If your web store is getting zillions of hits a month, business must be skyrocketing, right? Not necessarily. Many online retailers monitor visitor traffic as a measure of their stores’ success. But two Wharton researchers explain why high traffic streams to a website need not necessarily mean an increase in business, in a study that they assert is one of the first in-depth analyses of online visiting behavior using Internet clickstream data. In their paper "Capturing Evolving Visit Behavior in Clickstream Data,"
If your web store is getting zillions of hits a month, business must be skyrocketing, right? Not necessarily. Many online retailers monitor visitor traffic as a measure of their stores’ success. But two Wharton researchers explain why high traffic streams to a website need not necessarily mean an increase in business, in a study that they assert is one of the first in-depth analyses of online visiting behavior using Internet clickstream data.
In their paper "Capturing Evolving Visit Behavior in Clickstream Data,"Peter Fader, a professor of marketing, and Ph.D. student Wendy Moe point out that commonly used measures of web store successsuch as number of hits, page views, and average time spent at a siteprovide only generalized information about customers, lumping together everyone from brand-new Internet surfers to serious repeat buyers. In search of more detailed information, the authors scrutinized clickstream data for two powerhouse websites: A leading online bookstore and a popular online CD store from 1998 over an eight-month period. They came up with a unique model of consumer behavior at the individual level, which they say yields a new wellspring of knowledge to help e-businesses manage and market their web stores more successfully.
"A lot of sites right now are looking at their values and saying, ‘Our traffic is increasing, our company is doing great, our long-term future is wonderful,’" comments Moe. "But if you look at it on the level of the individual customer, that’s not the case." Moe and Fader found that individual consumers’ behaviorespecially in a changing environment like the Internettends to evolve over time. As they visit repeatedly, the behavior of site users starts to change. Moe explains, "As people start learning about a site, they start using internal knowledge to make their decisions: They don’t need to actually go to the store to access information. As a result, they’ll go to the store less frequently." Another evolving behavior to take into account, she says, is that "a lot of people are hopping on the Web and surfing just out of novelty. But eventually that novelty’s going to wear off, and people will go to a website only when they need to."
The Moe-Fader model reveals that evolving individual-level behavior patterns appear actually to contradict the conventional larger-picture perspective. Specifically, the summarized data for each of the two leading e-commerce sites they examined seemed to suggest an acceleration in customer visits. Yet their analysis suggests that the typical household is in fact experiencing a gradual slowdown in visiting rate over time. How could this difference be explained? The authors found that an increasing number of new visitors were coming to each site over time, masking the possible slowdown or even dropout of many experienced visitors. This effect, say Moe and Fader, could have dramatic implications for managers who neglect to examine their data at a sufficiently close level: "If such a pattern were to continue, future prospects for the store would appear less promising, especially when the arrival of new users inevitably begins to taper off."
How is it possible to track the evolution of consumers’ visiting behavior in such detail? For their study, the authors mined Internet clickstream data collected by Media Metrix, a New York-based digital media measurement company. Media Metrix maintains a panel of some 10,000 participating households who have installed special software on their personal computers so their Internet behavior can be recorded, pageview by pageview, over time.
Using this data, Moe and Fader were able to glean cross-sectional variations in store-visit behavior as well as changes over time in consumers’ behavior as they gained experience with the two sites. "Thanks to rich new sources of data such as Media Metrix," they point out, "we can now examine behavioral phenomena that would be impossible to study using more traditional sources," such as grocery store scanner data or traditional retail loyalty cards, both of which only record visits that end in a purchase.
Besides uncovering significant facts about the evolution of customer behavior at web stores, Moe and Fader identify a valuable new target market. They state that the most dramatic demonstration of their model’s validity and usefulness is its ability to pinpoint important differences in purchasing behavior across households. Besides confirming that customers who visit a particular online store frequently tend to buy something when they doa maxim that has long made frequent shoppers a prime target segment–their model also confirmed the researchers’ complementary hypothesis: Households in the process of increasing their visiting rates over time are more likely to buy something during a visit to an online store than households that are slowing down.
When you consider these two effects together, say the authors, it becomes clear that households that combine high frequency with an upward evolutionary trend in visiting behavior have dramatically higher conversion rates (the number of purchase transactions divided by the total number of visitors) than all other households. And for e-commerce executives, to whom measuring and managing conversion rates is becoming increasingly crucial, this is an important finding.
Web stores can apply this more refined segmentation approach in many ways, says Moe. Rather than look at the entire mass of visitors who come to a site, e-commerce executives can look at individual visiting patterns and identify people who may be visiting fairly frequently, and at an increasing rate. Those segments of consumers tend to be the most promising buyers, notes Moe. Online retailers should work on getting visitors to come at a more frequent pace, perhaps by offering benefits for frequent visitors or by providing learning mechanisms that encourages them to interact more often with the website.
Web store owners could also use this more closely targeted information when trying to manage online traffic more efficiently. Moe gives the example of a method already used by a popular website. "Victoria’s Secret frequently redirects visitors into either a faster server or a slower server, because they just have too much visitor volume to handle it all. People who are more serious buyers are redirected to a more efficient server. Our model can detect whether or not someone is more likely to buy, and that person can be treated a little differently in terms of how they’re being served."
Moe notes that the model holds up well, too, when it comes to looking ahead: "Our model helps people forecast their web traffic in the future," she says. For both retail sites examined, the model tracks future visiting patterns extremely well, staying within 5% of the actual data during a four-month period.
Moe believes that e-business owners who heed such marketing insights will gain advantages over those who do not. "It’s been easy to get a business going on the Web. The stock market has been pretty good, and investors have been supportive. But eventually–and I think it’s starting now–the market will settle down. And for industries like the book market and the CD market, a shakeout is coming. In order to come out on top, you have to really understand your customers and better meet their needs. Companies will have to better manage their web traffic, and understand when customers are going to buy. Those are the ones that will come out on top."