Database Marketing: Transactional Modeling for Beginners

Transactional data can reveal that two seemingly similar customers have strikingly different purchasing patterns

U.S. marketers mailed an estimated 14 billion catalogs last year, at an average cost of $.70 each. But 98% of those catalogs did not result in a sale. • So why do catalogers mail so many books? For one thing, for catalogers that are new to the industry, just getting down the basics of RFM, (recency, frequency, and monetary) may be challenging enough. In fact, only 35% of participants in the Catalog Age Benchmark 2000 on Database (April 2000 issue) perform RFM modeling.

But even those marketers that use RFM should realize that it’s not necessarily enough to generate the best results. Although this industry standard can yield a wealth of information about the purchasing history of your customer base, used alone RFM doesn’t give you the big picture of your customers’ overall purchasing behavior, across a variety of categories, and across many catalogs.

Actions speak louder than words

Transactional data based on actual purchase history can give you a clearer picture of customer behavior than data based on one purchase or on survey information. It’s important to remember that what consumers say they will do is not always the same as what they will do.

A consumer may have filled out a survey in which he checked off a box indicating that he likes to play golf. And perhaps he does like to golf, but he rarely has the time to do so anymore and certainly doesn’t need any more equipment. Nonetheless, his name was put on a file of consumers interested in golf, and golf-related marketers began renting his name. Soon he was awash in golf-related offers, which may have piqued his interest at first, but he didn’t buy from them. And after receiving them repeatedly, he began to toss them into the trash almost immediately.

Such is the drawback to compiled lists. Compiled list data can work well for niche marketers targeting people with lifestyle changes, such as new parents. But often the information on lists gathered from survey and credit-card information, product registration and warranty cards, and public records can be improved with enhancements from a transactional model. Had the marketers in our example known that the golf-loving consumer hadn’t made a single golf-related purchase from a cataloger in years, they would likely not have wasted their money mailing to him.

Transactional modeling uses purchasing information gathered into a cooperative database from a number of marketers. Because the data are based on transactions the consumers in the file have actually made — when they bought, what they bought, from whom they bought — they are generally more predictive of what those consumers will do than nontransactional information. And because the data are culled from numerous marketers, purchase aberrations are easier to detect.

Let’s say a consumer buys an inexpensive gag gift from a catalog. She will no doubt soon receive other catalogs selling low-end gag gifts. These mailers may have figured that renting the first catalog’s buyers file provided enough transactional data. What they didn’t know is that our consumer bought the gag gift as a one-time-only purchase on behalf of her nephew. All of her other catalog purchases during the past few years had been from high-end furnishing and bedding catalogs — which would have been more obvious had the low-end catalogers that rented her name been able to view her relationships with a multitude of marketers rather than with just one.

Not for prospecting only

Using transactional modeling provides many benefits in prospecting, but you can use transactional modeling to improve your house file results as well. Product information, amount spent, and multipurchase behavior are key metrics when choosing which names from your customer file to mail to.

But these internal selections tell only part of the story. House file modeling based on transactional data adds valuable information to your existing customer information to give you a total picture of a buyer’s purchase behavior across multiple channels.

For instance, two of your customers may look exactly the same on your house file. A profile of their previous purchases within your database might reveal that they both live in an affluent zip code, are first-time customers, and spent $100 on apparel from your catalog two weeks ago. But add transactional data from a co-op database to your records and you might learn that one of the customers spent $1,000 on apparel from catalogs, Websites, and retail within the past six months; spent $300 in gifts during the past year; and purchased $600 worth of home and garden merchandise. The other customer, however, bought only $200 worth of clothing during the past six months. The information can help you decide how frequently to remail to one customer vs. the other.

In short, transaction data reveal behavioral details regarding:

  • what your customers actually purchase vs. what they say they purchase.
  • when they make their purchases. For instance, are they primarily holiday shoppers, or do they shop year-round?
  • how much they spend on those purchases.

You may be able to improve your results by basing your marketing decisions on information compiled from the most profitable households’ purchasing behavior across categories. As you plan for the coming year, consider transactional databases as a tool to distinguish your customers’ buying patterns and enhance your marketing strategies.


Brian Rainey is senior vice president/general manager of DoubleClick’s Broomfield, CO-based Abacus division, which manages the Abacus Alliance database.

Who are the industry’s stars of tomorrow?

For an upcoming article on cataloging leaders of the past, present, and future, we need your input.

Fax your picks to 203-358-5823, or e-mail them to [email protected].

Deadline is March 30.

As for the Costs…

Cost is a straightforward aspect of transactional modeling; the cost of using the typical Abacus modeling product is $70/M. But when it comes to analyzing the benefits of modeling, hard-and-fast numbers don’t exist. The resulting lift in response rates varies widely, depending on the size of the file and the parameters set by the client: A mailer in the prospecting mode, for instance, may be willing to take a loss to acquire names, and therefore may establish more liberal criteria. Then, too, the client’s previous results and methodology will influence how much of an improvement it will see.