Readers of both the popular and trade press are regularly bombarded with high expectations for online consumer shopping and buying. For as they seek publicity, research suppliers rush to outdo one another by broadcasting key optimistic findings. Forrester Research predicts that by 2004, 49 million U.S. households will spend $184 billion dollars online. e-Marketer predicts 67.2 million Americans will purchase goods and services on the Internet by 2002. Jupiter Communications foresees that 85 million Americans will spend $78 billion online in 2003.

n addition to being large – as a reference point, Forrester estimated that just 17 million households bought online in 1999 – online shopping estimates and forecasts vary greatly. The Boston Consulting Group estimates that 1999 U.S. consumer online revenues totaled $36 billion. But the Direct Marketing Association pegged the same quantity at about 11 percent of that total: $3.9 billion.

Why do the forecasts of different research vendors vary so much? And what must happen in the business and consumer environment for these extraordinary growth projections to be realized?

Some of my recent research has been directed at answering these two questions. I’ve found that variation in forecasts can be attributed to variations in the approaches used to obtain them. However, analyzing these differences is very difficult, since suppliers do not always provide details about how they obtained their forecasts. Furthermore, I believe that the extraordinary growth projections cannot simply be achieved with a continuation of existing trends.

Why do the forecasts of different research vendors vary so much?

Analysts produce forecasts by applying a specific methodology (or a group of methodologies) to a set of data under a certain set of assumptions. Methodologies, data, and assumptions can vary in numerous ways.

The wildly optimistic online shopping forecasts probably do not take into account the impending plateaus of PC and Internet usage.

Judgmental methods – like expert judgment – are probably the most common forecasting methods in use today. Experts know a tremendous amount about a specific market and bring all their cognitive resources to bear on the problem at hand. Of course, each expert has an idiosyncratic set of experiences and attitudes. One may be more familiar with online book sales; another may be more familiar with online travel. One may have a tendency towards optimism; another has a tendency towards pessimism. (Viewers of financial news networks like CNBC will recognize certain pundits as congenital bulls or as perpetual bears.) As such, comparing the news releases of different online experts can be akin to comparing e-apples and e-oranges.

any market analyses are made via customer surveys. These surveys are centered on a question of the form “Do you intend to….?” While it is likely that a plurality or majority of people responding in a positive manner to such a question is indicative of a bright future for the concept, research has shown that people are notoriously bad at predicting their own future behavior. So when research suppliers use intentions data to make precise forecasts, they are likely overestimating true results.

Time series analysis involves the extrapolation of historical data. However, forecasting the adoption of a phenomenon as new as online shopping is very difficult because historical data are very sparse.

Causal methods express demand as a function of a set of potential causal factors. Under these methods, data are plugged into statistical procedures, which produces a forecasting model. Such a model can then be used to forecast demand if the future values of the causal factors are either known or could be forecasted in another way.

Obviously, different methodologies can lead to different forecasts. In addition, practitioners employing different meth-odologies may use different means of processing data. After all, research suppliers generally do not open their methodologies to public scrutiny, presenting another problem for objective analysts. Some may provide a flavor in their web sites and promotional literature. For example, Forrester describes its methodology as being based on interviews with consumers and business executives. eMarketer claims to enter data from a wide variety of published, publicly available sources into a proprietary aggregation model. Other companies, including the Gartner Group’s DataQuest division, are much more secretive. Repeated telephone calls to Gartner failed to produce an informative response.

These companies argue that their competitive advantage depends on their methodologies remaining proprietary. This is unfortunate, for it is impossible to evaluate a forecast without knowing the methodology and any assumptions it harbors in detail. Even those companies that hint at their procedures do not give sufficient detail. Forrester does not explain how it arrives at a projection of $184 billion spent online in 2004. e-Marketer does not post its proprietary formula. Gartner tells us nothing.

A significant feature of online buying is that it is contingent on other innovations. One cannot buy online without being online.

Do companies need to keep their methodologies confidential to be credible? Not really. Even if a methodology were to be completely revealed, a competitor would still need the raw data used by the research supplier in order to duplicate the supplier’s forecasts. Research suppliers can produce proprietary forecasts by maintaining proprietary data.

There’s another reason why suppliers should explain their methodologies more fully. Most suppliers have an economic interest in making optimistic forecasts. These firms all either sell other research or provide consulting services to Internet commerce companies. Forecasts of huge volume and high growth not only attract attention, they may recruit new players into Internet commerce and hence increase the universe of potential clients. But for any forecast to be credible, the methodologies, data, and assumptions behind it must be beyond reproach. Unfortunately, clients and investors have not made these firms accountable.

nsatisfied with the forecasts provided by the usual suppliers, I decided to develop model-based five-year “forecasts” of online shopping and buying. My forecasts are not based on what people say they will do or on some secret proprietary model. Rather, I applied standard time-series-based marketing methodology to U.S. Census data and survey data collected from people who describe what they have already done.

I further focused on forecasting the number of people participating in online shopping, as opposed to sales revenues, for two reasons. First, sales revenues comprise the number of people and how much they spend. By focusing on a simpler construct – the number of people – I have a better chance at success. Second, marketing scholars have a standard time series-oriented technology for forecasting the number of people who have adopted an innovation: diffusion models.

Diffusion models produce a lifecycle curve for a particular innovation – be it the microwave oven, the cellular phone, or the Internet. The premise behind these models is that an innovation is adopted by a small, select group of adopters in the population based on mass media communications. These adopters, called innovators, then influence others to adopt via word-of-mouth. As time goes on, and more people adopt the innovation, all non-adopters are subject to the same type of word of mouth. This process continues until all members of the population who will eventually adopt the innovation have done so.

significant feature of online buying is that it is contingent on other innovations. One cannot buy online without being online. Furthermore – recent technological developments notwithstanding – one generally cannot have access to the Internet without having a personal computer (PC). The diffusion approach allows for interrelated innovations. The models I developed allow for each of the variables – PC access, Internet access, and online shopping – to impact and to be impacted by each other.

The data I used come from two sources. The U.S. Census Bureau captures data on PC and Internet access. (See: www.census.gov) A private Internet marketing research firm, Cyber Dialogue, provided data on PC access, Internet access, and online shopping. Cyber Dialogue’s methodology, in contrast to other suppliers, is described in complete detail on its web site (www.cyberdialogue.com). It involves multiple, random-digit-dialed surveys per year. As such, the company surveys both users and non-users of the Internet. Its data on PC access and Internet access are remarkably consistent with those of the Census Bureau.

So what did I find? Well, my study produces the following generalizations:

1. By the year 2004, at least 60 percent of those U.S. residents having access to PCs will have bought over the Internet.

2. Online consumer purchasing in the U.S. will grow about 1/3 over the next year or two and 20 percent for a year or two after that. This amounts to a 150 to 200 percent increase in the number of online buyers within five years.

3. Personal computer and Internet access are approaching plateaus in the United States.

4. Within five years, 90 percent of the population with PC access will have Internet access as well.

The first generalization above suggests that more and more Internet users will use the medium for shopping. The second generalization suggests that this growth will indeed be considerable. It is in line with Forrester’s projected growth for the number of households participating. But my number significantly lags the Gartner Group estimate that online purchasing will rise from $20.5 billion in 1999 to $147 billion in 2003 – a 600 percent increase in four years!

I have a challenge for research suppliers whose projections are even more optimistic than mine: "Show me the methodology!"

The third conclusion – that Internet and PC penetration may be plateauing – is very informative with respect to the future of online shopping. For without a corresponding increase in PC and Internet users, the growth in the number of online shoppers is limited. The Census Bureau data are very clear on this. The number of new PC and Internet users each year is now decreasing. The Census Bureau estimated that in September 1999, 59 million Americans used the Internet at work or at home.

ot surprisingly, other research suppliers have published substantially higher estimates. Nielsen/Netratings estimated that 118.4 million Americans had Internet access in December 1999. This figure seems unusually high. And a search of Nua Internet Survey’s collection of “How Many Online?” studies confirms that the top five estimates all belong to Nielsen subdivisions. The vast majority of estimates are in fact much lower – in the 50-75 million range.

Can the Optimistic Forecasts be Achieved?

The wildly optimistic online shopping forecasts, like those of the Gartner Group, probably do not take into account the impending plateaus of PC and Internet usage. Many in the industry simply take for granted the continued galloping growth of Internet access. For example, Nielsen/Netratings last December claimed that between October 1999 and November 1999 the number of Americans with Internet access rose a stunning 5.4 million. But a recent Cyber Dialogue report found that one third of U.S. adults believe they have no need at all for the Internet, and that a significant number (estimated at 27.7 million) have tried the Internet and found they have no use for it. The Census Bureau data support this latter assertion.

So if the optimistic forecasts are to be achieved, something must change in the pattern of Internet access. Possibilities include:

Everyone who purchases over the Internet does ALL of his or her buying online;

Some discontinuity has to occur to expand the population that has Internet access; or

A vehicle other than the PC will have to be used to access the Internet.

The first of these possibilities is very unlikely. After all, there are a significant number of goods that consumers must experience before they buy them; tailored clothing, fresh fruit, and antique furniture will be very difficult to sell over the Internet. The second possibility would require the Internet to penetrate the lower income strata of society. Thus far, however, the lower cost of computers and free-PC model have yet to accomplish that goal. Finally, firms like Nokia are forecasting that five years from now, most Internet access will take place through handheld devices, such as the Palm Pilot or cell phones. The question remains as to whether such developments will expand the Internet user base or simply shift usage from one medium to another. (I believe the latter is more likely.)

A Challenge to Research Suppliers

Indeed, the more one crunches the numbers, the more difficult it becomes to square the sunny projections of research suppliers with more objective analysis. So I have a challenge for research suppliers, such as the Gartner Group, whose projections are even more optimistic than mine: “Show Me the Methodology!”

Look. It’s likely that both they and I are wrong. That’s the nature of the forecasting game. But executives and analysts need to know which e-commerce forecasts are the most reasonable to use for business planning. And the reasonability of any forecast depends on the credibility of its assumptions, integrity of the data, and soundness of the methodology used.

I’ve posted my major assumptions and the details of my methodology on Stern’s web site – here. Anyone who finds my assumptions and/or methodology unreasonable is free to reject my forecast.

ut the usual suppliers provide no such opportunity. What is their methodology? Do they project increased purchasing per customer? If so, how much? Will Internet access trends revert from declining growth rates to increasing ones? And if so, will they come through a medium other than the PC?

The suppliers of the data consumed so willingly by the media and the business community do not attempt to answer such questions. They provide a number and a mysterious black box, and ask us to take their results on faith. And they are slow to alter their projections. In response to questions from a Wall Street Journal reporter, the Gartner Group indicated that it was lowering its online shopping revenue forecast by about 35 percent through 2003 because research had revealed an unanticipated pullback in venture-capital funding for Internet retailers. The Wall Street Journal article appeared on May 25, 2000. Yet on September 15, 2000, Gartner still displayed its original optimistic forecast on its web site.

The Internet has empowered consumers of everything from airline tickets to stocks by improving disclosure. Today, an immense amount of previously hard-to-get information on everything from corporate finance to government operations is now available gratis on the Internet. And with every passing day, more data is available to businesses and consumers.

Given these developments, it’s more than ironic that some of the biggest boosters of e-commerce continue to operate behind a cloak of secrecy.

Joel H. Steckel is chairman of the marketing department at Stern. He acknowledges the assistance of Jill Grummert of Cyber Dialogue, Inc., and of Stern Professors Yannis Bakos, Sergio Meza, and Lee Sproull.