David Raab Article Archive

Today, most firms building a marketing database want an “open” system–which in this context, usually means a standard relational database management system like Oracle, Informix, or SQL Server. The primary motivation is a desire for compatibility with other corporate systems, which themselves typically use these databases. Building the marketing database on the same platform makes it easier to share data, query tools, update processes, hardware and support staff. As corporate systems departments play an increasing role in marketing system development–often fitting the marketing database into a larger structure encompassing some sort of data warehouse–these considerations weigh more and more heavily. At least from the systems department’s perspective, they overcome the significant performance advantages offered by non-relational “proprietary” products like Customer Insight AnalytiX, Harte-Hanks P/CIS or MegaPlex FastCount.

But it turns out that proprietary technology is not quite dead. The inconvenient fact still remains that conventional relational databases are singularly ill-suited for the complex analytical queries and hierarchical segmentations required by most database marketing applications. This has lead to a quiet resurgence of proprietary techniques–tucked inside of “open” solutions, where they give a performance boost without sacrificing over-all compatibility. Specifically, these systems use a conventional relational database to store their data, but load it into proprietary structures for actual manipulation.

Some vendors, including Mercantile Software Systems and Retail Target Marketing Systems, load the entire database (or selected elements) into their proprietary structures, and do nearly all their queries, selections and reports in the proprietary portion of the system. Other vendors, including Marketing Synergy, create new extracts each time a group is selected for a promotion. They then do the final analysis, segmentation and manipulation on the extract.

One-by-One Campaign Manager (Paragren Technologies, 703-758-1090, www.paragren.com) also does its final segmentations on an extract from a relational database. The extracts are created using RACER, a tool developed by Paragren to execute very complicated extractions in a single pass through a relational database. RACER lets the system administrator define “metadata” that includes both elements in the underlying relational database and derived elements defined through complex calculations or conditions. End-users create file segments by selecting a series of elements from the RACER metadata list, and defining the value ranges and other logical conditions that each element must meet. This approach shields the end-user from writing traditional Structured Query Language (SQL). But it also means that complex conditions and calculations can only be included if the administrator has defined them as metadata elements in advance.

RACER data extracts can be created as flat files, star schemas, aggregated relational data sets, or preloaded files for Oracle Express, MicroStrategy DSS Agent and other multidimensional databases. The files it creates for One-by-One Campaign Manager are in still another format, which stores compressed attribute data for each customer. These files are loaded into the Campaign Manager’s campaign definition module.

This module creates a matrix showing file segments down the side, promotion packages across the top, and the number of names allocated to each package for each segment. The allocation is done automatically. The matrix will also show a target output quantity per segment as defined by the user.

After the initial matrix is created, the user can eliminate some segment/package combinations, change the allocation within segments, or let One-by-One create an optimal allocation based on model scores, expected response rates, or other values. Names that qualify for multiple segments have a single record in the extract file, so they never appear as duplicates. The system will ordinarily assign each name to the first segment it qualifies for in the matrix. However, users have to option to place several segments in a “priority group”, and to have duplicates distributed randomly among segments within each group. After the user makes adjustments to the matrix, Paragren will recalculate the quantities in each cell of the matrix, working against the extract file at up to 1 million records per minute. This is much faster than working against the relational database directly.

Users can also split a campaign into multiple drop dates, define campaign performance measures, and specify up to 64 digits of attribute data to appear as a source code on customer records when they are output. The output itself is created through another RACER extract, based on the record allocations specified in matrix.

The system stores reference information at the campaign and package levels. By December 1997, it will add facilities to support multi-step promotions and to attach contents, such as letter text or a telemarketing script, to each step in a promotion. contact strategies, is scheduled for December release. The December release will also include a job scheduler to execute scheduled promotions automatically and a module to simplify data extracts and importing of model scores.

One-by-One provides some standard response analysis reports, but largely expects the user to do reporting in third-party multidimensional tools, loaded with databases created by RACER. The system does not allow users to browse the names and addresses of records in a selection.

Paragren’s products run on Windows 95 or Windows NT workstations and on applications servers can be Windows NT or Unix. The relational databases can reside on any platform that allows an Open Database Connectivity (ODBC) connection. The databases themselves are created outside of the system, although Paragren offers design and development services. On a mid-range server, RACER will extract from 350,000 to 750,000 records per minute.

Pricing for One-by-One is based on the server and number of users. The first-year cost for an entry-level, six-user license would be about $300,000, with 20% maintenance in later years.

Paragren was founded in 1995 by principals who were previously associated with the Express multi-dimensional database product. It released One-by-One in 1997, after beta tests with large clients in telecommunications, financial services and utilities. The system has been installed at six clients and fifteen sites. Paragren was purchased by APAC, the large telemarketing firm, in mid-1997.

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.