The Unified Theory of Database Marketing
Database marketing is all of a sudden the new holy grail. What with customer relationship management, data warehouses, one-to-one marketing, data mining — a number of techniques are capturing management time, attention and bucks, and they are all built around customer information.
So now we have an opportunity to pick up some of the powerful approaches that database marketing has been developing over the past 50 years and apply them to our Web businesses, to generate substantial value.
That said, there is a general misconception abroad about what database marketing can do. I will never forget the comments of a very senior but very uninformed sales vice president at a company I was associated with several years back. My job at that company was to help apply database marketing to an old-line business that had lots of data but little clue of how to put it to work.
This sales executive, who was extremely influential, indeed powerful, was in a position to be a valuable supporter of the project. But to hear him talk about database marketing was to cringe. Here is the kind of stuff that came from his mouth: “This stuff will let us find blue-eyed, one-armed paperhangers who graduated from Yale, and then we’ll be all set.”
Yipes, what’s a database marketer to do? The spirit is willing but the flesh is weak. I know that somewhere, deep inside, he “got” what database marketing is generally about, but he had no clue as to the art of the possible.
So I am inclined to lay out a few principles of what database marketing can and cannot do — what database marketing is really about and why it is so powerful.
The first thing you have to keep in mind is data availability. My sales vice president’s confusion was that he did not understand that marketers have only a limited amount of information about customers and prospects to work with. You may have your customers’ transaction histories, some demographics you have gathered or rented and, if you were smart, your promotion history.
Then, database marketing boils down to three major marketing applications:
Profiling for look-alikes. Profiling usually supports customer acquisition. This application analyzes current customers — or, better yet, your best customers — and identifies their characteristics. This is called a profile, and it might be generated by a cluster-type analysis (CHAID or CART) or a multivariate regression analysis, where various characteristics are identified as having either positive or negative correlations.
Once you know that about your customers, or your ideal customers, you then can apply the model to vast universes of unwashed prospects, select the closest matches and communicate with them. In theory, and usually in practice, these modeled “look-alikes” will be much more likely to become good customers for you than the great unwashed universe as a whole. Profiling uses analytics to identify strong prospects, saving vast amounts of money that might be wasted on promoting to the uninterested.
This approach also can be used in reverse — profiling unprofitable customers, for example, and using those characteristics to suppress or avoid doing business with undesirable new customers.
Segmenting, and treating the segments differently. This application is used primarily for customer retention marketing. The idea here is that not all customers are of equal value or have similar preferences or buying patterns. So a smart marketer will analyze the customer base regularly, divide the base into groups and set up programs, policies and processes that allow them to be treated differently to increase the likelihood that they will buy more and stay longer, or whatever is the marketing objective.
The classic segmentation analytic is RFM, where the customer base is regularly grouped based on the recency, frequency and monetary value of their transactions with the marketer. RFM is easy to perform with the kind of data most sales operations have lying around, and it is extremely powerful at identifying individual customers by value.
There are hundreds of other ways to segment, too, depending on the needs of the business. Segment by purchase channel preference, or by sales coverage, or by prior purchases. By geography, language or demographic characteristics. The only requirement is that the segment needs to be real, meaning that people in one segment are truly different from another, and that the segment is useful, meaning that there is real business value to be had by grouping people this way.
Predictive modeling. Some may argue that this application is nothing more than another type of profiling. I would not disagree. But it is such an important contributor of value to the whole notion of database marketing.
Predictive modeling is used to identify the likeliest people among a customer or prospect universe to respond to any given marketing campaign or offer. It considers not only the characteristics, but also the purchase or response propensities of the buyers. It is perfect for upselling and cross-selling.
Predictive models sound a bit like voodoo, but they are based on a simple concept quite like the one used in profiling: finding look-alikes. First, you identify the people who did buy that product, say, or who bought through that channel, or who bought at that certain point in their relationship with you. Identify their characteristics, which include not just who they are, but also when, where and what they tend to buy. Then apply those characteristics to the rest of the population, looking for the likeliest prospects for the campaign or program.
If you happen to have data about one-armed paperhangers, who knows, it might be predictive of future behavior. The point is, database marketing is very straightforward, it makes a lot of sense, it is a powerful marketing technique, and it deserves to be better understood.