Models Every Marketer Needs

As companies continue to improve CRM programs and marketing databases, they are often faced with an embarrassment of riches when it comes to opportunities for building predictive models. While virtually any behavior that can be tracked and measured can be modeled, many marketers get lost in endless modeling projects that never seem to deliver the promise of more efficient and profitable marketing programs. Companies in such a trap can benefit by focusing on modeling the three key points in a customer life cycle — response or acquisition, attrition or renewal, and repeat purchase.

Response or acquisition models

Perhaps the grandfather of all direct marketing models is the ZIP code model. But many marketers have overlooked this powerful tool, even though it proved itself long ago. With the explosion of individual and household data available on internal marketing databases and third-party compiled databases, many modelers focus on household or individual response models, despite their higher data and implementation costs.

If you don't have a ZIP code model of customer acquisition or response, you should build one — now! ZIP code models have many desirable properties that can improve marketing decisions and make them the workhorses of any DM program.

First, they are relatively easy and cost-effective to build. As long as you have ZIP codes for current customers or previously mailed prospects, you can summarize this data by ZIP, overlay relatively inexpensive demographic and psychographic data at the ZIP level and use simple, straightforward regression techniques to build a model.

Second, ZIP code models are easy to score and maintain. Since the characteristics of a ZIP code change slowly over time compared with household or individual data, ZIPs should not have to be scored more than once or twice a year.

Third, implementation is easy. Once ZIP codes have been scored, you can rent prospect names or select current customers for future mailings from only the best, highest-scoring ZIPs. If you're a regular renter of vertical or subscriber lists, you usually can select your best ZIP codes at little or no additional cost.

Finally, an often overlooked benefit of ZIP code models is that they can be used to target other media, including local print, broadcast or outdoor ads. For marketers with retail locations, these models can identify and even assist in site selection by identifying “pockets” of adjacent, high-scoring ZIP codes.

While ZIP code models won't deliver the predictive power of a household or individual level response or acquisition model, their lower cost, ease of use and flexibility make them one of the most valuable tools in any company's direct marketing arsenal.

Attrition or renewal models

Many direct marketers tend to focus on response and acquisition and leave retention to customer service or other departments. However, from publishers to credit card issuers, it's clear many companies don't have an acquisition problem — they have an attrition problem. It's not unusual for firms to lose half or more of their customers within the first year of a relationship. Attrition modeling gives marketers the ability to identify their most “at risk” customers and target them for proactive customer care and other retention activities.

Attrition models typically use logistic regression or other binary modeling techniques to predict the likelihood a customer will cancel an order within a specified time period. It's important to make this prediction far enough in advance to try to “save” such a customer.

Regarding data, there's good news. Customers' own behavior is generally the best predictor of customer attrition. If you've already invested in a customer database complete with historical purchase, transaction and payment data, it's unlikely you will need to buy outside data for attrition modeling. While demographics can provide a profile of attritors, it generally doesn't offer much additional predictive value over in-house data.

Finally, while attrition models can be tricky to build and require an experienced statistician to maximize performance, their ROI can be quite high. Even small, incremental decreases in customer attrition can compound over time to produce a more profitable and loyal customer base.

Cross-sell or upsell models

It's common sense, but if you sell more than one product or service, you can significantly increase marketing ROI by turning single buyers into multibuyers over time. Cross-sell or upsell models are built to identify the customers representing the best opportunities for additional sales. By modeling past buyers of different products or services, you can build a model or a series of models that allows you to decide the next best product to promote to each of your customers.

While cross-sell models can target your best customers, they have another benefit: They will identify your least responsive customers for future promotions. This will allow you to “rest” these customers from future mailings and greatly reduce mailing quantities. Given the recent postal rate increase, these cost savings alone may be enough to pay for the model.

These models also can save you from “burning out” your list. Rather than mailing every customer every offer every time, these models can manage contact strategy based on each customer's predicted tendency to make repeat purchases.

Like attrition models, these are usually built using binary modeling techniques, and also perform best on internal customer data rather than external demographics. These models can be difficult to build, though, so once you have identified the product or products to model, gather data, find an experienced analyst and get started.

While there are many other models, from lifetime value to sophisticated multivariate segmentation types, don't invest in those until you have built the ones discussed here. These models are tactical rather than strategic in nature and will provide an immediate lift to your bottom line.

If your goal this year is to boost new customer acquisition, get started on a ZIP code model. If you decide instead that it's time to increase the value of your existing customers, then focus on building customer attrition and cross-sell/upsell models.

Joseph M. DeCosmo is president of DeCosmo & Associates, a Burr Ridge, IL-based data mining and marketing research firm.