David Raab Article Archive

The vision is simple: make each customer interaction contribute to the optimal long-term relationship. But you need complicated systems to make it happen: a consolidated database to bring together all customer information; integrated touchpoints to deliver consistent messages; and a strategy manager to deploy customer strategies and measure their results. Today, most companies have some type of consolidated customer database. A smaller but growing number have fully integrated their touchpoints. Only those who have passed these first two milestones are ready to tackle strategy management.

Since just a few firms have reached the third stage, it’s no surprise that strategy management software is still immature. In fact, no complete strategy management system is truly available. Instead, marketers must consolidate separate elements including data mining, optimization and other analytical tools to identify possible strategies and assess their performance; campaign managers to define strategies in operational terms and assign them to customers; and interaction managers to execute strategies as interactions occur. Of these three categories, analysis and campaign management are reasonably well established because they are also needed for conventional outbound marketing. But real-time, cross-touchpoint interaction management only becomes possible when consolidated customer data and integrated delivery systems are available. So interaction management systems are quite new indeed.

The key function of an interaction manager is to tell the touchpoint systems what to do as an interaction occurs. In theory this might mean taking over all decision-making from the touchpoint; in reality, most touchpoint interactions are routine tasks that require no particular marketing input. Even so, the interaction manager needs a way to become aware of events that are occuring at the touchpoint and to capture data about those events that might be relevant in choosing the best response. Of course, it also needs a way to communicate its decisions back to the touchpoint system for execution. In addition, the interaction manager must access other data about the customer such as history, demographics and campaign assignments, and apply whatever decision rules are dictated by the relevant marketing strategies and campaigns.

The distinctions among marketing strategies, campaigns and rules are likely to shift–or perhaps even dissolve–as these systems evolve. Today’s interaction managers generally accept segment and campaign codes generated by other products, but do not import campaign structures or individual rules. They also rely on external campaign managers to run conventional outbound communications. It remains to be seen whether the interaction managers and campaign managers will remain distinct or merge in the future.

Allink Agent (Harte-Hanks Data Technologies, 978-671-6000; www.harte-hanks.com) is one of the few interaction managers available today. Agent can scan a stream of operational transactions and select those worth responding to, execute complex rules to select a response, and send this decision to a touchpoint or other system for execution. Agent is built on powerful technology from a company named Gensym, whose primary business is monitoring and reacting to large volumes of continuous inputs for tasks like running oil refinery. Gensym technology lets Agent read regular operational transactions as they are generated, instead of requiring them to explicitly notify Agent of specified events. This means Agent can be deployed without making internal changes to the operational systems or slowing their performance. Most interaction managers do require modifying the operational system to generate special transactions, raising a key barrier to implementation.

Of course, integration can be avoided only when the operational system does not deliver the results of Agent’s decisions. For example, Agent might scan the transactions that generate a monthly telephone bill and select customers to receive an email from an account manager. This could be done without changing the billing system itself. But when the decision must be delivered through the operational system–say, a marketing message printed on the billing statement–some direct integration is needed. Agent modules are available to help move data in and out of touchpoint systems using a common application program interface (API).

Agent also possesses sophisticated transaction evaluation capabilities. These let it make decisions that draw on data from three sources: the current transaction, such as Web session clickstream; an external relational database such as a data warehouse; and an “interim” file of critical information maintained within Agent itself. This interim file can store key information such as continually-updated summary statistics, sequences of events, and behavior patterns. The ability to monitor behavior against anticipated patterns is particularly powerful. It lets Agent react either when an expected behavior does occur–say, taking advantage of a sales opportunity–or when it doesn’t–perhaps suggesting an attrition risk that should be addressed.

While Agent’s technology lets it monitor and react to behavior patterns, users must identify these patterns and their significance outside of the system. Similarly, Agent provides powerful graphical flow charts to help users specify complicated decision rules and combine these in sophisticated structures, but it will not suggest the rules themselves. These limits are common to most of today’s interaction managers, although some are beginning to integrate automated data analysis. But even by today’s low standards, Agent is limited in how it helps users manage and evaluate rules: reporting is pretty much limited to how many times a rule has executed and what conditions led to this event. The system does not attempt to measure rule effectiveness or outcomes, or to compare performance against expectations. Instead, it relies on users to set up standard test vs. control structures and perform such evaluation externally. The system does include sampling and splitting functions to support such tests.

Agent runs on Windows NT and Sun Solaris servers and uses Java programs for its transaction monitoring and rule definition components. The base system costs $200,000 to $500,000 depending on the size of the installation. Modules to integrate with touchpoint systems and prefabricated rule sets for marketing objectives such as attrition management or cross-sell cost $50,000 to $75,000 each. The system was released in June, 1999. It has been deployed at two customers and sold to about ten more.

* * *

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.