Dynamic CRM

Improving customer dynamics for a better bottom line

By Barry Grushkin

Relations between people change and are mulitfaceted, the same as between customers and businesses. Using multiple dynamic customer models to target markets specifically to improve customer loyalty can effectively increase your revenue.

Anthropologists have long been saying that the categories used to divide the world affect how the world is seen and consequently, the actions taken. But different categorizations and categories are better in different circumstances.

In this column, I have regularly pointed out a quantitative equivalent - differing data aggregations show differing results and imply differing action rules. Now I will elaborate to show how both multiple and dynamic ways of looking at customers can combine to markedly improve the bottom line.

The goal of marketing and sales has always been to affect or take advantage of customer behavior. Reasons for this objective have included gaining more customers, trying to sell more products to existing customers, trying to sell products with greater margins, increasing regularity of purchases, or increasing customer longevity.

Checking Out

One common practice involves static cause-and-effect relations. For example, customers are sometimes categorized by lifestyle or socioeconomic markers in an attempt to find which might predict increased interest in certain product mixes.

Physical retail stores have been successfully using the immense body of simple cause-and-effect research for a long time. For example, the type of background music played in a store can affect sales by as much as 15 percent. Lighting, floor location, and height on shelf are well-known variables that also affect product sales. Some physical store practices can have direct, testable, virtual-store correlations: impulse buying at the checkout counter and the "Visa effect."

All those magazines and candies near a grocery store checkout are put there because it is prime buying space - customers pick up and buy a higher percent of items there than in any other place in the store. A recent trial showed impulse buying increased at an e-checkout as well.

The Visa effect shows people somehow think of credit as funny money. When you ask people to guess the value of items and credit card images are anywhere in sight, they consistently guess higher. Tests with online auctions for items with identical Web page displays, except for a credit card logo in the corner of one, showed the credit card logo induced faster and higher bidding.

Customer Segregation

A long-standing practice related to cause-and-effect approaches has been market segregation, where companies target products based on a customer's expected economic behavior.

We are all familiar with an airline's vastly higher prices if you don't stay over a weekend or don't book a few weeks in advance. The airlines have found this method to be a reliable way to actively segregate business travelers who are willing to pay much more for tickets than price-conscious, leisure travelers. (Price inelastic vs. elastic behavior is the jargon that economists use.)

Airline companies have clearly identified a way to specifically change one product - transportation - into two: For short-notice, intra-week flights they can charge premium price, and for the more marginally profitable, but essential time (weekends) they can keep the flights full. They have also found a way to segregate the customers by preventing the less price-conscious customers from nonetheless getting the discount.

Telecommunications companies extensively research predictors of who will churn and who will be loyal so that they can segregate their customer base as well. Obviously acquiring loyal customers costs more, but they are also worth more once acquired. Knowing who is who prevents wasting money on the wrong people.

Retail customers have often been segregated by store; for example, people interested in value more often go to discount stores, while those interested in quality and service go to higher-end stores and boutiques. Not surprisingly, the latter will sometimes end up paying more for items they could have found more cheaply at a discount outlet.

One limitation, however, that is common to all these approaches is the assumption that each customer fits exactly into one category and does not change. Instead, some of my recent work has shown that using multiple and dynamic ways to look at customers can be highly predictive and useful. I believe research into what makes customers shift categories is the mainly undiscovered mother load that data miners have yet to explore. The following research deals with the dynamics of customer loyalty.

Shifting Loyalties

A specialty retailer's goal was to increase customer spending at its online store. In advancing a data mining solution, I first asked the question, "How should I think about and differentiate the customers?"

Linking to a commercial customer information database had many problems. The objections included privacy issues, questionable reliability, the limited number of matches and inconclusive matches with current customers, the cost, the categories sometimes seemed artificial, and the information, when available, was only usable for existing customers, not prospects. Instead, we decided to focus on measures of loyalty itself.

We explored a number of differing schemes for dividing the customer universe. We looked at clustering on other available indicators that would justify some even definitional dividers of customer commitment. We wanted something more detailed about the customer process than simply acquisition, persuasion, and conversion. We developed two systems of categorization: a browser-based minimal, fair, good, very good, and excellent and an economic-based minimal, fair, good, very good, and excellent.

The browser-based categories segregated customers based on differing levels of time spent at the site, content browsed, and the numbers and kinds of items placed in the shopping cart within the last 30 days. The economic-based categories relied on differing spending and payment patterns over the last 30 days.

For example, a browser-based minimal was someone for which we had a cookie and an email address but who had spent less than five minutes last month at the site and did not place anything in the shopping cart. A very good for the economic-based category was someone who had spent between $200 and $300 last month on higher-margin items.

In this study, we focused on dynamics. When and who moved from one category to another and what could be done to influence these changes. We certainly wanted customers to move up the ladder - the minimal to become curious, good to become very good, and so forth. Furthermore, we wanted to find ways to prevent customers from moving in the other direction.

From recent historical data we produced the baseline transition matrix: What was the probability of any customer moving from one category to another over time? We had 535-transition matrices for both browser- and economic-based information, which showed a lot interesting things.

For example, only 5 percent of minimal customers became curious the next month. Each month the number of very good customers that became excellent was fairly equally matched by those going the other way. What could be done to improve this?