Of course I may have some ideas on how to do that, don’t you know.
the Online Customer’s Next Want
By ERIC A. TAUB
Marketers have always tried to predict what people want, and then get them to buy it.
Among online retailers, pushing customers toward other products they might want is a common practice. Both Amazon and Netflix, two of the best-known practitioners of targeted upselling, have long recommended products or movie titles to their clientele. They do so using a technique called collaborative filtering, basing suggestions on customers’ previous purchases and on how they rate products compared to other consumers.
Figuring that out is not so easy. For one thing, people do not always buy what they like. Someone may buy a sweater for their grandmother even though they dislike it and would never get it again. Similarly, a person who rents a movie may actually detest it but knows her child likes it. Or a film that was seen on a small airplane screen may garner a lower rating than if it were seen at a large multiplex.
The search for a better recommendation continues with numerous companies selling algorithms that promise a retailer more of an edge. For instance, Barneys New York, the upscale clothing store chain, says it got at least a 10 percent increase in online revenue by using data mining software that finds links between certain online behavior and a greater propensity to buy.
Using a system developed by Proclivity Systems, Barneys used data about where and when a customer visited its site and other demographic information to determine on whom it should focus its e-mail messages.
For instance, an e-mail message announcing sales might go to those Web site visitors who had purchased certain products or types of products in the past, but who had done so only when the items were on sale. In the simplest terms, if someone buys only when something is on sale, but never buys anything in December, then the e-mail sale flier might not be sent to that customer in December. “There is a digital trail of interest left by customers,” said Sheldon Gilbert, Proclivity’s chief executive and founder.
The observation about sales could be integrated with other behavior. Does the customer buy only when an item reaches a certain price? Is the customer more likely to buy on a weekend or during the week? Must it be organic material? An algorithm would weigh those behaviors to determine the likelihood that someone will open the e-mail message, and once opened, decide to click through to the site and buy the product. The more data, the better it gets at predicting, says Proclivity, which is based in New York.
“One customer found that 10 percent of its population accounted for 60 percent of bargain sales. So on the day of the sale, you can send a full-price ad to everyone else,” said Mr. Gilbert.
Barneys experienced at least a 10 percent increase in online revenue, as compared to control groups, said Larry Promisel, Barneys’ vice president of e-commerce. It found 20 percent more customers would purchase once sent the targeted e-mail messages. The company has saved money by not sending e-mail letters to customers unlikely to buy.
Not only are sales increasing, Mr. Promisel said, but with the store focusing on customers with items they are likely to buy, its clientele feels that it understands their interests, which increases good will.
Still, the problem of knowing what people want is hardly solved. While Netflix has persuaded almost five million subscribers to provide two billion movie ratings to its site, the company still has trouble figuring out exactly what somebody will like.
“I wish I could tell you that our recommendations system was reliable, but it’s not perfect,” said Reed Hastings, Netflix’s chief executive.
At best, Netflix knows that if someone rates a particular drama highly, it can predict what other drama they might like by correlating one’s rating of that film with others. “But if I know your taste in drama, I do not know your taste in horror,” Mr. Hastings said.
As customers value selection and rapid delivery more than recommendations, the company is not that worried about its prediction system. Even a 10 percent improvement of its ratings system has not been possible. Netflix has offered a $1 million prize to anyone who can do that, but to date, only slightly better than a 9 percent improvement has been achieved.
“Using as much information as you can is very important,” said Yehuda Koren, an AT&T Labs researcher, who was part of the group that achieved the results. To do even better, Mr. Koren would “track all clicks, the movies that people searched for, the pages they jumped to, their mouse movements,” information that Netflix does not now collect.
Doing this type of analysis, Mr. Gilbert of Proclivity believes, would stop retailers from sending out buying recommendations based on outdated information.
“I still get e-mails from Amazon recommending books based on the Jared Diamond titles I bought three years ago,” he said. “But I get nothing about my interest in gardening.”