The Power of Database Marketing
by Scott A. Neslin, Albert Wesley Frey Professor of Marketing
Philadelphia retailer John Wanamaker once said, "I know half of my advertising is wasted. The problem is, I don't know which half."
Database marketing tells you which half is wasted by identifying the customers for whom your marketing works, and those for whom it doesn't work. It can link marketing data to customer purchase information and help you eliminate wasted advertising. It can predict which particular marketing campaigns will work on which customers. Telecom companies use it to reduce customer churn—that is, customers leaving one cell-phone company for another. Banks use it for cross-selling, identifying which holders of a checking account would be most likely to purchase an IRA, for example. Retail companies use it to identify responders to a special offer, or to predict how profitable a customer will be over a year or over a lifetime. It also aids customer channel management, letting companies direct customers to the right sales channel—Internet, catalog, or personal sales—to reduce costs without eroding customer loyalty.
Formally defined, database marketing is the use of customer data to enhance marketing productivity through more effective acquisition, retention, and development of customers. That is, it helps manage the customer relationship, and for that reason, I have begun referring to database marketing as Analytical CRM (customer relationship management).
The key to database marketing is manipulating huge quantities of customer data. The three types of data are purchasing data, marketing data, and customer characteristics. Purchasing data is also called transactions data. For instance, a catalog company will have a detailed database on what you have ordered over a long period of time. Marketing data records the company's marketing efforts. The company will have records on how many catalogs they have sent you and what offers they've made. Customer characteristics include such things as your age and income level. Companies may ask you directly for this information, or purchase it from a third party that collates U.S. census data.
A variety of statistical techniques, some proprietary, others publicly available, have been developed to manipulate such customer data. The goal is to predict whether the customer will respond, whether the customer will churn, and what product the customer is likely to buy next. Even small increases in accuracy "multiply through" to create huge gains in profit, because these small gains are over millions of customers. It's very much like the retail store that operates on a small margin but makes it up in volume. Some of these statistical techniques are simple variations of regression analysis. However, some of the more sophisticated "data mining" techniques improve significantly over the basic methods.
I researched the value of these different statistical techniques for predicting customer churn. Telecom companies on average lose about 2.5 percent of their wireless customers a month to churn. To retain these customers, some take a reactive approach: The company waits until a customer calls to say, 'Goodbye, I'm switching.' Then the company makes the customer an offer. It's not likely to be successful. The customer has already decided to switch. It takes a big bribe to convince him or her to stay. A more proactive approach is to identify ahead of time customers who are at risk to churn, and provide them with a reward or incentive that convinces them to stay with the company. The challenge of course is to identify—predict—how likely each customer is to churn sometime in the future. With poor prediction, you'll waste a lot of money rewarding customers who didn't need the reward. However, if you can identify which customers are likely to churn with just a little bit better than random accuracy, you can save a lot of money. The issue becomes: How can these predictions be as accurate as possible?
To test different statistical methods, I organized a tournament. I provided 30 to 40 statistical model builders—some academics, some company practitioners—with a database covering 100,000 customers of a certain cell-phone company. Each entrant used a different statistical model to predict customer churn. I then compared the accuracy of their different methods. The most important finding of this study was that the method did matter. There was, in fact, a big difference. The accuracy achieved by the best entrants would have meant millions of dollars to the telecom company.
How do you find the best technique? If you're a company, you need to do your own testing. Try out different algorithms on your own customer database. Then run a market test, making an offer based on one set of predictions and tracking how many customers respond. This means that companies must always be on the lookout for the "next best thing." This takes time and effort, but our study illustrated it's worth your while to do so.
Database marketing is highly accountable, unlike a lot of marketing. Take TV advertising. We believe it's working, but there's very little understanding of how well mass marketing really works compared to the amount of money that is spent. With database marketing you can test your predictions. Your predictive model might say that people with certain characteristics are most likely to respond to a certain kind of catalog. Take 100,000 of your customers and divide them into 10 groups—deciles—ordered by how likely the model predicts they are to respond. Within the first decile, the top 10,000 customers, randomly send 5,000 catalogs. Do the same with the bottom decile. The other half of each decile doesn't get the catalog: They are your control groups. Then observe over the next few months to see what these customers buy. The difference between the top decile and the bottom decile won't be big, but it should be apparent. To the extent that the top decile sees large gains versus its control group, your predictive modeling is working well.
Again, one way database marketing asserts its value is to identify "small" efficiencies that multiply out over millions of customers to produce huge gains in marketing productivity You might improve the response rate to a marketing campaign from 1 percent to 5 percent. That's a small gain. But multiplied over hundreds of thousands of customers, it can mean millions of dollars. Today, we have the data, the analytical techniques, and the frustration with mass-marketing that is driving the growth of database marketing.
Scott A. Neslin is the Albert Wesley Frey Professor of Marketing at Tuck.
mba
tuck
dartmouth
database marketing
|
