Database marketing has emerged as one of the primary reasons for building a data warehouse. But, too often the people asked to build these systems--if not the prospective users themselves--lack a firm understanding of exactly what they are expected to accomplish. This, of course, is a sure-fire recipe for disaster.

It doesn't have to be this way. Database marketing is a well-established industry with roots stretching back thirty years. An entire infrastructure exists of service and software vendors, consultants, conferences, publications and associations. In fact, the very depth of the industry's heritage poses a challenge: because so many people have been doing database marketing for so long, many different applications have evolved within the general category. This means that even industry veterans must listen carefully to understand what someone means when they say they want a "database marketing system."

Here is a rundown of the major applications people might have in mind when they describe database marketing. One note of caution: while different applications originated in different industries, most organizations today expect their database marketing system to support several, if not all, of these functions.

Householding

Arguably, this is the mother of all database marketing applications. The term originated in the banking industry, where each product (checking accounts, savings accounts, certificates of deposit, etc.) typically ran on an independent transaction processing system. Banks needed a way to bring together all accounts belonging to the same family, or household, so they could avoid duplicate mailings and gain a complete picture of the relationship. The processing to do this involves sophisticated matching algorithms that can deal with inconsistent, incorrect and missing data; standardization to meet postal regulations; parsing to identify data elements and business terms; and business rules peculiar to each institution. Specialized vendors including Innovative Systems Inc., Harte-Hanks Data Processing, OKRA Marketing (now Harland) and Customer Insight Company developed the software to do this, originally on a service-bureau basis and later on desktop systems. Often they fed the householded files into an independent marketing database system referred to as MCIF (Marketing Customer Information Files). The householding process resembles the consolidation required for data warehouses, and many of these vendors are also selling their tools today for that application.

Campaign Management

Catalogs, book clubs and non-traditional direct marketers such as financial services firms need systems to simplify the production and evaluation of their mailings. "Campaign management" systems provide end users with the ability to identify the best prospects for a given mailing, export them to a mailing list or telemarketing file, keep a history of what was sent to whom, match responses to the original promotion list and analyze results. These systems must allow non-technical users to define the type of complex selections that tie SQL in knots. (For example, "Take a 100-name random sample of customers in the top 10 percent bracket with more than two complaints in the past three months who have not been contacted by customer service.") They also involve a level of detail--specific transactions by customer--that overwhelms multidimensional database tools. These requirements lead some campaign management vendors to offer proprietary database engines, often using an inverted or "columnar" data structure. However, most recent systems work with standard relational databases. The recent systems have also shifted their focus from one-shot mailings to automated programs that continually assess each customer and send appropriate messages based on specific events or time sequences. This often involves a daily update of the marketing database, compared with the monthly standard of the past. Unlike a "pure" read-only data warehouse or data mart, campaign management systems also generate data such as promotion history and response indicators.

Prospecting

Credit card and insurance marketers, whose profitability depends on careful screening of their customers, often build prospecting databases on large segments of the population by merging information from multiple sources. Business marketers also build smaller databases of people in their industries, and catalog marketers have recently begun to build prospect files by pooling lists of their customers. The consolidation process used to build these systems is essentially the same as the householding process used in other marketing databases or data warehouses. However, the large volume of information--hundreds of millions of name/address records--poses special challenges and often leads to these projects being outsourced to vendors such as Acxiom, Harte-Hanks or Dun & Bradstreet. There are also special security and accounting issues, since some data is restricted to certain uses and data owners must be compensated as their information is employed. Prospecting databases typically have a simpler file structure than customer-oriented marketing databases, since there is little transaction history. But marketers often link a prospect file to a customer database, so they can exclude customers from promotions and find prospects who resemble current buyers.

Point-of-Sale Data

Retailers have many of the same requirements as other database marketers, but they face additional challenges related to the anonymity of their customers and the volume of their data. The anonymity problem has largely been solved in the grocery industry through the now ubiquitous "loyalty cards" that are scanned along with each customer's transaction. Vendors of grocery database marketing systems such as RMS Inc. and S2 Systems have developed tight links to major point-of-sale (POS) products to import their data easily. These systems also make it easy to reward customers based on their purchase volume--a key element in encouraging use of the cards in the first place. Loyalty cards are being adopted less rapidly by other types of retailers, who are less willing to offer volume-based incentives and often can rely on credit card numbers as a de facto customer ID. But regardless of how they capture the data, retailers must find effective ways to evaluate huge amounts of detail on a customer-by-customer basis. The customer dimension makes the problem even harder than a "conventional" retail data warehouse, which tracks sales by item by day at the store level. The traditional solution among retail database marketers has been to assign each product to one or more merchandise groups, aggregate results by group for each customer and day, and discard the detail. Some vendors now also store the raw detail and allow custom aggregations as needed. Early vendors, such as Retail Target Marketing Systems and RMS Inc., used proprietary data structures for fast access to the aggregated data, but newer systems rely on merchant relational databases.

Predictive Modeling

In the telecommunications industry, the primary initial justification for many database marketing systems was higher customer retention. The database would accomplish this through statistical models that identified customers who are likely to discontinue their service, thereby allowing the marketers to take targeted preventative action (such as making special offers that would be too expensive to offer to everyone). Traditional direct marketers, who had been dabbling with statistical models for years but often faced data preparation times stretching into months, also looked to the marketing database as a repository that would streamline the modeling process. Support of modeling requires both an extract mechanism to feed data to modeling tools such as SAS or SPSS and a deployment mechanism to place the final scores on the customer records for use in selections and analysis. Deployment can be accomplished either by importing scores generated by the modeling tools or by calculating the scores within the marketing database itself. The latter solution is technically more demanding, since there must be some way to import or duplicate the model algorithm and to execute complex calculations within the marketing database itself. But in-place calculation may be more practical on large databases where extracting and reloading millions of customer records would be time consuming and costly. Today, some database marketing systems also provide the capability to actually build the models themselves--typically using a quasi-black box approach suitable for non-technical users. Although the results may not be as reliable as conventional statistical models, this approach is particularly useful for marketers who lack access to statistical experts or need to deploy many models very quickly.

Loyalty Programs

Ask an airline executive about database marketing, and chances are she will tell you about her rewards program. Originally designed to track and reward frequent flyers, these programs are now nearly universal among hotels, auto rental agencies, cruise lines, casinos and other segments of the hospitality industry. Unlike most database marketing--or data warehouse--systems, loyalty programs have a major transaction processing component: credits must be accumulated, rewards must be debited and customer inquiries must be answered. The systems must execute complicated and frequently changing program rules, attempt to control fraud and track outstanding liabilities for accounting purposes. They must also be tightly coupled to the reservations and ticketing systems. Not surprisingly, loyalty systems don't look much like traditional marketing databases. But today, hospitality marketers are increasingly using their loyalty data for conventional database marketing activities, such as campaigns to sell additional trips or retention programs for at-risk customers.

Contact Management

From the simplest electronic Rolodex to the most advanced customer-asset management system, these applications share a database of customers and prospects, a history of past interactions and an ability to schedule future contacts. These systems are most widely used in business-to-business marketing, either as personal contact managers, centralized field-sales automation systems or integrated marketing systems that encompass field, telephone, direct mail and other contacts. The range has recently been extended by firms like Scopus and Vantive to include customer service, help desk and other non-marketing activities. Like the loyalty systems, these products have a transaction-processing flavor, particularly when used in a call center or help desk environment. But most also have a basic campaign management capability, although they often lack the complex query and segmentation capabilities of dedicated campaign management products. These systems also tend to have very limited data cleansing and householding functions, largely because their users are expected to maintain the data themselves as they work with individual accounts.

Enterprise Access

As the inclusion of help desk and customer support systems in the last category suggests, the line between marketing databases and operational systems is increasingly blurred. In fact, even though database marketing developed from different applications in business, financial services, retail, hospitality and telecommunications, marketers in all of those industries are now working on the same task: integration with their customer contact systems. Whether the medium is an automated teller machine, airline gate agent, hotel desk clerk, grocery store cashier, customer service call center, long distance operator or tech support Web site, the goal is to give each customer appropriate, consistent marketing treatment. This is strategically important for the kinds of weighty reasons that pay for boardroom consultants' Mercedes: global competition, customer retention, product commoditization, brand differentiation, market fragmentation, and so on. What it means from an IT perspective is that the database marketing system must exchange data with operational systems in near real time, must continually execute batch processes and scoring algorithms to create new contact instructions, must incorporate individual "campaigns" into long-term marketing strategies and must measure behaviors beyond simple response or non-response. Because full implementation of this vision might involve hundreds of simultaneous marketing campaigns, the systems also need to manage the details of internal marketing operations (budgeting, approvals, start and end dates, and even project scheduling) to make sure the promotion materials are ready when needed. Products including Exchange Applications ValEX, Prime Response Vantage, IBM Target and EDS/dbIntellect Marketing Toolkit are all built for some or all of these functions.

It should be clear by now that a marketing database is much more than another name for a data warehouse or even a data mart. While there is considerable overlap in the technologies employed to build and manage these two types of systems, marketing databases also have their own unique requirements--particularly regarding updates and complex, ad hoc queries against huge volumes of detailed data. Understanding these requirements and the specific applications users have in mind is essential to project success.

Register or login for access to this item and much more

All Information Management content is archived after seven days.

Community members receive:
  • All recent and archived articles
  • Conference offers and updates
  • A full menu of enewsletter options
  • Web seminars, white papers, ebooks

Don't have an account? Register for Free Unlimited Access