Using the Right Kind of Data (2,213 words)
Finding your best customers with the right kinds of data
The tumultuous growth of the Internet has shown that catalogers that traditionally relied solely on print media now have new ways of reaching customers and prospects. However, anything new requires some degree of change: The world of the Web caters to customers' whims and offers to satisfy those desires on demand. That means now is the time for catalog companies to get sophisticated about what they sell and to whom. Can you leverage the information in your database to provide customers with more of the products they want at the right times? What about prospects: Are there better ways to determine if the catalog you just spent $800/M to create will end up in the hands of an interested and responsive person?
Unless your company is a start-up with very little history or you have yet to capture one iota of data since the launch of your first book, you can find out a lot about your customers by simply segmenting the records in your database. With the help of proper modeling techniques, you can find your best customers, identify similar prospects and develop campaigns that maximize your marketing dollars.
Start With What You Know
There are so many variables as to why people buy that you need to use the database to its fullest, asserts Betty Camenzind, a member of the marketing department at Oriental Trading Company, an Omaha, NE-based catalog firm that sells novelty items and decorations in bulk to schools, churches, businesses and consumers.
Order a complete work-up on your files. Find out how your customers break down by ZIP code, demographics, psychographics and RFM (Recency, Frequency and Monetary value).
When it comes to relying on data that will be truly predictive for future campaigns, you want to focus your efforts on meaningful segmentation. This means paying careful attention to contact history and products purchased—actual behavior, says Sam Koslowsky, vice president of strategic analytics for Harte Hanks Data Technologies in New York City. RFM is a staple, he says, but it's not as powerful as past behavior.
Chris Lynde, vice president of Experian Direct Tech, Broomfield, CO, explains: If you can build a better picture of your customers and prospects with relevant data available, you can be more predictive of how people will buy from you.
Koslowsky agrees, stating that, typically, a model contains between five and 15 data elements. These concepts apply to all catalog operations, but business-to-business (b-to-b) firms take analysis a step further.
Jen Sprague, vice president of marketing at IMarket Inc., Waltham, MA, explains that b-to-b marketing is different than consumer marketing because "you're not only modeling the behavior of the contact in your database, but also the behavior of that company site. If a particular contact leaves the company, it doesn't necessarily mean that the behavior of the site changes." This leaves b-to-b catalog companies with many data variables to assess.
Go for Help
Outside sources of data on the more than 100 million households across the country can help add clarity to the patterns you've already identified.
You can overlay your database with demographics, psychographics and geographics that will help you better understand the lives customers lead—and hopefully their needs and wants.
For example, a household in an affluent suburb with two working parents and three small children should differ in shopping behavior from a young, single college student, with no kids, who rents an apartment on campus with friends.
At the Oriental Trading Company, sales for the flagship catalog tend to be event-driven, so outside sources of compiled data do little to help Camenzind and her colleagues predict future behavior.
On other titles that offer more general home decoration products, like Terry's Village, the buying patterns are more regular and thus lifestyle data can help analyze trends, Camenzind explains.
Because b-to-b catalog companies don't have psychographic data on contacts, says Sprague, they need other data on their customers' and prospects' make-up, such as number of employees, decision hierarchies, SIC codes, number of offices, etc.
Attaching demographic data to determine whether sites are part of a bigger network can provide a better level of predictiveness on future behavior; for example, companies may purchase products at individual sites or at the headquarters only.
Sharing the Wealth
Another option is to investigate one of the cooperative databases created specifically for the catalog industry. The Z24 database, managed by Experian Direct Tech, offers more than 200 data attributes provided by at least 350 catalog companies and Experian's compiled sources.
When Experian converts a client to the Z24 database, says Lynde, it runs a series of profiles on the company's database—from RFM scoring to regression analysis to profile models. Then, statisticians score the file against the entire cooperative database to see how it stacks up. The result is 35 different reports that serve as the basis, not only for interpreting who its best customers are, but what segments of the co-op to test.
Lynde suggests catalog companies might want to look at key buyer segments, such as first-time buyers. You want to convert these customers into more profitable relationships, but you don't really know enough about their behavior to predict a mailing's outcome. Cooperative database participants can run these names against the master file and append data that provides more insight.
Or, you can turn to trusty demographic and lifestyle data to give you a leg up on how to get the second sale.
Proponents of co-op databases like the access to multiple sources of behavior and transaction data that can give them a better idea not only of who's buying from catalogs, but where and what they buy from the competition. However, one of the biggest critiques of co-op databases is the notion that participants are saturating the same names with marketing offers.
It's been said, defends Lynde, that the average consumer name resides on about 250 files nationwide, from publishers to credit card marketers to catalog companies. Essentially, that means the majority of the names rented are at play in the market.
If everyone is already mailing these names (the case when companies focus on renting multi-buyers), competition within a cooperative database should not be an issue. The objective is to learn as much information about your customers to be able to make them the right offer at the right time, says Lynde.
When evaluating the efficacy of a cooperative, explains Lynde, many catalogers like the biggest files with the highest match rates so they can append more data to their customer records. However, match rates can be tricky in that the data you append can be old and thus not entirely useful.
A better criteria, he advises, might be the recency of data available for appending.
Define Your Priorities
Before you can build a model, you must determine what outcome you would like to reach. Do you want to increase response per catalog or up the average order size? What about reactivating dormant customers or cultivating customers who could be spending more money with your company?
Harte Hanks' Koslowsky explains that the data used for each model depends on the campaign objective for which names will be selected. For example, reactivation of customers could involve a test mailing to inactive accounts to obtain response data. Further analysis will then determine what kinds of people ordered and help build a model to isolate other customers who fit the mold.
Customer behavior changes are certainly important to consider when modeling, says Lynde, to identify opportunities that can be incorporated into future mailing efforts. A catalog company planning the fall/holiday book would want to carve out response data from similar time periods to determine the characteristics of consumers who buy during this season.
This is just the case for Camenzind and Oriental Trading Company. The marketing department has found that customers are specific to each title, so it develops models for each catalog to plan for more productive campaigns.
At the same time, response has been analyzed as seasonal, so the company also uses models for each major season that sees a catalog drop, plus one extra model for the core book. These models earn back their keep in longer seasons, where the company can more precisely determine how many catalogs it can profitably mail to each contact.
You can also develop models to optimize cross-selling between different catalog titles. Clone modeling, explains Koslowsky, assumes you have customers who exist on both files; you could pull out these customers and then model for other customers who share the same characteristics.
So far, the emphasis has been on finding profitable names, but models also help improve a mailing's response by eliminating unprofitable names.
Experian Direct Tech's ABC Scoring System appends information from multiple sources, and scores names, placing them into one of three groups:
• Group A: top group yielding a 30-percent lift in response;
• Group B: middle group yielding an average list performance;
• Group C: bottom group yielding a 30-percent lower-than-average or worse list performance.
To improve response on a catalog campaign, companies are urged to drop names that land in Group C. For example, catalog firm Norm Thompson scored some of what it considered its best prospecting lists of single buyers and found that the "C" group performed 36 percent below the control group of rentals.
Sprague is a strong proponent of cutting the wasteful names in catalog mailings. "Some marketers still have the mind set that you have to mail a certain size drop, say 50,000, every time when a smaller, more targeted campaign would be more successful.
On the b-to-b side, RFM scoring is most useful, especially for companies that sell commodities, says Sprague. Once the database has been broken up into deciles of the best customers to the worst, then the goal is to move marginal customers up from one decile to the next.
"If your best customers are in deciles 0, 1 and 2, then you want to take a close look at what differentiates deciles 3 and 4" into not-so-perfect customers, she notes.
Build the Model
A successful model is only as good as the data used to build it. Having enough data on which to base a valid decision is crucial.
Koslowsky maintains that a test mailing or selection from a file must yield at least 1,000 records for analysis.
"And," Lynde adds, "there must be at least some history attached to a record before we will score them in our modeling process. We typically look for at least two transactions."
With all the data assembled, the model builder can analyze components equally or weight them individually to place emphasis on the most predictive data.
To check a model's accuracy, a statistician can take the response results from a previous campaign and crunch the numbers on only half the file. From this analysis, a model can be built that you can test against the second half of the file. The closer the model's results to the actual response, the better the model, says Koslowsky.
A similar avenue, say Camenzind, is to "back test" the model by running it against all the names from a prior campaign. Those names it identifies as top performers can be compared to recorded results.
Over time, a model can be tweaked as response to each drop is analyzed. At Oriental Trading Company, Camenzind and colleagues periodically take their models' temperatures by dropping campaigns where only half the names have been modeled. If these highly selected names do not pull better than the other half of the file, they know the model is degrading.
She adds that any major alteration in the way you do business, such as a change in your product mix, will spell a need to update or re-create your models.
More Predictive Future
Obviously, the most relevant data for catalog companies comes from their own house files and those of other catalog firms.
However, Lynde feels strongly that marketers can learn a lot about consumers' behavior by bumping their files against a different industry's database.
He cites the case of Smithsonian magazine, which took subscriber files from 1989 and older, cleaned them and passed them against the Z24 database; it then mailed all the matches on the last 24-month buyers and received a 21-percent lift in back-end response.
"In the past, separate entities were created for catalog buyers, magazine subscribers, book buyers, fund-raising donors and more," says Lynde. With the recent advances in technology and the creation of large-scale data warehouses, all relevant data can now be linked together allowing easy access to transactional tables across multiple industries. We've proven there's a direct correlation with many of these transactions and our objective is to now create dynamic environments where all data can be modeled and analyzed to reflect a much more rounded picture of our customers. After all, Lynde says, the more robust your data source is, the stronger your modeling foundation will be.
Camenzind does caution catalog companies to watch out for "analysis paralysis." The computer can only tell you so much about your database before you have to use your intuition and skill as a direct marketer to make sound targeting and mailing decisions.
The average consumer name is thought to reside on 250 files nationwide.