimize for response and ROI while at the same time cutting costs?
  • Winback/Reactivation — Would you like to target the former customers that are most likely to convert back to active, paying customers?

2. What Metrics Will Determine the Success of the Model?

Once you have determined the purposes and goals of the model, the next step is to decide which metrics will be used to measure success. Commonly used metrics measure engagement levels, response, pay-up, ROI, or some combination thereof.

We cannot overstate the importance of determining the success metrics of your data model upfront, as this is how you will ensure a clear and concise focus moving forward.

3. Who Are You Trying to Target?

Needless to say, it is important that the data provided for a model has to resemble the ultimate population you are trying to reach. The most common reason models don’t achieve their objectives is due to the fact that there is a discrepancy between the file used to build a model and the target audience.

Obviously, things like demographics, lifestyle, behavior, and interest all play a key role, but those factors are not enough to get a good picture of your customers. Often there are small overlooked factors such as marketing or geographic restrictions, recency of the transactions, and channel specific customers (i.e. internet only buyers) that, when it place, could impact the model.

4. Is the Input File from a Specific Recency?

It is always good to take into account the recency of the target audience you are looking to build the model from. Generally, our modeling experts say that it is best to use at least twelve months of your most recent responders to factor out seasonality for acquisition or campaign optimization models. The data needed for a Winback/Reactivation-focused model is typically more involved.

5. Is the Planned Offer Associated with the Campaign Hard or Soft?

Your input file should also take into account the type of offer you are advertising. There are two basic offers: a hard offer and a soft offer.

  • Hard offer — a push to buy the product or service right away. Hard offers may be accompanied by a heavy promotion such as a discount.
  • Soft offer — a push to get an initial response that will begin to build a relationship with individuals, with hopes that subsequent marketing efforts will lead to a purchase. Typical soft offers include subscribing to emails, to newsletters or to request more information.

It is important to take the type of offer into account because if your input file is filled with individuals who typically respond to soft offers, there may be more barriers to get them to convert. Conversely, if you have a file filled with those who respond to hard offers, you could be doing yourself a disservice by sending already established buyers a soft offer.

6. How Will the Offer Be Redeemed?

Finally, the way in which your offer will be redeemed helps to determine the kind of data model your input file is best suited for. After all, you ultimately want your campaign to generate action, so you are going to want to have a particular redemption method that is suitable for your audience.

Offers can be redeemed in a number of different places such as online, in-store, by mail or over the phone. You want to make sure your file, and ultimately your target audience, is receptive to the specific offer in the channel to which you are applying the model. For example, if your offer is redeem by phone only, make sure you supply customers to the modelers that have already demonstrated a willingness to contact you by phone. Conversely, if you are sending an offer to be redeemed online, make sure your target audience consists of tech-savvy individuals who will be likely to take advantage of the online redemption option.

Conclusion

In order to build a successful and accurate Predictive Data Model, it is imperative that you have a clear goal in mind, metrics in place to measure success, and that you provide a file that strongly represents your intended target market. Above all, communicating to your modeling experts any and all potential marketing restrictions that could constrain the model, such geography, will ensure that your model will meet your marketing goals.