Appliance Factory

GE database churns out thousands of DM solicitations per week

A WHOPPING 99% of GE Service Management Inc.'s revenue comes from direct marketing solicitations generated by its database. “If direct marketing stops, business shuts down,” says database manager Steve Sheltz.

GE, of course, doesn't want that to happen. Earlier this year, Sheltz brought in Lanham, MD-based Decision Software to design a marketing system from the ground up for the division, which oversees interactions with 50 million GE customers who have purchased 80 million appliances.

Sheltz sees his database as a virtual factory that churns out 680,000 solicitations per week across e-mail, telemarketing and direct mail channels. A given marketing program might include as many as six attempts to contact a customer, and there can be up to 295 variations among the division's various programs during any given week.

But having a database alone isn't the magic bullet, Sheltz warns. The database is a failure if users hate it, won't use it, or if a firm's information technology staffers are the only ones able to use it. “Marketing people don't want to learn [database query language] SQL,” he notes.

Decision Software president Jeff Lawler notes that the program his company designed for GE Service Management hides the complexity of the database from users. Instead, it presents them with segmented data appropriate to each type of campaign, whether a renewal program, a warranty expiration notification or a cross-sale attempt.

Databases are often structured to generate responses to specific marketing questions, adds Sheltz, such as segmenting customers by purchase data or geographic location. They might often not be designed to easily manage the wider variety of queries that may come up as a company's marketing operations change. He offers several tips to facilitate data manipulation.

For instance:

• Marketers should store name and street-level address data separate from the data used in answering marketing queries. These can be linked back to the other data with unique customer ID codes. Street addresses take up a lot of space, bog down the processing time, and — unlike ZIP code and state data — are not used in analysis.

• Don't use “nulls,” such as not coding a gender-unspecific name as being either male or female. Nulls, Sheltz says, are often overlooked when marketing queries are written. Better to leave the field blank.

• Overly complex source codes are difficult to use in data mining operations. A code that has a source, a year, a product and a campaign coded into a dozen-character alphanumeric string can't be sorted or manipulated, unless each attribute is also broken out in other fields on the data entry. “You can't build a query off a piece of a field,” Sheltz says.

• Be ever diligent in applying standards across different databases, cautions Sheltz. A marketer that uses “gender” in one set of records and “sex” in another is going to run into data migration and manipulation problems.

• Finally, be willing to share information and processes with an outside database contractor. “Don't be afraid to give full disclosure,” says Sheltz.

Sheltz and Lawler spoke at the Summer 2001 National Center for Database Marketing conference in Chicago.