Online retailers have had a tough year. Venture capital has dried up, Wall Street is screaming for profits and the prices of many online stocks have collapsed to a fraction of their highs over the past two years. CDNow Inc., the online pioneer in compact discs, was swallowed up by Bertelsmann AG, the German media conglomerate, after it found that promotions and other discounts aimed at luring customers to make purchases were eating up cash at a dangerously fast rate. Some bricks-and-mortar retailers such as Wal-Mart are re-thinking their Internet strategies or proceeding very slowly. Many still believe that online sites are an economically efficient medium for retail, because they can draw millions of customers to browse thousands of products. But making a sale happen, with a profit and with consistency, has been a problem. Against this backdrop, the study of when and how an online visitor becomes an online purchaser has taken on a new urgency with just about everyone, from corporate managers in department store chains to hotshot web-site developers. Getting a person to visit an online site isn’t enough anymore. Converting them into purchasers is the acknowledged endgame. “Everybody buys into the importance of conversion rates but everybody is wringing their hands over why they are so low,” says Wharton marketing professor
Fader, in a research paper co-authored with Wendy W. Moe of the University of Texas titled “Which Visits Lead to Purchases?: Dynamic Conversion Behavior at E-Commerce Sites,” analyzes conversion rates based on visit and purchase data at Amazon.com. The researchers conclude that generating an initial purchase at an online site is relatively easy. Subsequent purchases, however, are not so easy as shoppers become more familiar with shopping at a given online retail site. Over time, a shopper’s tendency to buy again declines as the sheer novelty of purchasing at that website wears off. This is a surprising – and very troubling – development for online retailers.
Realizing this dynamic, Fader and Moe developed a “conversion model” that breaks down the buying process into three components: (1) the shopper’s “base rate” of purchasing during his or her first visit in search of a particular product; (2) the accumulating effects from non-purchase visits (i.e., browsing sessions) that help drive the customer towards an eventual purchase; and (3) the threshold of resistance that the customer must overcome before making the commitment to purchase something at a given site. All three of these behavioral elements vary across customers, and the latter two also vary over time. Thus, for instance, the purchasing threshold may come down as a customer commits to additional repeat purchases at one site.
With this behavioral decomposition available, the web-site manager can get a more accurate reading of the basic retail efficiency of the site and, in real-time, pinpoint the reasons why a site is attracting shoppers or why it might be turning shoppers away. Adjustments such as website design elements or well-placed marketing messages can be used to improve the site’s weak links.
In practical terms, web-site managers can use the model to identify online visitors who may finally be enticed into the check-out aisle with a promotion, such as a pop-up coupon. This allows retailers to spend their marketing dollars with laser-like accuracy, instead of the carpet-bombing marketing approach that has been used the past two years. The model also can be used to “manage traffic” in a web site by swiftly identifying a visitor who is about to purchase an item and ushering him or her quickly to a high-speed server.
With an estimated $11.6 billion in expected purchases this holiday shopping season, according to Jupiter Communications, the online companies who can most wisely and efficiently spend their marketing dollars and manage their traffic will be the strongest competitors.
“We show that that there is a clear, but highly complex relationship between visit patterns and purchase patterns,” says Fader, who emphasizes that the model is built upon solid theoretical grounds and does not simply look at numbers. “This is not a data-mining expedition.”
Fader and Moe’s research uses so-called “clickstream” data i.e., the sequence of every web address (URL) that each shopper visits over a period of time, to generate their conclusions about shopping behavior. But Fader says the basic behavior patterns that the conversion model is built upon are not new; they have been identified by other researchers in other contexts that usually required extensive survey work to uncover them. For instance, the two professors have divided online shoppers into four categories whose definitions could be easily transferred to bricks-and-mortar stores. They are the directed-purchase shopper who knows what they want to buy and won’t waste time doing it; the search-deliberation shopper who wants to buy something but is proceeding cautiously; the hedonistic browser who does not have a specific purchase in mind, but indulges in the joys of shopping and may, if the circumstances are right, buy something; and the knowledge-builder shopper who is gathering information for an eventual purchase.
What’s different now, though, is the ability to closely and unobtrusively follow a person as he browses and actually buys online. “From a behavior perspective, this is nothing new,” Fader says. “In the past, we knew these different patterns were occurring, but we could never really see them in places like the mall.” He calls online clickstream data that gathers individual clicks on a web site “putting an electron microscope” to retail behavior.
The Fader-Moe research is based on an analysis of more than 4,000 households who visited Amazon.com over eight months in 1998. The clicks were gathered by Media Metrix Inc., one of the leading online research companies. In sum, Amazon was highly successful capturing that first online purchase, possibly because of the novelty of shopping online. But the “threshold” for repeat purchasing increased sharply as the shopper became familiar with the site, contrary to the predominant beliefs held by e-commerce managers. (Nevertheless, Amazon continues to enjoy very high conversion rates – generally over 10% – in sharp contrast to most of its competitors.)
“Amazon and CDNow are now starting to grapple with the problems our model foreshadowed ,” Fader says. “The premise that ‘if we build it, they will come’ is only partially true. They may come in droves, but they won’t buy anything very often.” Since the data is two years old, Fader says some of the parameters related to the research have probably changed. But, he adds, “today’s (online) behaviors are the same, just in different proportions.”
Fader has been studying visit patterns at online web sites for several years. The same problems confronting retail web sites – making visitors commit to a purchase – are being experienced by other web sites such as content providers and portals. His is the first paper in the marketing literature, he adds, to offer such comprehensive linkages between these types of “profitable commitments” and the visit patterns that lead to them.
