Catalog Tech: Campaign Management Software–aCase Study

Jun 01, 2003 9:30 PM  By

Catalogers looking to implement a customer-centric database marketing program (and who isn’t?) face two big hurdles: consolidating transactional data into a usable marketing database, and finding the most effective way to leverage their database resources. Creating the unified database structure can take the form of a comprehensive data warehouse or several smaller, more focused and more manageable “data marts” consisting of subsets of enterprise data.

In either case, you need to optimize the database to serve the needs of both an analytical team and a marketing department. This most often requires consolidating multiple sets of transactional data (for orders, cancellations, lost demand, items shipped, returns, payments, and so on) with some form of data summarization and reconciliation (to bring together data from different accounting periods and different sales order systems, for instance).

Of course, in most cases, the transactional system doesn’t take care of all this; you do — or more precisely, a “middleware” platform that you implement handles these chores.

The second big challenge — actually doing something with the data — involves a division of labor. Typically, one or more highly trained statistical analysts work with the database to perform a variety of analytical exercises for modeling the data and segmenting the customer file. Finally, when this has been done (though technically it is an ongoing process), the marketers can get down to the business of creating a series of promotional efforts and campaigns.

Inevitably, marketers have questions about the customer file that can’t be answered without further manipulation of the database. In the worst-case scenario, a marketer will pass a query back to the analytical team, who will then undertake the analysis and pass the answer back to the marketer. Obviously, the easier it is for the marketer to perform the analysis independently, the better.

Lands’ End’s choice

When apparel cataloger Lands’ End decided to develop a unified view of its customers in the late 1990s, it had already begun to address the first of these challenges with an enterprise data warehouse. But with five catalogs, a growing Website, corporate sales, and several other sales channels, all organized as separate business units, the Dodgeville, WI-based marketer still had several loose ends to account for. In fact, in addition to the data warehouse that the cataloger had created itself, Lands’ End had a separate IBM DB2 database for maintaining promotion history and was relying on flat-file extracts from an IBM mainframe for managing some of its customer data.

In assessing its situation, Lands’ End decided that it did not have the comprehensive multichannel view of the customer that it desired. Segmentation was an iterative or repetitive process that slowed down campaign management, and there was too much of a disconnect between customer segmentation and promotion targeting. Compounding matters, the design and execution of marketing campaigns required high-level analytical skills. It was cumbersome to manage a hodgepodge of sometimes conflicting business rules in order to interpret the data. And the technology platform in place was not an “open systems” environment — in other words, it wasn’t universally compatible with other systems.

After doing a thorough needs analysis for campaign management software (using both inhouse IT resources and outside consulting services), Lands’ End sent a request for proposal to 11 solutions providers. The five finalists were scored on functionality, implementation methods, support structure, financial viability, customer references, and industry rankings.

The winner: the Affinium Campaign and Affinium Model suite from Waltham, MA-based Unica. “Ultimately, we were looking for business functionality in a system designed for marketers, not programmers,” says Lands’ End vice president of direct marketing David Johnson, who was involved in the selection process.

What makes a winner

One of Unica’s strengths is its ability to access data from a wide variety of legacy data structures. Unica’s universal dynamic interconnect (UDI) tool, bundled with the system, is a “metadata” middleware platform that allows database administrators to map virtually any data source and import it into the system. As a dynamic interface, the UDI supports the integration into any campaign of last-minute suppression lists, opt-outs, or new customers or segments not stored in the enterprise data mart. The dynamic nature of the UDI means that it relies on user-definable and editable rules rather than on hard coding by a programmer.

The UDI eliminates the need to set up a specific, proprietary application data mart. In fact, there are no proprietary platform requirements for any components in Unica’s open-system suite, which runs on any UNIX or enterprise Windows platform. The system Lands’ End implemented uses an IBM DB2 database in conjunction with Affinium Model and Campaign modules running on an IBM RS-6000 AIX (UNIX) server platform with a two-tier structure relying on Windows-based user client workstations.

The Unica suite allows quantitative analysts to spend most of their time on such strategic database functions as data mining and segmentation modeling, giving marketers full responsibility for campaign planning and design. Affinium Campaign has a flow-chart-style interface for creating, assigning, documenting, and sharing campaign plans in a comprehensive planning environment that consolidated what had previously required juggling of multiple components.

With Affinium Model and Campaign, customer segmentation has become a much more flexible process for Lands’ End. The company now has more data to work with from more sources with fewer manually driven, hard-coded procedures to follow. Best of all, the system supports the creation of much more targeted marketing strategies and more automated campaigns that help to highlight productive segments and suppress offers to underperforming customers.

Automated multiwave campaign strategies also support more sequential campaign “triggers” for stream-based or “longitudinal” campaigns requiring a series of contacts — for instance, a phone call to follow up an e-mail, and a personalized letter to follow up the phone call. The choice of e-mail would determine the script for the phone call, and the phone call script and results would determine the specifics of the letter. The system also supports highly customized and individualized mailings that have reduced overall production costs and enhanced revenue per piece mailed.

On the modeling side, where the statistical analysts live, Affinium Model consists of four components:

  • Customer Valuator, which predicts RFM and profitability by customer over time;
  • Market Segmenter/Profiler, which identifies significant dependent variables in customer and transaction data;
  • Response Modeler, which does the actual modeling to predict how customers will respond to specific promotions; and
  • Cross-Seller, an “affinity modeling” tool to predict which products a customer will buy next based on purchase history.

A “wizard” walks you through setting up the modeling process, using automated sampling and selection plus normalization of data. The system automatically applies a range of statistical tools such as genetic algorithms, stepwise search features, and cross-validations to determine the most accurate type of model for a specific campaign.

You can specify quick, intermediate, extensive, or custom modes for search methods and functions. Statistical methods incorporated by the system include linear and logistic regression, RFM analysis, back-propagation neural networks, CHAID and CART decision trees, Naїve Bayes, and K-means with leader clustering. (The last two, which use probabilities as well as correlations to predict results, are less well known than the others but are becoming more popular because they require only one pass through a large database to produce results, rather than multiple passes.)

Once built, models can be deployed and customers scored within the system or by using UDI-supported export options with either C or SAS source code or the Unica Data Mining Engine, an application programming interface for embedding Unica production models into other applications. Unica provides a two-day training course for each of its components, as well as customized and advanced training sessions.

The Affinium Implementation Methodology is run by Affinium or an implementation partner, with systematic assessment of objectives, a formal data integration process (to set up the UDI, among other things), and assistance in creating and implementing an initial campaign.

In Lands’ End’s case, there was an initial 14-week pilot project using live data that was tied to a final “product acceptance” phase of implementation.

Unica’s applications, which also include Plan and eMessage, cost nearly $100,000 per module (depending on specific configuration). While any application, like Model or Campaign, can be run alone, most Unica users start out with two applications and add others over time.

You don’t need a lot of return to justify that kind of investment. But what you do need is a commitment to making use of the tools once implemented. A suite like Unica Model and Campaign will not only make your current work much easier but will also enable you to pursue new approaches to multichannel database marketing. That’s where the real payoff resides.


Ernie Schell is president of consulting firm Marketing Systems Analysis, based in Southampton, PA.

Other Options

Unica Corp. (www.unicacorp.com) isn’t the sole provider of campaign management software and related analytical modules. Nor might it be the best for your company. Here are several other providers to consider if you’re in the market for such a system:

Aprimo (www.aprimo.com)

Doubleclick (www.doubleclick.com)

E.piphany (www.epiphany.com)

Marketing Pilot (www.marketingpilot.com)

NuEdge (www.nuedgesystems.com)

Oracle (www.oracle.com)

PeopleSoft (www.peoplesoft.com)

Pivotal (www.pivotal.com)

SAS (www.sas.com)

Siebel (www.siebel.com)

Yellow Brick Solutions (www.yellowbricksolutions.com)