Harvard Infrastructure Conference, 23-25 January 1997
Numerous groups will sell you advice on getting rich on the Web; they discuss online sales of information, retail catalog shopping, advertising, consulting, and connectivity. What will actually pay for the Web? What is the `killer ap'?
This paper contains a great many conflicting numbers. The different predictors of future revenues differ; even the measures of current success differ. No effort is made to resolve the conflicts, since knowing the spread in estimates may be of value to the reader.
My personal projection for getting rich: connectivity.
What is going to pay for the Web? Why should Web site providers continue to do the work of writing, drawing, and coding, plus bear the cost of equipment and communication lines? Justification for Web sites includes many reasons which involve no direct financial gain, such as self-promotion. However, many site builders are hoping to get rich, despite costs that may run over $1M for a large professionally designed set of corporate web pages.
Among the possible models of finding wealth on the Internet are:
This paper discusses some of the estimates for the amount of money that people might pay for each of these services. The traditional business motivations are greed and fear. So far, a great deal of Web activity could be said to be motivated by fear: `our competitors have a web page.' Is it likely that greed will take over? And will it actually result in riches?
Mail-order catalogs are the model for online sales of objects. In the United States in 1995, catalog sales may have been $53B out of total retail sales of $2.2 trillion [Ziegler 1995], or perhaps they were $60B out of total retail sales of $1.7 trillion [Sanders 1996], or perhaps $46B out of $2.3 trillion total retail sales [Cwiklik 1996]. Online sales, by contrast, are still small: maybe $324 million sold online in 1995 (Wall Street Journal), or maybe $575M (Jupiter Communications), with other measures and forecasts ranging down to $160M. Most distressing is that some of the forecasts were higher than the actual numbers reached; online shopping has not boomed as rapidly as some forecasters thought. 1996 is estimated at either $518M or $1B [Stoltz 1996].
There are something like 100,000 retailers on the Web, or maybe only 10,000. The most active shopping seems to be for airline tickets, computer parts, CDs and books. As examples, NECX Direct (computers) sells about $1M per month; CDNow (audio disks) sold $2M in 1995 and expects $9M total sales in 1996; Amazon Books expects to sell $5M worth of books in 1996. Looking at a more traditional retailer, the Sharper Image sold about $250K online in 1995 and expects to sell $3M in 1996.
To consider a typical retailer, one online store is `She Sails,' which features clothing with a nautical theme or for use on boats. The owner, Ursula Kuehn, described her choice. If she prints and mails a catalog, costing $20,000 for 12,000 copies, she can expect 1 or 2 percent of those who get it to order something. Her Web page cost about $3,000 to set up and $300 a month to keep going; and it has 200 hits a week, producing about 2 orders a week of around $200 each. If we imagine amortizing the initial cost of the web page over 2 years, the cost per order from either the paper catalog or the Web is about $100. Clothing is a particularly hard sell on the Web. The HERMES survey of the University of Michigan Business School reports that only 2% of its respondents buy even casual clothes online; nobody buys formal clothes online; by comparison 12% buy computer hardware and 13% buy books.
More people want to buy books online, and bookstores, particularly Amazon Books, are prominent in Web sales publicity. The specialty store Pandora's Books says 30% of their customers are online. CD sales on the Web are also popular; specialty items that are hard for retail stores to stock can be easily handled by online book or CD operations. The CD vendors often allow you to access samples of the music \- listen to thirty seconds and then buy the record if you like it.
Another online operation is `Peapod,' which sells groceries to about 13,500 people (starting in Chicago). 80% of their users order on a computer. They charge $4.95 per month, plus $6.95/order plus 5% (note that they are delivering groceries, which are not usually sold by mail). In 1995 they grossed $16M, but their balance sheet showed a $6M loss. They hope in the long run to change the grocery business by taking advantage of the lower costs of warehouse space as opposed to supermarket space; they predict $750M of revenue for the year 2000.
Despite the publicity given to sales of books and CDs, Nevertheless, Jupiter Communications says that half of all online sales are airline tickets. Anyone using the Web also knows that a lot of pornography is advertised, and Forrester claims that about 10% of Web sales ($50M/year) are porno. Only 14% of Web shoppers have actually bought anything; there are a lot of browsers. Other bad precedents are the Prodigy florist, which sold $4M in 1994 but has not thrived as Prodigy has lost business (in 1996, Prodigy had 31 online merchants selling $35M). There used to be a set of shopping CD-ROMs produced; but such operations as `Compuserve CD,' `2 Market,' `Shopping 2000,' and Spiegel's CD-ROM have either stopped producing their CD-ROMs or retargeted them in some way.
Forecasts for shopping remain optimistic. Here are some forecasted volumes from Forrester:
Real stores have advantages over any kind of Web experience. Looking at any shopping mall, most people are in groups, not alone. Shopping is partly a social experience, and so far Web shopping doesn't replicate that. For some items, such as clothes or tools, the ability to try them on or handle them is important in making a purchase. And retail shopping provides instant satisfaction; no waiting for the delivery to arrive.
Web shopping for information, rather than objects, avoids the problem with delivery. Not only can the Web deliver information immediately, but for many purposes the digital information may be more useful than some kind of paper bought at a bookstore or newstand. Electronic information sales are of course a substantial industry today, although many of the sales are to professionals (e.g. librarians or stockbrokers). The information industry was about $13.5B gross in the United States in 1993, with much of this representing credit reports and other financial information [Hillstrom 1994]. It was showing a 25% growth rate and it is an industry in which the United States has an enormous positive balance of payments. The table below shows US imports and exports of data and information services for 1993.
|United States Trade in||Data and Information|
|Whole world||Japan only|
An enormous amount of information, of course, is distributed via the Web. About 200 GB of text is now online, perhaps the equivalent in volume of a 250,000 book library. Walnut Creek distributes about a terabyte a month from its ftp site, equivalent to about 5,000,000 copies of a magazine. As everyone knows, however, Web information is extremely variable in quality, and although copious in some subjects (current sports events, computer parts, and so on) extremely sparse in other areas. Ian Irvine (head of Reed Elsevier) said that Elsevier journals rejected 80% of what was submitted to them, and that those rejected manuscripts were what you found on the Web. ``Information you get for nothing,'' he continued, ``is worth nothing'' [Milliot 1996]
Publishers have been reluctant to sell information on the Web. One obvious problem is the risk of piracy. It is estimated that something like $15B worth of United States software is pirated every year, and $2B worth of movies are sold illegally. A software game industry on PCs flourished briefly in the early 1980s but was destroyed by illegal copies, and book publishers worry that the first digital copy of a book sold might be the last one sold. Another problem is competition with existing channels; bookstores might choose to put books available on the Web down on the bottom shelf. In addition to the loss of money, publishers fear a loss of control; for example, the book `Le Grand Secret' (a report of Mitterand's illnesses) was banned in France as an invasion of privacy, but it is widely circulating on the Web. It is estimated that something like one half the global software market is pirate and one fifth of the global music market is private; and this does not count private copies which are not distributed (it is estimated that three times as much music is copied privately as is sold commercially).
Nevertheless the attractiveness of information sales has stimulated many attempts. Companies like First Virtual and OpenMarket systems have tried selling items one at a time on the Web. Others such as Commercenet also aim to support companies making such sales. However, to date various problems seem to be discouraging the use of item-by-item sale of information. Both users and publishers fear the lack of predictability in item-by-item sales. Journal publishers, for example, have traditionally relied on selling subscriptions, so that they collect their income in advance; they know before the issues are printed how many copies will be sold at what price and thus what size the journal can be; they have no problem collecting bad debts; and so on. Users, in their turn, gain a certainty of knowing what their expenses will be. If information moves to a model in which articles, or perhaps even paragraphs, are charged for individually with billing taking place later, both sides worry.
In addition, electronic commerce is not really well enough developed for the small amounts that would have to be charged for individual articles. Most billing on the Web is still via conventional credit cards, and if small amounts of intellectual content are going to be sold separately, some kind of micropayment scheme will need to be adopted. There are many proposals for small payment systems, e.g. Netbill [Sirbu 1995] and Commercenet [Tennenbaum 1995], but none yet that has general acceptance.
America On Line used to employ a kind of item-by-item payment; when AOL charged its users about 10 cents per minute, AOL would pay a small fraction of this (one or two cents per minute) to the information provider whose information was being read. The rise of the Web, with free information, has discouraged this model, which never paid very large sums anyway. For example, Time magazine got several time as much in the way of bounties paid to it by AOL for people who joined by filling out an ad coupon in Time than it did for people reading Time online through AOL.
For these reasons a subscription model is becoming more common. The `Electric Library' charges $10 per month; InfoSeek charges $5/month; and the Wall Street Journal is charging about $4/month. As yet it is unclear how well these companies are doing. NewsPage, also at $4/month, went from 55,000 to 15,000 users when it instituted charging, but rebounded to 38,000 users as of December 1995. The Wall Street Journal is still providing free access to those who use a particular browser, so it is premature to judge what fraction of its subscribers will actually pay in the long run. As of October 1996 they had 30,000 paid users of their `interactive edition' while retaining about 100,000 users who get it free [Weber 1996].
Subscription sales of information look on balance as if they are thriving. In fact, they pose a threat to traditional companies like Mead Data Central or Knight-Ridder's Dialog which are accustomed to charging much higher prices to professional librarians; in addition to CD-ROMs, librarians can now get some of the same databases from the cheaper online services.
The figure below, for example, shows the revenues and profits of Mead[Weber 1996] Data Central. As can be seen, although revenue growth has been steady, profits have not kept track. Although some kinds of online information (library catalogs, indexing services) have largely replaced the paper versions, it is not yet as lucrative a business as the sales of paper.
Many CD-ROM publishers have also had problems turning a profit; the costs of producing multimedia CDs are high and sales have been disappointing for many producers. Some CDs are very successful; CD-ROM encyclopedias, for example, have largely taken over the market for paper encyclopedias. Many games sell well. But most reference CDs do not sell in quantity and profits have been hard to find. The boom of 1993-1994 tailed off at the end of 1995, although a record 2600 CD-ROM titles were released in 1996 [Lasky 1996]. This did not prevent 96 CD-ROM sales from dropping further; in 3Q 1995 43.3M CD-ROMs were bought, but in 3Q 1996 only 38.9M discs were purchased. Forecasts were that 25% fewer discs would be sold in 4Q96 and that this would be 39% less spending per household. The average CD title sells about 28,000 units and brings in $638K. Problems of declining interest might repeat on the Internet.
Forrester Research has taken a skeptical attitude towards online subscriptions. They predict few will pay more than $20 per year and that few sites will attract more than 12,000 subscribers, i.e. revenue of $240,000 per site. This would not be enough to make much difference to a major newspaper or magazine; e.g. the Wall Street Journal has 1.8 million circulation of its printed editions. Even projecting to the year 2000, Forrester believes that a site will do well to get $900K per year from subscriptions.
Esther Dyson is known for saying that on the Internet information should be treated as if it were free, implying that money will be made more by selling advice, comments, ratings, consulting and the like [Ross 1996]. There are various ratings services available on the Internet today, both formal and informal. Several Web searchers, for example, advertise the fact that they review the material they index instead of just grabbing all possible URLs (to provide material of higher quality, or to assure the absence of pornography).
The most interesting form of advice services are the shopping robots. Andersen Consulting, for example, built a program called the BargainFinder. It looks at different CD store web pages, looking for the lowest price for a CD that you wish to buy. The problem with this service is that many CD stores block the robot, since it is doing straightforward price comparison and they wish an opportunity to compete on service, delivery, or something else.
An interesting advice technology is the use of cooperative ratings. Groups at both Bellcore [Hill 1995] and MIT [Maes 1995] have introduced technology in which users are asked to rate something, and the ratings are used to model the users as well as the data. Once you have rated thirty movies, for example, the system compares your ratings to those of other people, and looks for movies that it thinks you will like but that you have not seen. Bellcore has used this for movies and restaurants; MIT, under the name Firefly, has used it for audio CDs. A new company, Agents, Inc. (founded by some of the MIT researchers) is commercializing this technology. Sapient is another company which attempts to deliver health care information using agents.
How to make these systems commercially successful is a problem. There is a great deal of free advice on the Web, and it looks as if finding people willing to pay for better advice may be difficult.
In 1996 the most hyped possibility has been that the Web, like broadcast television, would be supported by advertising. Web users are generally wealthier than the average American (after all, either they or their employer must be able to afford a computer), and they seemed a desirable group to which to advertise some products, particularly high-tech products. In the last quarter of 1995 about $13M was spent advertising on the Web, with estimates for the whole year running to $55M or $64M. Predictions for 1996 made early in the year ranged from $44M to $74M to $110M in advertising revenues. Estimates made late in the year don't agree much better, as shown in the following table.
|Web ad spending 1996|
One problem has been that the price of advertising on the Web, as more and more sites try to host ads, has been declining rapidly. In 1995 people talked about 10 cents per "view" of an ad, coming down to 6 cents, and by mid-1996 to perhaps 2 cents. Higher rates are charged for a kind of targeted advertising service, appropriate for search services, in which the ad shown depends on the search word typed. As a trivial example, an automobile company might wish to have its ad shown whenever the user typed car as one word in the search. The table below shows some late 1996 advertising rates.
|Ad rates in cents per exposure|
Cheaper total costs are of course available from smaller publications; American Horseman Online charges $40/month. For comparison, the Morris County, New Jersey cable company charges $65 for 30 seconds on CNN with about 1000 viewers; this is the equivalent of 6 cents per view (although the TV commercial may be considered to have more impact than a web page image).
The next table shows the actual Web advertising revenues for the first quarter of 1996.
|274||Wall Street Journal|
|232||Discovery Channel Online|
|192||Mercury Center (San Jose Mercury)|
|181||Jumbo Coolest Shareware|
Despite these relatively moderate numbers, the estimates for future advertising are enormous. Here are the estimates from Forrester:
|Year||Ads on Web|
|Web ads in 2000|
|New York Times||$5.0B|
The largest amount of money being paid today is being paid by people to be connected to the Internet. How many Web users are there? Here are various measurements and projections. Estimates taken in mid-1996 ranged widely, as seen in the table below [Kantor 1996].
|How many Web users?|
|Computer Intelligence Infocorp||15M|
The most extreme forecast of users from Nicholas Negroponte: he predicts 1 billion users in the year 2000. That is comparable to the number of people today with telephone service.
How much do those people pay to get access to the Web? Revenue for ISPs (Internet Service Providers) was $125M in 1995. Again, prices are declining as the number of users increases. For example, America Online pricing is as follows:
|$5.95/mo + 5.00/hr over 1 1992|
|$7.95/mo + 6.00/hr over 2 1993|
|(no change in 1994)|
|$9.95/mo + 2.95/hr over 4 1995|
|$9.95/mo + 2.95/hr over 5 1996|
In the public eye, the most visible online provider is AOL (America On Line). Here is a chart showing the number of subscribers to various services, including a few of the traditional professionally-oriented online vendors.
|DJ News Ret.||Dow Jones||.233M||.240M|
These numbers are subject to enormous turnover. AOL, for example, has seen nearly a quarter of its subscribers quit in one 3 month period. In first quarter 1996, AOL gained 2.3M customers but lost 1.4M. This makes it difficult to judge how successful the marketing effort of AOL will turn out to be.
Whatever the details, however, if we imagine that in the near future there are 50 million people paying $20/month, or $240/year, to connect to the Internet, this represents a $12B industry. This is the largest revenue stream directly predictable for the Internet.
If people started making voice or video telephone calls on the Web, this would represent a huge boom in Web business. Trying to measure the size of the Web as compared with the U. S. telephone network is difficult. Counting switching decisions, the Web is already larger (since it switches every packet, with a few dozen bytes per packet, while the phone network switches a typical 3-minute voice call only once). If one counts by bytes transmitted, the Internet is something like one-hundredth the size of the long distance phone network. However, the Internet is doubling every year; this would suggest it will catch up by 2004 or so. The growth rate is hard to project: on the one hand, it has been accelerating with the attractiveness of the Web, but on the other hand fairly soon it will become impossible to double the Internet merely by connecting more households, and further growth must represent additional use by people already connected.
One well-hyped possibility is making voice (or fax) calls via the Internet. In principle, Internet telephony can reduce costs by packetizing and compressing the speech. For fax transmission, where it is possible to allow some delays, costs could be reduced even further. Voice transmission can not tolerate delays, although users at the moment seem willing to put up with fairly poor quality (in exchange for free service). There is a complex legal and regulatory issue revolving around the payment of `access charges' by Internet users. Internet connection by modem, with an average holding time of 20 minutes, is having an effect on a local phone network designed for 3 minute average call durations, and we do not yet know what mechanisms will eventually dominate Internet connectivity.
Here are the market shares of current Internet telephony vendors (this is for the sale of the software that permits the phone calls):
|Market shares in Web telephony|
|Year||Consumer users||Business users||Total||Software Bought|
If videotelephony on the Internet became common, its bandwidth demands would dwarf anything else. There is little experience for guidance; to judge from AT&T's attempt to sell Picturephone in 1970, it would appear that videotelephony may be less attractive to people than plain voice telephony.
There are indirect reasons for having a web page. These represent some kind of cost avoidance or miscellaneous benefits. As examples,
There are even stranger motivations; one company has proposed to pay people just to look at ads.
The strongest motivation in early 1996 seemed to be not to sell products, but to sell companies. Netscape reached a market value of over $5B on 1995 sales of $80M (Maytag, by contrast has a market value of $2.5B on sales of $3.3B). Spyglass was valued at $242M on 1995 sales of $10M. For a while there was a rush of new stock offerings for Internet startups. As one would expect, some went up and some went down.
What is the balance between the different sizes of the markets, as predicted?
Here are the forecasts from Jupiter:
Feeling that it would be churlish not to offer my own prediction, I suggest that the answer is 20% from ads, 20% from subscriptions, and 60% from connectivity. What we lack at the moment is an effective way for content providers to benefit from the money paid to connectivity providers. Connectivity does not have the economies of scale of content provision, and so its prices can be more stable. Perhaps in the end some model such as the German tax on blank tape will be applied to the Internet. Alternatively, a technological solution to the piracy problem might make subscription publishing on the Internet more feasible.
On balance, my guess is that just as telephony and the post office are larger enterprises than broadcasting and publishing, we shall also see inter-personal communication and access as more significant than shopping and advertising. Come back and read this paper in the year 2000.
Robert Cwiklik, ``Revisit the Cyberseers at www.hindsight.1996,'' Wall Street Journal, December 9, 1996.
[Hill 1995] Will Hill, Larry Stead, Mark Rosenstein, and George Furnas, ``Recommending and evaluating choices in a virtual community of use,'' Proc. CHI Conference on Human Factors in Computing Systems, Denver, Col., 1995, pages 194-201.
[Hillstrom 1994] The Encyclopedia of American Industries, Volume Two: Service and Non-Manufacturing, ed. Kevin Hillstrom, Gale Research, 1994.
[Kantor 1996] Andrew Kantor, Michael Neubarth, ``Off the Charts,'' Internet World, vol. 7, no. 12, pp. 44-51 (December 1996).
[Lasky 1996] Michael Lasky, ``Best CD-ROMs of 1996,'' PC World, December 1996, page 113.
[Maes 1995] Upendra Shardanand and Pattie Maes, ``Social information filtering: algorithms for automating 'word of mouth','' Proc. CHI Conference on Human Factors in Computing Systems Denver, Col., 1995, pages 210-217.
[Milliot 1996] Jim Milliot, ``Publishers still searching for profits in new media,'' Publishers Weekly, vol. 243, no. 1, page 22 (January 1996).
[Ross 1996] Philip Ross, ``Cops vs. Robbers in Cyberspace,'' Forbes, September 9, 1996
[Sanders 1996] Jared Sanders, ``At Last, Main Street.com is Opening for Business,'' Wall Street Journal, June 17, 1996.
[Sirbu 1995] M. Sirbu and J. D. Tygar, ``Netbill: an Internet commerce system optimized for network-delivered services,'' IEEE Personal Communications, vol. 2, no. 4, pp. 34-39 (August 1995).
[Stoltz 1996] Craig Stoltz and Carolyn Spencer Brown, ``wwwhy.//shop.on/line? We spent several weeks doing it -- and still aren't sure we know the answer..'' The Washington Post April 24, 1996, page R04.
[Tennenbaum 1996] J. M. Tennenbaum, C. Medich, A. M. Schiffman, and W. T. Wong, ``Commercenet \- spontaneous electronic commerce on the Internet,'' Proc. COMPCON conference, pages 38-43, (March 1995).
[Weber 1996] Thomas Weber ``Interactive Edition Attrats More than 38,000 Paid Users,'' Wall Street Journal Interactive Edition, October 9, 1996
[Ziegler 1995] Bart Ziegler, Associated Press, Dec. 20th, 1995.