1998 Jul 01
Interactive Relationship Marketing
David M. Raab
Relationship Marketing Report
June-July, 1998

By now, it is embarrassingly common observation that the Internet is well suited for relationship marketing. What’s less obvious is exactly which technologies you need to make it happen.

Start with the database. Surely this must be the foundation, just as it has been for non-Internet forms of relationship and database marketing? And surely it will look like the databases developed for these other applications?

Think again. Conventional marketing databases are designed for segmentation and analysis, not to execute real-time interactions. Analytical applications require data structures that make it easy to ask complicated questions and to scan large amounts of data quickly. These typically involve some type of “denormalized” design, of which the best known is the “star schema” popularized by data warehouse applications. These structures are quite distinct from the “normalized” designs used in conventional transaction processing systems, which are very fast at finding and updating records relating to a single account.

(“Normalized” structures store each piece of information only once, which means elements must be grouped into many different layers or tables; “denormalized” structures use a smaller number of tables but accept some redundancy. For example, a normalized design would have one table for household information such as street address and a separate table for information about individuals. A denormalized design would combine these into a single table of individual records, with the address information repeated on each record. The denormalized structure is easier to query, since there is no need to connect household records to individual records. But the denormalized structure is also harder to update, since several records must be changed to record a new address. This takes more time and leaves considerable room for error.)

One of database marketing’s grandest crusades over the past fifteen years has been to convince the larger technology community that marketers really do need a separate database with a different type of structure. This battle has finally been won, more because the much more powerful data warehouse developers had similar needs than because of anything the marketers themselves really accomplished. But now, in the sort of irony that is only amusing to people who haven’t spent a decade in the trenches, interactive marketing does require quick access and updates to individual records. That is, it requires a database using a conventional, transaction-oriented structure, not the finally-accepted separate marketing database.

Just to clarify matters–or possibly confuse them further: interactive marketers still do need their consolidated, denormalized, analytical databases as well. They will use these databases to find behavior patterns and market segments, to evaluate results, and to define the parameters of new campaigns. These are the traditional analytical uses of a marketing database. But non-interactive marketers also use this same database for their operational function, which is to select names to receive promotions. The analytical database can handle this operational application effectively, because its selections are made against the file as a whole in some sort of a batch process. But the operational application of the interactive marketers involves reading single records as the transactions occur. This is not something the analytical structure handles well. This is why the interactive marketers need a separate database for their marketing operations.

Another way to express this is to say that for interactive marketing, the focus shifts away from database to the application. This is a radical change in perspective from conventional database marketing, where building and maintaining the consolidated marketing database has always been the first and greatest challenge. (In fact, one of the increasingly common issues faced by conventional database marketers is figuring out exactly what to do with their centralized database now that they have finally managed to build it. This need for efficient strategy definition and implementation is driving much of the industry’s current software development. But that’s a topic for another day.)

The application-centric view of the interactive marketing systems is illustrated starkly by their ability to function without any external data at all. Many of the interactive systems are designed to make marketing decisions based solely on the behavior they observe during a visitor’s current session in the Web site. Of course, they prefer to keep a history of behavior over multiple sessions, but recognize the technology does not always allow the system to identify the same person from one session to the next. Even systems that do rely on tracking behavior over time are not necessarily able to integrate information gathered at the Web site with non-Web sources such as a conventional marketing database.

Part of the reason is technical–a visitor can often access a Web site without providing a name, account number or other identifier that will link to other company systems. But many systems ask visitors to register, thus providing information that could potentially be used to match against other sources. Even those system often choose instead to build their own internal database by gathering additional profile information directly during the registration process. Using an internal database makes it easier to design and operate these systems, since they have full control over the format and structure of the data they must access. It also involves an implicit judgement that these advantages outweigh the benefits of integrating with external data. Coming down on the side of efficiency over integration is the classic choice of operational application developers–and quite foreign to the traditional marketing goal of a single, consolidated database.

So even the level of the database itself, interactive marketing systems are profoundly different from conventional database marketing operations. Not only do they use a different data structure–operational rather than analytical–but they are less focused on building integrated databases than on using whatever data is available to manage interactions. Of course there are exceptions to the rule: products including Persimmon IT TargIT (800-546-7242; www.persimmon.com) can in fact access external databases directly. But most others rely on data they gather themselves or, at best, import from external systems through a batch process.

There is no small irony here: the most powerful systems for relationship marketing cannot easily access the full history of a customer relationship. But the database is not the only technology involved in interactive relationship marketing. Next month’s column will look at others.

* * *

Last month’s column began a look at the technology used for interactive relationship marketing. It showed that the marketing database, which plays a central role in conventional relationship management systems, is much less important in the interactive world than the interactive application itself. So just what are those interactive applications, anyway?

In broadest terms, these products generate responses that are tailored to the actions of Web site visitors. This involves identifying the set of possible actions, creating a list of possible responses, and defining rules that determine which response goes with which action.

Products differ in how they handle each of these capabilities. The “actions” they consider may be limited only to events during the current Web session, or may include a richer context provided by actions during previous sessions, information the visitor has provided through Web registration or surveys, and information derived from non-Web sources such as a marketing database. Many products use this information to place the visitor in a market segment and then apply different rules to different segments.

Due to the quasi-anonymous nature of Web contacts, it is a significant technical challenge to link a visitor with any information beyond the current Web session. The most direct solution for visitors to identify themselves through registration or use of an existing identifier such as an account number. This makes it easy to store and retrieve their information and probably makes the most sense in the context of an open, two-way marketing relationship. Nearly every interactive relationship system can work with registration data.

A more passive alternative involves placing a “cookie” on the visitor’s PC. This is a small file that stores a record of past behavior at the site without requiring specific action by the visitor. However, cookies are problematic for several technical and privacy reasons. A more acceptable Open Profiling Standard is being developed by the World Wide Web Consortium’s Privacy Preferences Project (commonly referred to as P3P). Details on P3P can be found at www.w3.org.

A key element of the Open Profiling Standard is that it will be common across applications–potentially allowing different Web site owners to access the same visitor information. Developers including Engage Technologies (978-684-3884; www.engagetech.com) and Firefly (617-528-1000; www.firefly.net) have already developed their own “portable” profiles, which are intended to serve cooperative networks of sites that will use them to improve ad and message targeting. These profiles use standard formats and common survey questions, so information can be accessed by any interactive application. The information is encrypted to prevent outsiders from viewing it and users are supposed to consent to their participation. Microsoft recently purchased Firefly for this technology.

Linking Web-generated information with external files such as a marketing database requires either a common ID such as an account number or a name/address match similar to a conventional data consolidation process. Interactive marketing products generally leave any advanced matching to other systems, although many can import external data given an ID to match against. Persimmon IT TargIT (800-546-7242; www.persimmon.com) is unusual in its ability to maintain a live linkage between an external relational database and the Web application itself. This lets it assign segment codes in the external database and look these up as visitors enter the Web site. The more usual approach is to import the external data, generate the segment code, and store it on the user profile in the interactive application’s internal database.

The interactive systems also vary considerably in the rules they apply to their data. Most allow users to set up a hierarchy of rules use a combination of user actions and profile data to determine which content will be provided. Products like Intelligent Interactions Adfinity (703-706-9500; www.ipe.com), with roots in selecting banner ads, add extensive features to control when, how often and in what sequence each visitor sees different messages. Rubric EMA (650-513-3870; www.rubricsoft.com) and MarketFirst (408-261-6950, www.marketfirst.com) can include complex campaign flows, with branching sequences of actions and options for tasks like report generation and sending messages (such as sales leads) to people other than the visitor. Persimmon IT can also execute this sort of complex marketing campaign, and even offers the type of champion/challenger testing and detailed response analysis used by conventional database marketers. This adherence to standard database marketing procedures is quite rare among interactive systems, although it will eventually be necessary for properly controlled operations. Most of the interactive products (and their users) have not yet reached the state of maturity where they realize they need it.

One of the greatest challenges in setting up rules for Web-based interactions is sheer volume: the number of options and rate of change are so high that manually defining an optimal set of rules is impractical, and probably impossible. Aptex SelectCast (619-646-0121, www.aptex.com) and Nestor InterSite (401-331-9640; www.nestor.com) both use neural network models in combination with more conventional rules to determine which messages are most appropriate for a visitor. Somewhat similarly, NetPerceptions GroupLens (612-903-9424; www.netperceptions.com) uses “collaborative filtering”–rules derived from the behavior of people with similar preferences–to determine what each visitor sees. (While Firefly is also known for its collaborative filtering technology, it announced plans to discontinue it after the Microsoft acquisition.) Both neural networks and collaborative filtering offer the promise of rules that can evolve automatically with changes in user preferences, conditions and offers, overcoming the limitations of manual rule definition.

The final element in interactive relationship systems is the responses themselves. Most systems rely on the marketer to set up a database of “content”, which is typically a series of prebuilt Web pages that will be sent to visitors as the rules determine. Some, including Aptex, Rubric and MarketFirst, can also generate other types of messages such as e-mail, faxes or even physical mail. Aptex is particularly notable for its ability to “understand” incoming text messages and offer appropriate replies, while MarketFirst has an unusual ability to mix several different media (e.g. fax and Web) in the same document. This type of cross-media response is another critical challenge that must be met as customers interact with companies through an increasing variety of channels.

Perhaps the ultimate in interactive relationship management is to provide custom Web pages that are tailored to the individual visitor’s needs. BroadVision (650-261-5100; www.broadvision.com), Micromass intelliweb (919-851-3182; www.micromass.com) and Persimmon IT do this by assembling pages from templates whose contents are selected by system rules. This means that each visitor effectively sees a personal Web site, which is adjusted constantly as their interests evolve. The hope is this will prove so valuable that the visitor will keep returning, generating new opportunities to generate advertising revenue and make appropriate marketing offers.

No matter how advanced, even a personal Web site is just one element in an interactive marketing relationship. Today, no single product combines the most sophisticated methods for handling vistor actions, marketing rules and responses. Although some cooperation is possible among certain systems, most operate in isolation–due primarily to their separate internal databases. We can expect this to change as the industry matures and vendors realize that a single product cannot hope to lead in all these different areas.

* * *

David M. Raab is a Principal at Raab Associates Inc., a consultancy specializing in marketing technology and analytics. He can be reached at draab@raabassociates.com.

Leave a Reply

You must be logged in to post a comment.