New Rules of Mathematical Marketing

by Mark Klein

Originally published in Target Marketing Magazine, on 01 Jun 08

The marketing challenge all companies face can be divided into two parts: customer acquisition and marketing to existing customers. Today, search engine marketing is the dominant methodology for customer acquisition, and Google is the preferred vehicle. The equally important task of existing customer marketing has a growing consensus around a set of best practices called mathematical marketing (MM).

MM is the scientific process of marketing to existing customers based on a scientific understanding of how past customer behavior predicts future purchases. Key elements of the process include behavioral tracking, predictive analytics, behavioral targeting, testing, what-if analysis and results measurement. For comparison, the typical elements of SEM are keywords, ad impressions, clickthroughs, search engine optimization and Web analytics.

New developments that powerfully enable MM include software as a service delivery, visual math, faster algorithms and new test design methodology. Now that marketers have access to this toolset, the challenge is to apply the techniques and use the tools to create the most effective marketing campaigns.

Rule #1: Capture Transaction Data

Forget about MM unless you’ve got transaction data to understand what your customers do. There’s no behavioral tracking without it. Demographic, psychographic and firmographics data can all tell you something about your customers, but none of those are as useful as transaction data. Customers vote with their purses and wallets, not with their ZIP codes. Customer satisfaction data is highly questionable, too. There is a large body of evidence and academic research leading to the conclusion that people say one thing and do another. Typically, customer satisfaction data does not correlate with future buying behavior. Transaction data, on the other hand, leads to accurate predictions of purchasing.

Rule #2: Use Your Eyes

Analyzing transaction data to predict future customer behavior is not worth much unless you can interpret the results of your calculations. Unfortunately, few of us can translate tables of customer scores into actionable marketing campaigns. We are a visually oriented species and need a visual way to make sense of the mathematics.

Graphic representation of data is nothing new. But creating an accurate picture of 50,000 customers and predicting what they’re likely to do in the future is a more complex problem than simply creating a chart that tracks the price of IBM stock over the past decade.

Happily, this “complex problem” has been solved many times over. Whether you use behavior maps or some other technique, it is possible to make graphical representations of customer behavior that reveal clusters of similarly performing customers and suggest appropriate marketing campaigns. MM means not only doing a scientific analysis of customer transactions, but also presenting the analysis in visual terms so that innumerate marketers can understand what the analysis shows and apply the understanding to their campaigns.

Rule #3: Focus on Customer Needs, Not Just Product Features

Today, most marketing starts with a product focus. The marketer develops collateral that explains the features and benefits of the product or service to be sold, and then exposes all possible customers to this collateral. This method has been working for a long time and will continue to work. But an MM approach opens up an alternative method: marketing to customers based on what they individually need, want or are likely to buy. This is the core idea behind behavioral targeting (BT).

Making an appropriate offer is a great way to improve campaign results, whether it’s presented in an online ad, e-mail, direct mail or telesales campaign. BT delivers customer insight from purchase transaction data as well as, if not better than, clickstream data does. Not only does targeting find customers for your products, but it also helps spot potential defectors, identify candidates for win-back campaigns, pick out promising new customers and, in general, help you address customers throughout your customer set.

Rule #4: Test, Test and Test Again

Most companies, even sophisticated ones, don’t test their marketing campaigns before launch, even though they know that testing is one of the two best ways to improve response. For some companies, they don’t test because they don’t know how to do it. More often, they are put off by fears about the necessary time and/or cost. These fears are unnecessary; they stem from an outdated notion that for a proper test, only one variable can be changed at a time, using what is called split-run or A/B testing. That methodology is indeed time-consuming and expensive, requiring many subjects and trials.

Fortunately, a newer and better methodology exists, called Factorial Design testing. Not only can multiple variables be tested simultaneously, but the interactions, if any, between the variables can be determined, too. If the tests use e-mail, where campaigns are usually completed in 36 to 48 hours, powerful results are available in just a few days. Today, there is absolutely no reason not to test before executing a marketing campaign.

Rule #5: Only Good Surprises Allowed

The history of marketing is littered with fiascoes and failed campaigns, and no marketer wants to add to the debris. What marketers and their managements want is, in fact, the opposite scenario: the ability to predict the performance of a campaign before its launch. Properly done, MM supplies the necessary advance knowledge.

Crude BT delivers the names of customers likely to buy. Better targeting also predicts what products customers are likely to buy. More sophisticated analysis can roll up the individual customer predictions into reliable response rates and revenue for the campaign as a whole. With this information in hand, a marketer can adjust campaign parameters to produce acceptable results.

Variations in the actual results from the predictions should only occur when the campaign collateral is substantially better or worse than business-as-usual standards. Typically, the only surprises are good ones, where the results are significantly better than the predictions. Marketers who don’t use this safety net have no excuses for poor performance.

Rule #6: Look in Every Nook and Cranny

All too often sales teams meet their quotas by going back again and again to their best customers; ones they know are reliable purchasers. Many segments of the customer population are neglected, with a corresponding loss of revenue. A better approach is to design programs to address customers at all points along the spectrum.

“Along the spectrum” is a purposely vague phrase. An easy scale is revenue: Customers could be ranked according to their revenue contributions over some useful time period. The best customers should be near the top, but some of those ranked high on this scale could be there because of past purchases and could now be planning to switch to a competitor. Conversely, there could be newer customers near the bottom end who are vigorous buyers.

An alternative ranking metric is RFM. This covers some of the flaws in a revenue rank but has its own serious shortcomings, including being blind to product information and customer velocity. In addition, there are the usual uncomfortable trade-offs associated with a linear methodology. Two other useful scales are at-risk rank (likelihood to defect to a competitor) and loyalty rank (probability of making a purchase in the next period).

Using these overlapping metrics in combination opens possibilities for some finely targeted campaigns. For example, a win-back campaign to encourage a purchase from customers in danger of defecting might look for customers in an intermediate recency range (not very recent, but not so long ago that they are considered already lost). Further, you want customers with a revenue rank high enough to make them attractive and a risk score that says they are not already gone. Finally, you want customers with some products for which they have a reasonable purchase probability, so you can make an attractive offer to get them to purchase.

Another segment of often neglected customers is those a notch down from the best customers (who are likely to buy anyway) on a loyalty ranking scale, but not so low that getting them to buy is a big struggle. These second-tier customers are usually the lowest-hanging fruit for campaigns to boost overall revenue. A real strength of MM is its potential to uncover a multitude of profitable customer segments.

Rule #7: Keep Score

MM makes this rule easier than ever. Learn what works and what doesn’t. Measure how well you do along the loyalty spectrum. One of several useful plots is median revenue versus loyalty rank, as shown on page 37.

The median revenue distribution of this company is strongly skewed to the high end—most of the revenue comes from customers in the top decile. Since total revenue is proportional to the shaded area under the curve, this company’s best opportunity to grow total revenue is to get more from customers with loyalty ranks from 50 to 90. This is just one example of how keeping score with MM metrics can help guide marketing.

Mark Klein has been a theoretical physicist, software designer, marketer and a serial entrepreneur. Currently, he is CEO of Loyalty Builders LLC, which has just launched Longbow, a new direct marketing service. For more on this topic, check out Klein’s articles, “The Science of 1to1 Marketing” (http://www.longbowdirectmarketing.com/resources/library/the-science-of-11-marketing) and “The New Tools of Mathematical Marketing” (http://www.longbowdirectmarketing.com/resources/ library/new-tools-of-mathematical-marketing).