Digging for diversity

If you have ever used a credit card, subscribed to a magazine or signed up for a store discount card, your name is in a database somewhere.

It doesn’t pay to worry much, though. Like government agencies, businesses are much more diabolical in movies than in real life. Experts estimate that only 7 percent of companies are using the information they gather. Even fewer are smart about it — as anyone who has sifted through a pile of worthless junk mail well knows.

But that’s changing. In the past decade, the value of storing customer information has risen as new techniques have emerged to discover patterns and associations in the data. The young science of “data mining” promises companies the power to predict consumer behavior.

Unlike mutual funds — which come with the warning that past performance is no guarantee of future results — people’s past behavior is usually a pretty good predictor of what they’ll do next. If you rented an action flick last Friday and the Friday before, chances are you’ll be stopping at Blockbuster this week for another.

Transaction data — what you bought and when — is a powerful way to predict future shopping decisions, said Usama Fayyad, a onetime NASA scientist who left Microsoft to create Bellevue, Wash.-based digiMine. “The only way you can understand your customer base and serve it better is to crack the code of behavioral data,” Fayyad said.

Though people like to group each other in neat categories based on race, age and ethnicity, data miners delve for connections based on behavior.

“Humans like to think about demographics: young, old, Asian or what have you. It’s easier for the human mind to understand, but behavioral data is a more powerful predictor,” Fayyad said.

The buying preferences of a black man and a Hispanic man who both love steaks and action flicks might be more similar than those of two black men with opposing tastes. “Behavior is the most powerful diversity,” Fayyad said.

But there are a few data mining companies that do look at demographics, including Geoscape International of Miami. General manager Cesar Melgoza said Geoscape helps identify potential customers by race and ethnicity for clients that want to target those markets specifically.

His company’s tools can help provide lists of people by country of origin, ethnicity, race and socioeconomic status. Geoscape looks at the full name of a person and assigns probability-like scores to each portion to figure out country of origin. For example, if a person has a traditionally Hispanic first and last name, he is much more likely to be Hispanic.

But that’s not enough. Next, the name is matched to geography. For example, a man named Vincent Lee can more accurately be classified as Chinese if you know that his neighborhood is home to many Chinese people.

The next step is matching the name against other existing lists of people and businesses that are known to fit certain categories. For example, if a person already subscribes to a publication in Spanish, he can be more accurately grouped as Spanish-speaking.

Direct marketing, telecommunications, financial services and publishing companies are all Geoscape customers. “You want to address people in the way they want to be addressed and the way they will respond,” Melgoza said.

Chris Ragusa, president of Estee Marketing Group Inc. of New Rochelle, relies on Geoscape to connect her clients with people who speak Spanish. Her customers include magazines, music clubs and nonprofit groups looking for donors.

The standard response rate in the direct mail industry is 2 percent. In the Hispanic market, for offers presented in Spanish, a 3 percent return is common, Ragusa said.

DigiMine’s Fayyad said it’s tempting to think you can market to a group just by labeling them “Spanish-speaking people,” but it would be wrong. “The real world is more complex,” Fayyad said.

Understanding your customers is about learning what they like to buy, not which racial or ethnic group they fit into, said Todd Gutschow, co-founder of San Diego-based HNC Software Inc. “It’s not: ‘Are you African-American or Caucasian?’ It’s: ‘Who are you?’ We can begin to treat people based on their individual behavior,” Gutschow said.

Most of the data that companies gather is blind to race, said Robert Grossman, director of the National Center for Data Mining at the University of Illinois at Chicago.

Billing records, warranty cards, coupons, store discount cards, phone surveys and Web questionnaires all help companies paint a picture of their customers that’s less about race and more about income and buying habits.

Persuading companies to actually use the data they collect is an ongoing challenge. “Companies do collect and store a heck of a lot of data, but if you ask them what they do with it, the answer is typically nothing,” Fayyad said. “Companies have been starved for meaningful views into their data.”

Those views can be surprising — as Fayyad demonstrated for Nordstrom. “I’ve never grabbed a large data set and failed to find something interesting in it,” he said.

DigiMine researchers were startled a few months back to see a huge number of people search for “bellybutton rings” on Nordstrom’s Web site. Curious about the odd spike in requests for that particular item, which Nordstrom didn’t stock, managers did a bit of research. It turned out that a recent newspaper advertisement featured a model sporting a bellybutton ring.

Nordstrom decided to stock bellybutton rings both on the Web and in its stores. The “very expensive bellybutton rings” were hot sellers, Fayyad said.

The Web has been a rich field for data mining because shoppers leave digital trails that reveal which products they clicked on and purchased.

In the physical world, retailers like to think they can analyze their customers without the help of computers. But a large retail store might have tens of thousands to hundreds of thousands of items for sale. “It’s very hard for a human being to analyze this data,” Fayyad said. That’s why so much marketing involves cruder strategies such as bulk mailing. Data miners like to call their science the cure for junk mail.

Financial services companies are probably the most sophisticated users of computer technology, including data mining. HNC got its start in the 1980s detecting fraud among credit card users by analyzing transaction data. That’s why today if you make a purchase that’s out of the ordinary, you’re likely to get a call from your credit card company ensuring that it’s on the level.

Using data mining techniques, researchers were able to determine that certain shopping patterns indicate fraud. For example, a very minor transaction at a gas station followed by a very large purchase for a big-screen TV raises a red flag. Credit card companies have learned that thieves will often test a card to see if it works before spending the big bucks.

Today, HNC is using the same techniques to predict behavior. Using a card holder’s master file that contains demographic information such as age, marital status, credit score, income and other factors, HNC can predict whether that consumer might be open to certain offers, said Dan Rich, a senior product manager for HNC.

If a customer has a high incidence of tennis-related spending, he might receive an offer on a special package to the U.S. Open, for example. “People think of it as invasive, but the companies that will succeed will be the ones that provide the services that we want,” HNC’s Gutschow said.

What people fear most is the possibility of merging databases, like comparing credit card purchases with subscription information and bank records. Though most reputable companies shy away from “list aggregation,” as it’s called, it happens.

Customers haven’t rebelled so far because it’s a quid pro quo relationship. A customer gives up some information in exchange for a benefit, such as a discount for flying the same airline again and again.

People will have more opportunities as they are judged on their own habits, Gutschow said. Analyzing a particular customer’s own financial behavior instead of his ethnic group’s could lead to a green light for a loan. “One of the trends that motivated the acceptance of data mining was one-to-one marketing,” Gutschow said. “Just because a group exhibits a behavior doesn’t mean an individual will.”

By: Julie Moran Alterio

Source: The Journal NewsPDF

Leave a Reply