Predictive Tactics: Five Database Modeling Strategies

If you think you’ve done all you can to improve the success rate of your list selections, you should stop reading right here. But if you suspect there’s more to be done, try database modeling. It’s a tool that can benefit nearly every catalog in terms of fine-tuning list selections — both on housefiles and outside lists.

Here are five tactics to consider for boosting the effectiveness of modeling:

Tactic #1: Recognize which kind of models can be most useful for you and will give you the biggest bang for the buck.

According to Bryce Connors, director of the Consulting Services division for Unica, a Lincoln, Mass.-based provider of enterprise marketing management solutions, four types of models are most applicable to catalogers:

• customer segmentation models, which break out customers based on their similar characteristics (e.g., behavior, demographics);

• response models, which predict who in the database is likely to respond to certain types of mailings and offers;

• models with cross-selling capability that predict which next product a customer likely will buy based on his or her past purchases; and

• lifetime value, which examines customers’ past behaviors and looks forward to predict who may be the most valuable customers in terms of sales over time.

The response model is the most basic, and therefore would likely be a starting point for many catalogers, Connors says. “This model gives you, the mailer, answers on whom to mail or not mail so you can narrow your audience, reduce costs and trim circulation. The mail vs. no mail — zero or 1 — outcome you learn from this type of model will give you the biggest bang for your initial modeling investment. You’ll see a difference right away in your response levels and reduced costs.”

Tactic #2: Use a model to help refine your mailing strategy to cut unwanted circulation.

Printing and postage are big cost outlays for a catalog. A model can help you cut such costs by reducing unwanted or unnecessary circulation.

WearGuard-Crest, a large business-to-business (b-to-b) mailer in the uniforms supply industry, was able to do just that using Unica’s Affinium Model, part of the Affinium Suite for Enterprise Marketing Management. WearGuard-Crest has used Affinium to build what it calls “response-spend likelihood” models, which help it to target only those individuals likely to buy from a catalog. It’s based on past purchases, region and other demographics. The cataloger reduced its direct mail costs by $2 million last year without a reduction in response volumes, says Kelly Fiedler, director of product marketing at Unica.

Use modeling to help refine your housefile mailing strategy and guide your decision on how far to go into your file to maximize profitability. Depending on your goals, mail strategies and frequency, you could vary the mail depth accordingly, Fiedler says.

Tactic #3: Use “what-if” analyses to help you focus on the most profitable portions of your housefile or other portions of your mail file.

This is especially effective when dealing with, say, gift recipients and catalog requestors. Says Connors, “We already know these are potential buyers, and we know where they live. They have a good likelihood of being profitable customers. So by comparing them to your housefile we can look for the most similar, and therefore most likely to be profitable, segments of those groups to mail first, then next and so on.”

The same strategy can help you find new profits in your database of former buyers. Says Connors, “Look at the back of the database: your 48-month names. There’s a chance to reactivate them.”

Of course, you don’t want to mail the whole group at once, he says. It’s a tactic you test into from the model. “After you find what looks like a sweet spot, a glimmer, then test it. Then you mail a little more.”

Tactic #4: Use information gleaned from the model to help you better target your catalog creative and marketing offers.

Using predictive modeling you can extrapolate the results of version-tests of creative (e.g., different covers) to identify the best target customers for each creative tested. There’s also the potential, if you have the data at your fingertips, to do other interesting things, such as identify key customers and print specially targeted offers on catalogs using print-on-demand techniques.

“We all appreciate that the catalog-planning cycle is long, and you can’t always do things on the fly like test creative,” says Connors. “With the right data there are things you can do that you couldn’t do before and that are of real importance for long-term customer value.” For instance, he says, “You can look at value-tier migrations, and extending an offer to upper- and mid-tier customers to keep them at those higher buying levels so they don’t become lower-value customers. That kind of creative needs data [to back it up].”

Fiedler says other applications include examining when is the best time to make an offer and what’s the best week to mail. “For example, when should Customer A get your big book? This may differ depending on when they came into the buying cycle, what business they’re in, if they’re a b-to-b customer, or on other factors like seasonality.”

Tactic #5: Analyze data from your models to see if they lead to new or different merchandising directions.

“Supplying more data and analyzing those data open up other possibilities: from identifying new cross-sell opportunities to uncovering the idea for a new specialty catalog for a certain customer subsegment,” Fiedler says. “It’s customer analysis using a combination of data mining and predictive modeling.”

Connors adds that by using a combination of types of models, you can look at customer segments to find new homogeneous groups of customers. You may have had preconceived notions of who your customers are, he notes. “It becomes a cycle where this is supported by the mailing of only those segments and those kinds of outside lists.”

Fiedler concludes: “Being able to analyze different types of customers and what they buy helps uncover trends in customer behavior.”