From Web Visits to Firm Orders

“Our project demonstrates that, at a minimum, firms should keep track of their click data, starting with who is on the website and for how long,” says Jan Van Mieghem, a professor of managerial economics and decision sciences at the Kellogg School, who undertook the study with his Ph.D. student Tingliang Huang. The research also indicates which types of clicks are the key factors in site visitors’ decision-making. That information, the pair concludes, will allow vendors to predict the probabilities that specific visitors will order products featured on the websites, as well as the likely amounts and timing of those orders.

The extraction of operational value stems from the time between potential customers’ visits to non-transactional websites and their resulting placements of orders, or lead time. By gathering data on those lead times—which greatly exceed the negligibly small click-order intervals in e-commerce sites—vendors can use the visits to predict orders and thereby make plans and adjustments before the actual orders arrive.

Such a capability presents obvious business benefits—in controlling inventory or production planning, for example. “Matching supply to uncertain demand is challenging for many firms, and mismatches are costly,” Huang explains. “If vendors can predict future sales better, they can significantly reduce the costs of the mismatches.”

Correlations in Non-transactional Websites

Researchers have extensively studied the correlations between e-commerce consumers’ online behavior and their purchasing propensities. But little has been done to understand those correlations in non-transactional websites, such as those commonly used by business-to-business (B2B) firms. “What we believe is novel is linking the clicks to the operational perspective,” Van Mieghem says. “This is probably the first empirical study of this kind.”

The research is significant because of the popularity of non-transactional websites. “Their use is surprisingly large in B2B sales,” Van Mieghem says. The approach makes sense for many companies selling relatively small numbers of products that must be customized, that require extensive testing by buyers before they decide to make a purchase, or that involve detailed price negotiations before a sale is completed. Many websites of this type contain little more than online versions of their printed catalogues.

The Kellogg researchers set out to discover whether tracking visitors’ clicks on the websites has any value in forecasting future transactions—and if it does, in quantifying the relationship between specific forms of click behavior and sales. “The reason for our analysis was not just to confirm the relation between click behavior and purchasing behavior but also to identify and quantify the key factors,” Van Mieghem explains.

To do so, Van Mieghem and Huang used information on the North American market of a single, anonymous company that sells industrial products globally through its non-transactional website. The company’s CEO—a Kellogg alumnus—gave the pair access not only to click data but also to sales information about customers’ accounts. “That’s a sensitive issue for companies,” Van Mieghem says. “We were fortunate in having such a close collaboration with the company.”

A Variety of Variables

The research involved trolling through the entirety of what Van Mieghem calls “noisy data” to discover the key factors related to future orders. “There’s a variety of click variables you can look at, but only a few seem to be driving most of the information,” Van Mieghem explains.

Based on their analysis, he and Huang reported in their paper that “visitor online click behavior is indeed providing the firm useful information to predict future ordering probabilities.” Specifically, they found that both the frequency of site visits and the number of visits to important web pages correlated in complex ways with the propensity to order products. For example, the longer visitors stayed on the site, the more likely they were to order from it. That propensity starts to decline, however, after a certain length of stay. In addition, the click behavior of new customers differed from that of existing ones.

Detailed analysis of the correlations produced a regression equation quantifying the likelihood that a particular click will lead to a purchase. “The company had been using click striking to detect the ten most important sales prospects they would cold call,” Van Mieghem says. “We are prioritizing each of these customers in terms of the likelihood that they will buy. We can even predict how large a purchase will be and when the purchase is likely to occur.” The analysis also provided insights into website visitors’ strategies—whether, for example, they will decide to buy immediately or to wait in hopes of getting a better deal but at the same time delaying their consumption of the product.

Broad Application Does the result of research on a single non-transactional website have broad application to all websites of that type? “Our findings must be interpreted cautiously, given the limitations of our study,” the researchers note in their paper. Nevertheless, Van Mieghem says, “I’m confident that a similar type of analysis can be done for any site, though the coefficients will differ. I expect the overall picture—our longitudinal analysis—will work.” Indeed, he and Huang believe that vendors with non-transactional sites that have a reasonable lead time between visits and order placements can also benefit from the same analysis of visitors’ click behavior.

Huang summarizes the overall conclusion of the study: “It turns out,” he says, “that click tracking typically brings win-win outcomes for the firm and its customers, especially compared to traditional operations and marketing strategies studied in the literature.”

Related reading on Kellogg Insight

Beating the Bottlenecks in E-Commerce: Effectively allocating Web system capacity

People are Talking: Anticipate product performance through online discussions

Learning from Zillow and Zoots: Improved performance through service inventory management