David Raab Article Archive

Relationship Marketing Technology
David M. Raab
Relationship Marketing Report
August-October, 1998

Call it relationship marketing, customer intimacy, database marketing, one-to-one, or whatever–the idea that companies should tailor their actions to individual customers is now largely accepted as conventional business wisdom. But companies seeking tools to implement the concept find themselves accosted by a wildly different vendors, each promising the One True Way to marketing salvation. The resulting confusion has become a major impediment to industry success.

It doesn’t have to be this way. While stampeding vendors will always raise a lot of dust, the main features of the terrain are still visible underneath. In fact, the marketing database landscape contains three distinct regions: the swamp where it all started, the plains that most are still trudging across, and the mountains of the not-so-distant future. Marketers are spread across all three like a long wagon train whose head is starting into the hills while the stragglers still drag through the swamps far behind. As in any journey, picking the right equipments depends on understanding the terrain you’re in now and where you’re headed next. So here is a look at the major sets of database marketing technologies.

I. Campaign Management

The concept of database marketing developed primarily in the direct marketing industry, which can trace its own roots to the 1890’s and even earlier. But the earliest database marketing software appeared in the mid-1980’s, when something called a Marketing Customer Information File (MCIF) began appearing on desktops in bank research departments. These systems contained a consolidated customer database that was built by linking all accounts that belonged to the same household. Householding was important for bankers because their operational systems treated each account independently, so there was no easy way to get a complete picture of all the relationships the bank had with a given customer or household.

Service bureaus had been creating consolidated customer households files for banks since the late 1960’s. These were typically built quarterly and used to create massive printed research reports on customer behavior. Sometimes the files might also be used to create mailing lists, but this was a slow and expensive process when all the work had to be done on a mainframe by a service bureau’s staff. What was new in 1985 was the files had moved from the service bureau mainframe to the banker’s PC desktop. This meant that incremental queries and selections were now essentially free. At first, this let the researchers ask and answer many more questions about their customers’ activities. Over time, it let bankers develop increasing numbers of selective, well-targeted promotions. So what had originally been a market research tool evolved into a campaign management system whose primary goal was to reduce wasted marketing expenses by allowing tightly targeted mailings.

MCIF systems were in a pioneer phase from about 1985 through 1989, and then sold rapidly from about 1990 through 1994. These were small departmental systems that ran pretty much independently of the corporate IT groups. The main systems (Marketing Profiles, Customer Insight, OKRA, and Harte-Hanks) were quite similar: all used a proprietary database, ran on stand-alone PCs or local area networks, and captured limited amounts of account, household and sometimes individual-level data. They could execute complicated queries, generate lists of customers selected for a promotion, and store a copy of each list for later response analysis. As the focus to shifted over time from research to campaign management, vendors added multi-segment selections, automated creation of promotion history files, tracking of promotion costs, and return on investment analysis. Truly robust campaign management became available only at the end of these systems’ major growth period, and some vendors never fully implemented them. Numerous other campaign management systems appeared after 1994, nearly all using standard relational databases and many targeted at specific industries such as retail or newspapers. Quite a few still survive, although none has sold large numbers of copies.

The early MCIF systems established a set of expectations for what a database marketing system should be like. These included: end-user control without corporate IT involvement; near-instant response to complex, unplanned queries; tightly integrated tools for queries, selections, analysis and reporting; and one-shot campaigns whose results were measured by linking responses to the original audience. These expectations still frame many database marketing veterans’ evaluation of current products, even though requirements and technologies have changed dramatically.

II. Customer Management

Even as the MCIF systems enjoyed their heyday in the early 1990’s, a new vision began to develop. This can be called “customer management”. The idea was to shift from single promotions for specific products to executing the most appropriate sequence of contacts for a given customer. The goal was not just to consummate a single sale, but to maximize the over-all sales to each customer over time. The concept was usually described as share of customer or, more poetically, share of wallet.

At the same time, marketers began to realize that most contacts with customers actually came through day-to-day operational interactions–that is, the actual processes of buying and using company products and services–not through the deliberate marketing contacts received through direct mail, outbound telemarketing or sales calls. The thought then occurred that these operational contacts could themselves provide a channel to deliver marketing messages, through comments printed on billing statements or cross-selling during customer service calls. The idea of delivering marketing messages through operational systems is immensely attractive, both because the message is nearly free–since the contact will be occurring anyway–and because it allows direct targeting to individual customers. Since the idea of marketing strategies tailored to each individual customer was at the core of the new “customer management” concept, it dovetailed perfectly with the idea of delivering tailored messages through operational channels. Opening a link between the marketing and operational systems also provided a way for the marketing system to become aware of events that called from some marketing response, such as closing an account or making a major purchase.

Systems built to deliver on this concept–developing long-term customer contact strategies and executing them through operational systems–appeared as early as 1991. But it wasn’t until about 1997 that sales began to move beyond the pioneer stage to a broader market. In fact, it’s really just now (mid-1998) that the concepts are widely understood and the major competitors–firms like Exchange Applications and Prime Response–are positioning themselves for rapid growth. Most of the industry’s attention is now focused on this type of database marketing, often described as continuous or longitudinal campaign management.

Customer Management Technologies

Although continuous campaign management sounds like an incremental extension of old-fashioned one-shot campaign management, the underlying systems and technologies are quite different. There is still a consolidated customer database, similar to the old MCIF household file. But now instead of being the main focus of the system, this is primarily used as an analytical resource to help identify customer segments and plan long-term strategies. The real action now shifts to an intermediate database that holds a list of customers with their next-scheduled marketing actions, to be delivered through whatever channel the customer happens to appear in. In data warehouse terms, this resembles what is usually called an Operational Data Store.

The real technical challenge here is not so much building the list of planned contacts, as connecting that list to the different operational systems so they can read and deliver its messages. This is a major system integration task, and it means the marketing database is wrenched from the privacy of the marketing department to the bright lights of the corporate data center. Suddenly the proprietary databases that were optimized for desktop analysis are criticized because they don’t meet corporate standards for openness and compatibility; suddenly user desires for flexibility, access and response time are weighed against IT concerns for security, reliability and change management. Not surprisingly, the corporate IT issues often take precedence over the marketers’ more parochial preferences. The systems that survive in this environment are those that use standard relational databases and are open to integration with other enterprise systems. (Interestingly, the specialized proprietary databases have not so much vanished as retreated to inside the newer systems, where they execute high-speed queries and segmentations using data loaded from standard relational platforms.)

In fact, many of the newer systems do not even provide the data preparation and householding tools that were a key feature of the initial MCIF products. These systems assume the necessary database will be built with third-party software or it will already exist in the form of a corporate data warehouse. In a further decomposition of the original self-contained MCIF, the new systems also often rely on third-party software for queries, reporting and analysis–although they usually need to supplement standard SQL capabilities to meet marketers’ complex query requirements.

Vendor Consolidation and Alliances

Since implementation of these new systems is now a major project, external systems integrators and consultants take a larger role in the process. Like corporate IT groups, these new players look beyond current functionality to vendors’ long-term stability, technical architecture and product development plans. These concerns favor larger, more established vendors, leading to consolidation in the marketplace. Integrators and consultants also like to standardize on a single vendor’s product, to reduce learning costs on later engagements. This results in still further consolidation and leads vendors to pursue them aggressively as partners.

The vendors also now seek alliances with developers of integrated front office systems, which combine sales automation, call centers, customer service, and other contact functions. From the marketing software vendor’s perspective, these systems are a mixed blessing: they create unified systems to serve all points of contact (field, call center, service desk, Internet, etc.), thereby making it easier for customer management systems to execute their marketing policies across channels. But the scope of the front office systems also makes it easier for their developers to see customer management as a incremental enhancement to their core products. Thus, the new allies are also potential competitors.

In reality, modifying a front-office system to incorporate customer management is a significant change. The front-office systems are built to handle one-at-a-time transactions between individual customers and company reps. This requires a transaction-oriented data structure that is fundamentally different from the warehouse-type data structure needed for the analytical portion of the marketing database. But the distinction is easily missed during the sales process, especially since many buyers themselves are not fully aware of the issues. Nor is the problem insurmountable: once the front-office software developers recognize the need to build auxiliary analytical databases, they will be able to do this and integrate them with their core systems.

Whether the front-office systems present a major strategic threat to customer management software vendors is unclear. Although the front-office vendors clearly have the resources and motivation to develop competing applications, the front-office market is highly fragmented. This means the majority of the market would still be available even if several of the leading front-office vendors managed to lock out third-party solutions. On the other hand, the front-office market itself is consolidating rapidly, so access to clients of the leading vendors may prove indispensable. It may also turn out that the only practical time to install customer management is when a company is replacing its existing front-office systems. This would also make it vital to sell customer management systems to the front-office vendors’ new clients.

In any case, the proper strategy for customer management software vendors seems clear: to position themselves as a component that is easily integrated with any front-office system. This will encourage the front-office software developers to adopt the customer management vendors’ products, rather than building their own, and also let buyers consider purchasing independent customer management products. To the extent that easy integration requires building prepackaged interfaces to specific front office products, there is a cumulative business advantage for the customer management vendor with the greatest number of interfaces. This is yet another factor favoring larger vendors and leading toward industry consolidation.

Customer Management Functions

The internal functions of the customer management systems differ substantially from the features of earlier campaign management products. Most obviously, the new systems must let users construct campaigns involving a time element–that is, define a sequence of promotions to be executed at predefined intervals. This typically involves laying the campaign out on a flow chart, usually with branches that specify different paths depending on what the customer does after the campaign begins. A first rate system will automatically create the exclusions and tracking flags that move customers from one point in the flow chart to the next. A not-so-good system will require the user to write queries that create those exclusions and flags manually–something that is doable, but error-prone and a fair amount of work. An ideal system would also coordinate messages across multiple campaigns, to ensure a given customer is not overwhelmed with marketing contacts. This might be done by setting priorities among campaigns, defining limits on the number of permissible contacts, or allocating a marketing budget to each individual. Tools for this type of coordination are just beginning to appear, however.

Although flow charts and campaign coordination are desirable, the true minimum requirement for customer management is an automated job scheduler that will execute promotions as they become due. This is essential because customer management implies a much greater number of file selections than traditional one-shot campaign management. Often each campaign must be executed nightly and would itself involve multiple selections. This is the only way to determine which customers are due to be contacted and to react to the past day’s events. But the customer management selections themselves are often simpler than under campaign management. This is because customers are assigned to relatively large, stable segments, where they remain while a long-term contact strategy is executed over time. This contrasts with the old product-oriented campaigns, which used very detailed, custom-built segmentations designed to do the best possible job at finding the most likely buyers for a specific product. The shift away from elaborate, promotion-specific segmentations means the newer systems have less need for the complex ad hoc queries and data exploration. But marketers brought up in the old school still often look for those capabilities during their evaluations, anyway.

Managing long-term contact strategies also means a shift away from counting the responses to a specific promotion, to measuring cumulative changes in behavior over time. This requires the new systems to track a consistent set of fundamental business metrics that are independent of individual promotions or products. These include measures like customer profitability and non-purchase behaviors like service requests. Gathering these metrics may require automated feeds from accounting and operational systems, further increasing enterprise-level integration requirements. Analysis often uses multidimensional data structures and tools to look at aggregated behavior by segment.

Organizational Issues

While the technical implications of the customer management model are profound, the organizational impact is still greater. Since marketers now rely on operational systems and users to deliver the bulk of their marketing messages, they must coordinate with the operational departments to an unprecedented degree. This requires involvement in training and compliance monitoring, adjustments to compensation schemes, coordinated planning and scheduling, transfer pricing for internal resources such as service rep time, advance notice of marketing activities and even approval by operations of marketing offers. Although the real challenge is changing business processes to allow this coordination, there is an impact on the systems as well: they must add workflow, approval reporting, workgroup security, project scheduling, budgeting, requirements forecasting, and other administrative features. They also need to translate promotion messages into multiple media–that is, to deliver the same offer through a telemarketing script, Web page, billing statement, e-mail, or whatever other contact channel the customer might use. Work on this type of translation has barely begun.

III. Relationship Management

Although campaign management is just entering the stage of broad adoption, the outlines of its successor are now coming into view. Call this Relationship Management.

The key distinction between relationship management and its predecessor is that relationship management uses the marketing system to guide operational decisions, not just to define marketing messages. These decisions involve matters such as pricing, credit authorizations, or service levels. While the distinction may seem subtle, it actually implies yet another fundamental transition in technologies.

One reason is that involvement in operational decisions requires real-time interaction, so the marketing system can guide transactions as they occur. This differs sharply from even the overnight batch updates that characterize the most responsive of the older customer management systems. It involves a variety of technologies that can accept inputs and return results in real time.

Coordination with Operations

Even more important, guiding operational decisions requires somehow embedding the marketing systems within operational systems’ internal processes. This is a vastly more intimate integration than using the operational system to transmit a message. It means that the operational system must somehow check with the marketing system as it makes a decision, possibly modify its decision based on the marketing system’s response, and then continue on to implement the modified decision. In concrete terms, the difference between the two types of integration is that one means printing a message in a reserved area on your electric bill, while the other means changing the calculation of charges themselves.

Relationship management cannot be accomplished merely through changes within the operational systems, because the goal is to coordinate all interactions with a customer. This means the relationship management system must provide a single place to specify and implement marketing policies through whatever operational systems are involved. This is a new and much more difficult challenge than simply modifying operational systems by themselves.

Happily, organizations have other reasons to coordinate real-time activities of different operational systems. For example, they increasingly wish to link manufacturing, planning and logistics systems to allow just-in-time manufacturing and distribution. So the necessary underlying technology, sometimes called middleware or processware, is slowly being developed for purposes that have little to do with relationship management. (In a similar fashion, database marketers have benefited as data warehouse requirements forced standard relational database engines to improve their analytical performance.)

Other cutting-edge technologies, including component-based designs, object-relational database, and message-based communications, also facilitate sharing processing across systems. Enterprise-wide resource management systems like SAP and Baan may also make integration easier, since they leave fewer systems to communicate with. The enterprise system vendors could also pose a competitive threat similar to the front office systems; in fact, the enterprise and front office systems are already beginning to integrate directly with each other, creating still greater potential to squeeze out the independent relationship management systems. There is little the relationship management system vendors can do to block this risk, except possibly license their own technology–as some are now doing. But they can take comfort in the thought that few firms will commit fully to a single enterprise vendor’s system, so corporate IT departments will continue to press for technologies that move data among multiple vendors’ products.

One obvious observation is that for relationship management systems to take advantage of these new sharing technologies, they must themselves be compatible with them–something that could require massive redevelopment for the earlier generation of customer management systems. A more important observation is that the new relationship management systems are fundamentally transaction processing systems, not data warehouses like the original MCIFs or Operational Data Stores like the newer customer management tools. Being a transaction processing system imposes a set of requirements for performance and reliability that are well understood within the IT community but still new to marketing databases. These are essential, because the marketing system is now mission-critical: if you can’t process orders when the marketing system is down, because the marketing system determines your prices, then the cost of marketing system downtime is suddenly much higher than a missed mail date. Perhaps even worse, an error in the marketing system could mean customers are charged the wrong prices, with immediate, negative bottom-line impact. So all the traditional fail-safe, backup and checking mechanisms needed in other core transaction processing systems must now be applied to the marketing system as well.

These requirements imply that the physical design of the marketing database will resemble other transaction processing systems–typically a highly normalized relational structure. This is very different from the denormalized, analytical design used in data warehouses or the much simpler structures of an Operational Data Store. It also means the new marketing systems will be built and managed by transaction processing experts, not the data warehouse, marketing or analytical staff who have controlled previous marketing systems. So we see an entirely new set of issues, technologies, and people.

Analytical Functions

Although the executional portions of the relationship marketing systems will essentially merge with the rest of the company’s transaction processing infrastructure, these systems will also have a separate, large analytical component that designs the policies to be implemented and evaluates their results. Like the rest of the relationship marketing system, it will differ radically from analogous portions of earlier marketing databases.

Earlier systems relied on the concept of predefined campaign applied to everyone in a single segment, perhaps with a few branches for major contingencies. But the relationship marketing system will be involved with a myriad of operational decisions based on individual circumstances. It is quite possible under this scenario that no two customers will ever be treated exactly alike. This means that marketers will need analytical methods that let them treat each interaction on its own, while somehow predicting and optimizing the long-term results. It implies a new framework that moves beyond the sum-of-purchases approach of traditional lifetime value analysis to incorporate elements such as loyalty, referrals and service costs. So far, no standard model has emerged for this analysis, but interesting work is being done with theories developed to calculate values in the securities industry.

While the details of the new analytical process are unknown, its general outline can be guessed. It will start with a heavy dose of modeling, presumably against an analytical database similar to existing marketing databases or data warehouses. Users will first create a process model that defines the different interactions over time between their company and its customers and prospects. The model will include both operational events such as purchases and service requests, and promotional events such as marketing offers. Each event will have several outcomes and each outcome will have an estimated value. There probably be separate values from the company and customer perspectives–for example, allowing an out-of-warranty return may have a small negative value for the company but great positive value to the customer.

The next steps will be to identify customer segments and behavior patterns and to use these to build predictive models. With these models, users will be able to estimate the impact of applying different policies to different interactions. Based on their estimated values, users will pick specific sets of policies to test in the real world. They will then implement these and monitor results, making adjustments as needed to ensure the total value increases over time.

While this process is broadly similar to traditional “closed loop” database marketing, the number and complexity of variables will require a new set of control mechanisms. Indeed, the issue of control–how do I know what I’m really doing to my customers? How do I know I’m doing the right things?–is probably the greatest challenge in the relationship marketing model. Policies are no longer executed directly by the marketing system, as in campaign or customer management; instead, they interact with one or more operational systems before an action is presented to the customer. As in any complex system, predicting the result of these interactions will be difficult, and in some cases the system will yield unexpected results. Thus part of the infrastructure needed for relationship marketing will include advanced simulation capabilities–to show the expected net impact of a rule change, given all the other systems surrounding it. This will very likely involve advanced data visualization to make the results understandable.

Another part will be near-real-time reporting of actual results, to identify any deviations from expectations in time to make corrections. Given the volume of transactions involved, this type of deviation reporting will need to be highly automated. In fact, it is likely that the system will be empowered to not merely identify deviations, but actually to react on its own–if only by reverting to a previous or default policy. This will be necessary to quickly limit the damage from any mistakes. Once this kind of system is in place, it’s only a small step to letting the system initiate its own tests of new policies–say, by slightly varying prices or credit limits and then determining whether the results constitute an improvement. Given the huge number of potential variables, some sort of automated testing is almost inevitable if companies are to approach optimal results.

The final analytical challenge faced by relationship marketing systems will be evaluation. Each customer will be impacted by different policies in different systems, which will likely be adjusted at different times. This will make it very difficult to isolate the effect of any single rule. But there will be no alternative: because each customer will receive a unique combination of treatments, the option of assigning customers to large, persistent segments and comparing their over-all results will simply not exist. So marketers will have to rely on a combination of direct observation and sophisticated statistical methods to tease apart the individual factors accounting for customer behavior. Again, this will need heavy automation to cope with the volumes involved.

Changing Roles

If analysis is largely automated and execution is in the hands of operational systems, what role is left for the marketers? It seems they will revert to what marketers have always done best: developing product and service offerings that meet customer needs. This is a near-revolution for database marketers, who have had to develop increasing levels of technical expertise over the years as they built and operated their own systems. Those with a primarily technical bent will presumably move into the corporate IT groups who will now take over the day-to-day operation of the marketing systems. Those who remain, freed by automation from routine testing and evaluation, will be able to focus on the sort of major marketing innovations that automated optimization systems cannot create.

Marketing system vendors will face a similar change in roles. Although they will need to build interfaces for their systems interact with operational processes, the primary control over those processes will still reside elsewhere. Thus, the primary function of the marketing systems will migrate back to the analytical functions–process modeling, prediction, simulation, and evaluation–along with marketing administration functions such as budgets, planning, approvals and policy deployment. Given the great complexity of these functions in a relationship management process, this will be challenge enough.

Existing Systems

The specific technologies needed for relationship management are still being developed. The concept itself is so new that no complete systems yet exist. But solutions that address parts of the problem are beginning to appear.

One set of technologies involves methods to manage real-time interactions. These must both respond to immediate inputs and adjust automatically as behavior patterns evolve. Many real-time interaction technologies are related to the Internet, where real-time response is essential and the number of paths that a particular session can follow is effectively infinite. This has led to systems that autonomously assess a customer’s behavior and decide how to react. Practical applications include responding to an e-mail query, selecting Web sites in a search engine, listing books related to prior purchases, or presenting an appropriate banner ad.

Real-time technologies are also increasingly common in areas like fraud detection and credit risk analysis. One side effect of the desire for real-time interaction is an increasing need for real-time access to external enhancement data. This lets the marketing system look up additional information about a new customer or prospect during the course of the initial interaction. Several systems to allow this type of look up, both across the Web and against data stored in-house, are now available. These systems raise significant privacy issues that may eventually result in substantial limits on their utilization.

Online commerce systems offer another glimpse of the relationship management future, since these already offer near-seamless integration between operational and marketing systems. But in many cases, the entire e-commerce system is isolated from the rest of the organization, so its capabilities are not easily extended to other channels.

So-called “marketing automation” systems, such as Rubric and MarketFirst, offer a grand vision of cross-channel customer strategies combined with a practical focus on real-time Internet interactions. However, these systems are built primarily for business-to-business marketing, and are in fact largely limited to lead management. They are barely integrated with existing front-office applications and not at all with other operational systems. Whether they develop into something more robust or are simply absorbed into the front office products remains to be seen.

Oddly enough, one branch of the database marketing industry has been combining operational and marketing processes for nearly twenty years: the points-based loyalty systems used primarily by airlines and other parts of the travel industry. These systems have always been outside the mainstream of database marketing technology, precisely because their main features had to do with operational processes including points accounting, customer service, and distribution to front-line locations such as reservation centers, ticket counters and departure gates. It will be interesting to see whether lessons learned in these systems turn out to be useful for other types of relationship management.

Coincidentally (or maybe not coincidentally), the airlines have also been leaders in yield management–that is, real-time price adjustments in response to market demand. This process is arguably the most advanced existing example of delegating a key operational decision to an automated system. Yield management is attracting increasing interest in other industries, including hotels, auto rental, shipping, and telecommunications, and to some extent even retail. These are all industries with a strong interest in loyalty systems, so some form of marriage between the two approaches may yet occur. This would imply charging different prices to individual customers based on their personal history. In some ways this seems an obvious next step, although there are considerable technical, customer relations and perhaps even legal obstacles.

The Internet seems to threaten yield management by making it easier for consumers to compare prices. Yet this may actually accelerate the merger of yield management with loyalty programs by giving companies even more reason to develop programs that retain customers even when lower prices are available elsewhere. The Internet also provides a direct connection to individual customers, bypassing the travel agents whose order-placing role has been one of the practical roadblocks to customer-specific pricing in the travel industry.

Interestingly, the financial services industry–the traditional leader in database marketing applications–has shown little interest in loyalty systems or yield management approaches. If these systems do somehow represent the industry’s future, it is possible that leadership will shift away from financial services to another sector.

Conclusions

Three Generations

There are three substantially different generations of database marketing systems. The first generation systems were stand-alone MCIFs used for one-shot campaign management. These are largely obsolete, although many are still in use. Later campaign management systems, most using standard relational databases, are still appropriate in some circumstances. The concepts underlying campaign management systems continue to shape many marketers’ perceptions of what a database marketing system should be.

The second generation of systems perform “customer management.” These define customer-specific sequences of marketing contacts, which are distributed through both operational systems and traditional outbound media. Most of the industry’s efforts over the next few years will be devoted to selling and implementing second generation systems. At many companies, the organizational issues involved with customer management implementation will be much harder to manage than the technology.

Third generation involves “relationship management” systems that will guide individual operational decisions to optimize long-term customer relationships. The details of this approach are still being worked out and even if a fully developed system did exist, the market is not yet ready to buy it. The third generation will probably remain in its pioneer stage for another five to ten years. The one factor that might accelerate this is the Internet, which moves very quickly and requires many of the third-generation system capabilities.

The technologies, staff skills and organizational implications of each generation are profoundly different. This means that there is relatively little carry-over from one generation to the next–in fact, some old habits must be unlearned to prosper in the newer environments. It also makes it hard for vendors to convert to a new generation: indeed, the leaders of the first generation (Customer Insight, MPI, OKRA, Harte-Hanks) are definitely not the leaders of the second, and yet another set of companies (mostly involved in Internet marketing) are early leaders in the third.

Implications for Vendors

The implication for vendors is they must decide whether it is too late to enter the market for the second generation applications: while new systems continue to appear, their functionality is roughly the same as the industry leaders and the leaders have the advantage of position. (This is NOT to say that all systems are the same or that the leading systems are necessarily the best. There are some very interesting alternatives that any buyer would do well to consider. But the industry has reached the stage where marketing prowess rather than technical superiority will determine a vendor’s long-term success.)

Vendors who wish to focus on third-generation solutions should realize that the key technologies have to do with analysis, simulation and real-time interactions. Integration with operational systems, seemingly the most important challenge of all, will be solved by other portions of the IT industry for their own reasons. Anyone addressing third-generation solutions should also bear in mind they are blazing a trail for others and can expect little immediate return for themselves.

Implications for Companies

For companies that wish to implement database marketing, the simultaneous presence of these three generations causes a great deal of confusion. Hopefully this paper has removed some of that. The critical observation is that the three generations are truly quite different. This means a company new to database marketing can start with the current generation, customer management, without needing to master campaign management first. It also means that if a company is not yet ready for customer management, it should buy a modern campaign management system instead. This will best suit its current needs, and let it buy an up-to-date customer management system when it is ready for one in the future. Companies wishing to try relationship management should recognize that complete solutions are not yet available and prepare for continued experimentation instead.

Perhaps most important, companies should realize that customer and relationship management involve organizational issues that are ultimately more complicated and difficult than any of the technology. This may lead some firms to scale back their initial aspirations to more traditional campaign management, although competitive pressures will probably not allow this in most cases. Instead, firms need to address the larger issues head-on and to treat customer management as a business process, not a marketing program. In many cases, they will need outside resources who are experts in managing this type of change. While such resources can be expensive, the cost of a failed implementation is much higher.

* * *