March 2007 - Backend Business

Predictive Modeling: Is it for Everyone?

By Dr. David Castillo

There are several challenges that organizations face in their search for new customers and in targeting their marketing campaigns. Companies have a number of available channels for reaching their target audience. Depending on their pocketbook, they can use direct mail, e-mail, telemarketing, radio advertising or television, among others. Or, they can choose to use some sort of multichannel approach.

Obviously, there are specific costs associated with each of the available channels, as well as expectations on the response associated with each channel. Therefore, it is important to understand the target audience and the objectives of the campaign. Marketing campaigns that are not targeted are notorious for underperforming.

Knowing who your customers are in terms of demographics, lifestyles and purchasing behaviors allows you to develop messaging aimed at converting a prospect into a paying customer, or transforming an existing customer into a higher-value one. Without this knowledge, you run the risk of placing everyone in the same category. The challenge for businesses today, especially small businesses, is to maximize each marketing dollar spent. One way to achieve this objective is to employ predictive modeling as a means for approaching marketing as a quantitative function rather than a qualitative one. I refer to this process as marketing as a science, as opposed to marketing with a gut feel.

When I use the term predictive modeling, I am referring to the process of applying quantitative methods to focus and drive marketing programs. Whether it is a customer acquisition campaign, an upsell or cross-sell campaign or a loyalty program, predictive modeling employs quantitative methods for determining the target audience of the marketing initiative.

Predictive modeling is the process of developing predictors or variable factors, such as age and gender, to foretell the likely outcome of events. The term model implies that there is a relationship between the predictors. In other words, not all predictors have the same effect on the predictive power of the model.

Marketing analytics is not a new concept; many companies employ quantitative methods to drive their marketing programs. Some companies staff in-house analytics groups, while others outsource their marketing analytics needs to firms specializing in quantitative analysis. In either case, predictive modeling is performed by highly skilled analysts versed in the application of specialized techniques to problems associated with direct marketing.

Because of the skillset required, predictive modeling can be a time-consuming and potentially costly activity. Despite this, most companies that employ it in their marketing programs tend to achieve better overall campaign performance over those who use non-quantitative methods.

Predictive modeling often begins with establishing a customer profile from a company’s customer database. Customer profiling is the process of determining the key attributes and behaviors of the customer base. For example, a company may know that its customers tend to be married, middle-aged, have children, and that they purchase their products online rather than through retail outlets. Profiling also involves identifying the various customer segments of the customer base-specifically, segregating the top performing groups from the lesser-performing ones.

Marketing analysts employ an assortment of methods to determine the key attributes of current customers. Predictive models make use of those attributes in order to identify the target audience of the marketing initiative. Often, third-party data containing demographic and lifestyle attributes is used to enhance the customer records as a means for improving the overall performance of the predictive model.

Depending on the marketing problem being solved, the predictive modeling process can range from straightforward to complex. Moreover, the time and cost associated with developing and deploying the predictive model can increase significantly when third-party data is employed. It is not surprising then, that many small- to medium-sized companies have opted to deploy their marketing campaigns without the benefit of sophisticated predictive modeling.

Over the last two years, there has been increased pressure from the marketplace to develop more cost-effective and timely methods of predictive modeling. As a result, some analytic companies now offer Internet portals that allow companies to submit their customer files for analysis. Users upload their customer lists and are notified when the results of their analysis are complete. Many of these portal offerings deliver reports and analysis back to the user and make recommendations regarding prospect selection.

One of the key benefits of these portal offerings is that it makes predictive modeling available to those who previously were not in a position to utilize it. Companies that lacked analytic expertise, or that did not understand the process of developing and interpreting predictive models (or simply could not afford it), are now able to gain value from the automated process. (An aside worth noting is that business people utilizing predictive modeling portals are often not interested in tables of data containing mathematical relationships among the predictors in the model. Rather, they simply prefer to see ordered prospect lists and geographic maps identifying where the prospects are located relative to the business.)

Automating the predictive modeling process has another less obvious benefit. Human analysts who perform predictive modeling often are faced with deadlines and other responsibilities of their job functions. All too frequently, they find themselves in situations where they must deliver a predictive model in a timeframe that is acceptable to their customer. Consequently, they may not be afforded the opportunity to explore and refine the model as much as they would like.

By automating the process, there’s an opportunity to improve the fidelity of the predictive model. For a given timeframe, automation allows more exploration of the predictors and their relationship to the performance of the model. More iterations are possible, since the entire process is automated and does not require manual intervention. When human analysts are involved, they must balance their exploration with the reality of their deadlines.

The marketing strategy is the foundation of the marketing plan; the plan itself is the list of specific actions that are required to achieve the goals set forth in the strategy. These actions require knowledge of customers, competitors, business climate, etc. There is little argument that predictive modeling provides a method for improving overall campaign performance. However, is predictive modeling a viable option for every business? It is, with access to an automated predictive modeling service. With such a service in place, predictive modeling is no longer available exclusively to those companies and organizations with access to highly skilled analysts and customized technology tools.

For those companies that use automated predictive modeling services, there is a secondary benefit. No longer bottlenecked with specialists, predictive modeling capabilities can permeate the management-and even branch management-of marketing departments. It creates a form of data democracy that can lead to more and better campaign investments for everyone.

Dr. David Castillo is founder and CEO of CopperKey in Gilbert, Ariz. He can be reached at (480) 633-1966.


No Comments

No comments yet.

RSS feed for comments on this post. TrackBack URI

Leave a comment