The need to give advertisers better information on Web site traffic has spawned an industry of data collection software and services.

Modern capitalism owes an enormous amount of debt of gratitude to Luca Pacioli the mathematician whose treatise on double-entry bookkeeping, De Arithmetica, provided merchants throughout Europe with a system for keeping track of their physical and financial assets and enabled them to move out of a barter economy. What the world needs now, it seems, is another Pacioli who can invent a methodology for keeping track of the only assets most Internet entrepreneurs have to sell: the fleeting and disembodied visits of travelers on the World Wide Web.

As recently as a year ago there was really no way to measure Web site activity beyond the proprietary solutions--rarely more than educated guesswork--developed by individual sites. Even today, Internet advertisers are frustrated because they are being asked to pay top dollar without the familiar metrics that are used to measure effectiveness in the print and broadcast worlds. Web publishers and advertisers alike recognize the need for the Net equivalents of Nielson ratings, Audit Bureau of Circulations reports, and other reliable audience yardsticks if the Web is ever to become an ad-supported medium.

So the recent efforts of a number of new companies to provide meaningful statistics on Web site traffic could have something of the same import for the future of online businesses as double-entry bookkeeping had for medieval commerce. On the other hand, the complex issues involved in measuring something as ephemeral as a Web site "hit" won't be easily untangled. In a previous letter ("Adding it Up," May 20, 1996) we discussed some of the problems of translating print and broadcast concepts like frequency, reach, impressions, and demographics to the Internet. Such things as offline page caching, proxy servers, firewalls, and site mirroring create special problems when it comes to exact measurements. To take just one example of a situation peculiar to the Web, what happens if a user chooses to turn off the graphics to speed up browsing? But here's the irony: Because computer technology is so much a part of the online medium, advertisers believe that it ought to be able to deliver much more precise accounting than is possible in print or broadcast.

Hit parade

Certainly a Web site is capable of accumulating a vast amount of information. And over the last year or so, a number of tools have begun to emerge that can mine this data for the answers advertisers are seeking. Publishers have been using hits as a way to measure usage, but hits are not very precise. They only measure how many times a file was requested from the host server. When impressions are measured based on how many people actually clicked through to the advertisers site, the number drops off dramatically. Furthermore, measuring hits doesn't produce any demographic information about visitors.

Early Web server analysis tools were largely enhanced versions of Common Gateway Interface scripts that read log files and extracted usage patterns. Designed primarily for Web administration, they were not easy to use, nor did they go very deeply into the data. Over the last year, however, the sophistication of the tools available for extracting information from site logs has increased enormously. The newest tools can track exposures or clicks by hour and day, measure how many click-throughs were successful, and report where requests originated. In many ways this is simply the maturation of the Internet as an application platform. Newer versions of Netscape's Navigator browser can be polled to reveal, for example, the last Web site and incoming visitor checked in with, which allows Web masters to know the origin of their traffic. On the server side, vendors have steadily added more state to the HTTP protocol Web sites use, which means increasing the ability to retain information on requests for Web pages.

Service aces

Two basic business models have emerged from the face to commercialize Web site measurement and research. Service companies extract information from site logs and slice and dice the data to come up with standardized and customized report for the site owners; software companies sell products that enable sites to generate this information themselves. NetCount and I/Pro are the major companies following the service model. Initially an in-house operation of Web site developer Digital Planet, NetCount was spun out as a separate company about a year ago. Reflecting its origins NetCount markets its service in part through site development companies like Planet Genesis and Media Circus, which in turn receive a commission for providing NetCount solutions to their clients. The company claims to be the only service that provides its clients with an accounting of both the requested transfers for content on a Web site. The distinction is important because when a server is busy or a page takes too long to download, users may hit "stop" so that the requested information is never transferred. I/Pro, which has been offering Web measurement and analysis tools for about a year, has also built a consulting group that does custom work as well as issue reports.

Numbers Game For Web site managers who prefer to generate their won reports, a number of companies sell software that resides on the site's own servers and spits out statistical information that ranges from bare-bones to highly sophisticated. At the low end, Stamford, Conn.-based Software Corp. of America offers a $299 product called WebTrends, which provides query options and reporting tools that allow the user to produce customized reports. More powerful--and costly--products are being sold by Interse and two newer companies, Accrue Software and Andromedia. These products have a built-in database and can generate information similar to that provided by service companies although each company puts its own twist on what it measures.

Interse, true to its origins as a data mining company, promotes the ability of its market-focus software to go deeper into the data than its competitors, including the ability to analyze the path people take when they navigate a site. Andromedia's Aria system emphasizes real-time activity monitoring, recording and reporting data such as the type of browser and IP address of the client. Accrue is implementing its web site measurement at the router level, which allow its Insight software to combine hit tracking with more accurate metering of the packets flowing over the network. By closely monitoring packet flow, Accrue believes it can get a better picture of the exact end-user experience of a site. For example, Accrue's software can tell how long the user waited for an image to download before hitting the cancel button and suggest whether the download was too slow because of server speed or because the page contained too many graphics.

Household help

Coming at the problem form a different direction, PC Meter, a subsidiary of the consumer data specialist NPD, has developed a model based on monitoring Internet activity from the client side rather than the the server. Software on a consumer's computer monitors online and Internet usage, tracking which sites of domains are being visited as well as what pages are chosen and in what order. PC Meter know the demographics of the household using the software as well as the profile of each user. This allows the company to provide actual data about the age, sex, and income of the users it monitors.

Because PC Meter measures actual individual behavior, it can break down total visits into reach (the number of consumers) versus frequency (the average number of times each consumer is exposed). It can also account for such variables as page caching. Of course, any tracking system that is based on a sample of individual consumers is only as good as the methods used for selecting the sample and ensuring that the individuals chosen accurately reflect the wider population of users. PC Meter now tracks 10,000 households, selected for a representative distribution of age, sex, and income.

Another approach that seems to be gaining popularity is to concentrate more directly on the needs of advertisers and agencies that are starting to place online ads. Products from companies such as Accipter, FocaLink, and NetGravity automate ad placement on Web sites, thus preventing manual errors and reducing administrative costs. Scheduling systems enable Web sites to deliver guaranteed numbers of impressions to their advertisers. Advertisers in turn can view up-to-the-minute reports on how an ad is performing. FocaLink aims its product at advertising agencies who want to manage their clients' online presence. The company also provides advertisers with extensive profiles of individual Web sites, data about site content and traffic, ad pricing, and sales contacts at each site.

NetGravity and Accipter, on the other hand, have primarily targeted publishers and content providers with software that generates information for advertisers and agencies. Accipter, a new startup, hopes to make its mark by adding a demographic component. With Accipter software, a Web site can develop audience profiles--which the company calls "cyber-segments"--using such information as site registration information, domain type and the page of a site the visitor is on. This information can be used to deliver targeted customers to advertisers. A game developer with a new game for the Sega Saturn, for example, could target its ads to a Sega users group while another developer might want to reach only strategy game players.

Qualifying site visitors is a much harder task than simply quantifying them. While everyone who visits a site leaves behind a record of that visit, that record doesn't give an advertiser or publisher very much information. But finding out more about Net users has met with a lot of cultural resistance. I/Pro tried adding incentive to site registration by offering sweepstakes prizes for people who provided information for its I/Code registration system but abandoned the scheme recently because there were not enough entries.

NetCount has plans to launch a product called head Count which will track registered users and provide household of business demographic information. NetCount plans to work with sites where people have to register in order to get certain information. At C|Net, for example, you can browse without registration but are required to sign in if you want to chat.

Follow the Surf For advertisers, the ability to qualify site visitors would greatly enhance the efficiency for their ad dollar. DoubleClick, a media buying service that was spun off from the advertising agency Poppe Tyson, can track people through its network sites and infer from previous behavior what kind of ads they would like to see. Say you have spent time on sites in the DoubleClick network clicking on pages about gardening. Next time you sign on to your favorite Web site, if it is a DoubleClick site, the software checks your e-mail address against your profile and a banner for Smith and Hawken or a plant special from 1-800-Flowers would pop up.

One approach for gathering more detailed individual information is the Firefly Network intelligent agents model (see "In From the Cold," September 23, 1996) in which users fill out profiles and in return are connected to information that is interesting and relevant. We're also interested in the communication technologies developed by BackWeb Technologies, launched in Israel but now headquartered in San Jose, Calif., and Seattle-based Intermind, which are just now bringing out products that allow Web sites to create one-to-one relationships with users. (We'll report on this phenomenon in a future letter.)

The old fruit question

As you would expect from an industry this young, there are a lot of issues to be worked out. Of key importance to advertisers is making sure that apples are being compared to apples. Different sites use different methodologies, and there are no standards yet to determine what, for example, counts as an impression. Traditional standards associations such as ABC are setting up guidelines for the Internet, and some attempts to work out standards are being undertaken by the Internet Advertising Bureau. But the desire for standards collides with constantly evolving technology that requires new standards. We suspect it will be a while before these issues get resolved and that advertisers will have to take the usefulness of much online advertising on faith.

As a common measurement standard evolves, we suspect it will be more like the broadcast model (where companies like A. C. Nielson extrapolate number from actual consumer surveys) than the print model, which counts circulation. The Internet is more like television than it is like a consumer magazine, where passthrough circulation can only be estimated. With more information on the behavior of individual surfers, advertisers will be able to evaluate their campaigns using metrics other than gross numbers. Eventually, perhaps, that will get them close to the ad man's nirvana of being able to reach a person--and only that person--who is just waiting to learn more about his product.