Using Predictive Models, Part 3

|  June 14, 2005   |  Comments

A predictive modeling primer for marketers. Last of a three-part series.

This is the last of a three-part series that started with defining predictive models and how marketers can use them. Part two discussed recency, frequency, and monetary (RFM) models, as well as CHAID models. Today we'll move on to regression and other more "advanced" models.

Regression Models

Regression models are almost always more predictive than RFM and CHAID. So using them results in better targeting and higher response rates.

Regression analysis is something most of us learned about in the eighth grade. I developed my example of a regression model (predicting someone's shoe size based on his height) using Excel. The data sets are usually a bit larger in marketing applications. You'll need SPSS, SAS, or some other statistical package to conduct the analysis needed to develop the model.

Although CHAID analysis puts every prospect into a precisely defined segment, regression analysis uses weights, or coefficients, to score each individual prospect. The result is a set of scores with a higher resolution that provide better differentiation for ranking.

Neural Networks and Genetic Programming

More advanced methods of building the models can result in even better (i.e., more precise and accurate) predictions. To marketers, this means higher response rates and better return on investment (ROI).

Neural network and genetic programming techniques examine data in a more in-depth manner than RFM, CHAID, or even regression model analysis. By examining more combinations of variables (income, age, location, etc.), these advanced techniques can uncover more customer information than basic modeling techniques.

These techniques are often used to uncover hard-to-find relationships in data to detect fraud in industries such as credit cards, mortgages, insurance, and credit applications. Fair Isaac is one major provider in this arena.

On the genetic programming front, a program called The-Gmax helps marketers determine what useful patterns exist in data that they can exploit to improve response to their campaigns.

SAS and SPSS also have neural network software, and Unica has modeling software that incorporates aspects of many of the different types of models we've discussed in this series. To its credit, Unica doesn't try to hype the advanced software algorithms in its marketing materials.

Specifics of how advanced algorithms work is beyond the scope of this column. When you really dig into how these models work, it helps to have an advanced degree in math or statistics. But it's important marketers understand these types of models exist, and that once developed they can be used just like any other type of model. Like RFM, CHAID, and regression models, these are simply a methodology used for scoring and ranking prospects or candidates. Obviously, this is valuable to marketers.

Using Is More Important

Models built using regression analysis or any of the other advanced techniques usually aid marketers most. What's important isn't which algorithm you use but that you use predictive modeling to improve marketing.

No matter the model, it's important to understand the model is just an equation or method of helping the marketer rank and pick the best prospects for her campaigns. The statistical algorithm is used to select important variables and weights (as discussed above). The result is used to score and rank prospects.

The Results

I haven't yet seen a situation in which using a valid predictive model didn't improve response rates.

I worked with telecommunications companies to improve their targeting for customer acquisition and retention by developing models that use available data to predict which customers are most likely to respond to their marketing. I know educational institutions that use models to predict which students are most likely to have difficulty completing their courses. A major shipping company uses models to identify prospects for new services.

It still surprises me to find marketers (and there are many) who don't use modeling to their full advantage.

What models do you use?

Vote for your favorite products, services, and campaigns! The ClickZ Marketing Excellence Awards recognize ClickZ readers' choices for achievement and innovation in online marketing technology, solutions, and execution. Voting runs until Wednesday, June 22 (EOB, EST).

Brian Teasley is the leader of Teasley, a consultancy that helpsadvertisers, marketers and advertising agencies use data and analysis toimprove their marketing campaigns. Brian has over 14 years experience inengineering and marketing, and has worked for numerous Fortune 100companies. Brian also teaches a marketing course at New York University. Heholds a M.S. degree in Applied Statistics from Iowa State University and aBA in Mathematics and a BA in Mathematics and Statistics from St. OlafCollege.

Get the ClickZ Analytics newsletter delivered to you. Subscribe today!

Featured White Papers

US Consumer Device Preference Report
Traditionally desktops have shown to convert better than mobile devices however, 2015 might be a tipping point for mobile conversions! Download this report to find why mobile users are more important then ever.

E-Commerce Customer Lifecycle
Have you ever wondered what factors influence online spending or why shoppers abandon their cart? This data-rich infogram offers actionable insight into creating a more seamless online shopping experience across the multiple devices consumers are using.

Jobs

• SEO Specialist
SEO Specialist (Marcel Digital) - ChicagoSearch Engine Optimization (SEO) Specialist &nbsp; Marcel Digital is an award winning digital marketing...
• SEO / SEM Manager
SEO / SEM Manager (CustomInk) - FairfaxAre you a friendly, motivated, and inquisitive individual? Are you a positive, out-going leader? Are you...
• SEO Analyst
SEO Analyst (XO Group) - New YorkSEO Analyst @ XO Group About this Job, You and Our Team: The XO Group SEO Team is&nbsp;looking for you, a passionate...