RIGHT MODELS FOR E-COMMERCE -- CORN FLAKES & KETCHUP (October 25, 1999)

The remarks are based on a paper, "Repeat Buying in Cyberspace," that is being presented at a convention of the Institute for Operations Research and the Management Sciences (INFORMS®) in the Philadelphia Marriott Hotel on Monday, November 8 at 8:15 AM. The authors are Peter Fader, Associate Professor of Marketing at the Wharton School, and Bruce G. S. Hardie, Assistant Professor of Marketing, London Business School.

"Whether you're dealing with a web site or breakfast cereal, the dynamics are very similar: People shop, many buy again, but most eventually buy at a slower rate than they did during that honeymoon period," says Prof. Fader.

"In e-commerce right now, there's a constant inflow of new people and sales are going up. But if you track individuals, repeat purchases are a lot lower than what Wall Street is betting on. We're not at a point of unlimited high growth. We never will be. I believe that the 'bubble' that everyone is talking about may not burst (as some suggest), but it will almost surely slow its expansion and perhaps even contract over time as newcomers to the Web slow down."

Web Retailers Don't Analyze Data
Using the techniques of operations research and management science, Professors Fader and Hardie previously developed probability models that forecast sales of packaged goods at supermarkets. When they turned to e-commerce, they discovered that the earlier models' capacity to forecast based on electronic checkout information worked well with e-commerce sales data collected by such companies as Media Metrix. They have examined Media Metrix data from dozens of Internet sites, including Amazon.com, Barnes & Noble, CDNow, and Expedia. In addition, they have been working directly with several major Internet retailers.

Web merchants, despite their technological advances, lag behind more established retailers in one important area: Fader and Hardie say that most of the e-commerce companies they have observed aren't benchmarking or using the right types of analytical tools to discern and project buying behavior of new customers and repeat purchasers. The insights generated from many of today's e-commerce reporting services are akin to what packaged-goods manufacturers had at their disposal in the early 1960's.

"Only the more sophisticated websites are looking at factors like timing, choice, and counting -- factors that traditional retailers have examined for years," Fader says. "What many people fail to acknowledge is that, in many cases, the Internet is just another distribution channel -- albeit one of growing importance -- and thus it is natural that the same tools should apply to it as in the "bricks and mortar" world.

Prof. Fader says that nearly all web-based companies need to do a better job of decomposing their data into a series of related questions, such as: how long an online visit will last, when the visitor will return, if the visit will lead to a purchase and repeat purchase, and how much is purchased in terms of dollars and number of items. Answering each question separately provides a more informed view of consumer behavior -- and better forecasts -- than looking at the overall level of sales by itself.

Not all models work equally well, says Prof. Fader. "A naïve model may look at a customer who has visited a site five times and bought twice and conclude that that person is more likely to buy again than someone who has bought only once in ten visits," he says. "But this could be wrong. That second shopper has been coming back and looking. Perhaps there's a threshold, perhaps after shopping a few times the person will begin buying. Businesses don't give consumers enough credit for the types of browsing and information search that they require. You need benchmarks and good models to predict and understand such complex behavior."

Fewer Repeat Purchases
Predicting growth in e-commerce should be done cautiously, say the authors.

"A web business leader may say that 50% of people who buy today will be back to buy again sometime this year and claim that that's a good sign," Fader observes. "But that rate will decline. So in the second year, maybe 50% of that 50% will return. This kind of attrition occurs in an exponential manner. It's shocking to see how few people who bought at a site are still active customers one year later."

Two years down the road, he says, repeat buying rates at websites will typically be no higher than 30% of the initial triers. This rate is similar to purchase rates among packaged goods at the supermarket.

"That's my point," Fader says. "The only difference between the two is that you have lots of new people visiting a website every day. If you could get that rate of new people entering the supermarket, you would see sales of ketchup going sky high. Perhaps the valuation for Heinz would be as dramatically inflated as that of Amazon.com.

"It's ludicrous for web retailers to think that their rates will increase at that kind of rate forever," he says. "People are just not facing that reality yet."

Philadelphia Convention
Members of the Institute for Operations Research and the Management Sciences (INFORMS®) will hold a national convention place at the Philadelphia Marriott Hotel from Sunday, November 7 to Wednesday, November 10. The theme is "ORMS and the Quality of Life." More than 1,600 papers are scheduled to be delivered. Additional information about the conference, including a full list of workshops, is at http://www2.informs.org/Conf/Philadelphia99/ and http://www2.informs.org/Press.

The Institute for Operations Research and the Management Sciences (INFORMS) is an international scientific society with 12,000 members, including Nobel Prize laureates, dedicated to applying scientific methods to help improve decision-making, management, and operations. Members of INFORMS work primarily in business, government, and academia. They are represented in fields as diverse as airlines, health care, law enforcement, the military, the stock market, and telecommunications.