If you’re a marketer who’s not using predictive models to increase your response rates, you need to read this series (here’s Part One). In this second part of a three-part series on predictive modeling, we’ll discuss how models are used, and touch upon some of the algorithms marketers and modelers use to develop models.
Predictive models are simply equations that help marketers target their best prospects. You ought to use them.
How is a Model Used?
Models are used to score and rank prospects. A model by a retailer selling ice skates would probably incorporate age and location in the U.S. to help predict how likely someone is to buy ice skates. The company probably shouldn’t target 60 year-old men in the South.
While this example is pure common sense, in practice, the development and use of predictive models will always outperform a pure “common sense” approach to targeting. The reason is good models are better able to make correct judgment calls, and simultaneously take into account multiple factors and variables.
Two Common Types of Models
Predictive models vary by what they’re trying to predict (response models, churn/attrition models, high-value-customer models, credit risk models, etc.) and/or by the variables or algorithms used to develop them. Regardless of type, variables, or algorithm, a predictive model is a system for scoring prospects.
Marketers familiar with predictive models often know two common types: RFM models and CHAID models are popular in part because they’re both relatively easy to understand and to implement.
A common type of model in the catalog industry is an RFM model (Recency, Frequency, Monetary). Using these, a catalog company decides who to mail their next catalogs to based on the how recently and how frequently each customer has purchased, and how much money they spent. This is a simple, useful model.
All the RFM analysis I’ve seen was accomplished using either SPSS or SAS software. If your prospect list is small enough to be handled in Excel, it would be easy to score and rank prospects provided the correct data (for recency, frequency and monetary values) are available. If you have to do much data manipulation, you’re probably better off using a different tool.
While RFM models are valuable, other models will perform better, and other algorithms can produce valid, useful models.
A CHAID analysis (chi-squared automatic interaction detection) can incorporate recency, frequency and monetary variables, but can also examine other variables to increase predictive power. Many marketers like CHAID analysis because the result is a set of precisely defined customer segments.
A CHAID analysis looks through all the variables to find the one that appears to have the biggest single impact on a response rate (or whatever it is you’re analyzing). If your average response rate is 12 percent and the analysis finds your customers younger than 21 years of age respond at a rate of 21 percent, while the balance of customers respond at 7 percent, CHAID will inform you there’s a performance “break” at age 21.
If you were only going to use one variable, you’d probably only target prospects under the age of 21. But CHAID allows you to further segment prospects and use any and all predictive variables available.
The final result is a series of defined segments ranked by their predictive response rates. Marketers can thereby understand the characteristics of their best prospective segments and target those top groups.
SPSS has had a version of CHAID available for some time. These days, it’s not part of their standard statistical software, instead it’s part of their Answertree product. While SAS long ignored CHAID analysis (perhaps because other algorithms can yield better results), they do have a version of it in their Enterprise Miner software.
Using is More Important
In the final column in this series, we’ll cover models built using regression analysis, as well as a couple of other algorithms: neural networks and genetic programming models. The type of model you use isn’t as important as whether or not you’re using predictive models to improve your marketing. If you aren’t, you’re leaving money on the table. More accurately, your customer is leaving their money on their table — or they’re giving it to your competitor.
Marketers create personas to better understand their target audience and what it looks like. If marketers can understand potential buyer behaviors, and where they spend their time online, then content can be targeted more effectively.
What’s behind a successful data-driven marketing strategy?
One of the major challenges in the martech industry is getting the attention of prospects in a world where they are bombarded by content and emails on all sides.
Facebook is addressing one of the biggest missing pieces of its chatbot offering: analytics.