Over the past two years, an increasing number of vendors have offered software to help coordinate customer treatments across separate touch-point systems. Enough of these interaction management systems have appeared that it’s now possible to identify two subspecies.
Products including E.piphany Real Time, Black Pearl Knowledge Broker and Manna FrontMind focus on recommending the most appropriate offer to make in a specific situation. These systems are most obviously distinguished by automated model-building capabilities that help select the best offer.
Another set of products, including Revenio Dialog, Verbind LifeTime and YellowBrick PathWay Solution, are designed to deliver complex sequences of messages through different touch points over time. These products are distinguished by graphical interfaces that let nontechnical users design multistep, branching campaigns.
DD Series (Data Distilleries, www.datadistilleries.com) clearly belongs to the first group. It originated in work at a Netherlands mathematical research institute and provides a sophisticated infrastructure for building and deploying predictive models. This includes modules to extract data from source systems, manipulate and store the data in formats optimized for analysis and let expert users build templates for specific types of models.
These templates, which include specifications for the data elements, modeling techniques, quality measures, output formats and deployment destinations, are then made available to nontechnical users – that is, marketers – who can combine them in rules that select different offers for different customer segments.
This approach allows marketers to control most of the recommendation development process while ensuring that technical experts are involved in the areas where their skills are needed. Marketers who want complete independence may not like this, but most would agree that involving experts will yield better results and reduce the potential for error.
The system incorporates about 15 different modeling methods, mostly involving rule induction and decision trees. These were chosen for performance and because they are more easily understood than methods like regression and neural networks. This is important both to give marketers confidence in the models and because European laws sometimes allow consumers to request explanations of model-based decisions.
Rules in DD Series also can prioritize offers in real time for individual customers, based on calculations that determine the long-term effect of making different offers. This sort of dynamic prioritization is more sophisticated than the static prioritization – applying the same priority sequence to all customers – used by many interaction managers.
Another unusual and useful feature lets marketers do simple what-if calculations of the expected value of different rules, using the predicted success rate of individual offers in combination with user-provided cost and value figures. However, DD Series cannot easily manage random splits within rules, making it somewhat difficult to set up tests directly comparing the results of alternate approaches.
DD Series rules also could execute multistep campaigns, but only if users write fairly sophisticated XML scripts. This is not something most marketers will find attractive. The vendor plans to add a graphical rule builder in the near future, which will make the system a more realistic alternative for marketers seeking a multistep interaction manager.
Modeling and campaign-building are the most visible functions of an interaction manager. Still, the real heart of such systems lies in the technology that lets them work across multiple touch points. DD Series technology combines some extreme strengths and weaknesses.
The system is flexible in how it distributes recommendations to the touch points. Users can select offers for a set of records in advance, link to touch points to score individual records in real time or export rule logic to execute selections within the touch-point system itself. The system can communicate with touch points through a wide range of interfaces, including XML, COM, DCOM and MQ Series. In addition, the vendor has built connectors for specific products, including Siebel, Broadvision and Vantive. This adds up to a considerably broader set of choices than that provided by most interaction managers.
But DD Series’ technology to access customer data is less impressive. The system relies largely on real-time queries against whatever databases exist in the touch points or other external systems. This approach allows access to current data, but means performance may suffer if the other systems respond slowly or if data must be gathered and combined from several sources.
Other interaction managers often use the more direct approach of having the touch point send information about the current interaction along with the request for a recommendation. This can be implemented in DD Series but requires custom integration. Similarly, DD Series must read the underlying database to capture other information gathered at the touch point, such as answers to survey questions, or to capture the results of its recommendations. This creates similar problems of time lags and custom integration.
The current version of DD Series also lacks its own customer database, although the vendor plans to add one in the next release. This will store customer profiles derived from other systems and keep internal data, such as the list of offers a customer has already received. It will support real-time scoring and reporting as well as historical analysis of model performance.
The database should eventually let DD Series alert users when results fall below a targeted level and, perhaps, rebuild models automatically as new data becomes available. Currently, models are rebuilt in a batch process using data extracted into the analytical database. This is another contrast with other interaction management systems, several of which do adjust their models automatically as new results are received.
DD Series runs on Unix or Windows NT servers and is accessed through a Web browser. The system was introduced in 1998 and currently has about 15 to 20 installations. These include a mix of Web and call center implementations with as many as 10 million customers and 1,000 call center agents. Clients are mostly large banks and all are in Europe, although the vendor is planning to enter the U.S. market. Pricing begins at about $300,000 and can reach several million dollars, based on the number of customers, channels and users.
• David M. Raab is a partner at Raab Associates, Chappaqua, NY, a consultancy specializing in marketing technology evaluation.
