Internet Advertising: The Medium is the Difference

October 4, 1999

Xavier Drèze Fred Zufryden

The underlying full-duplex networked organization of the Internet transforms the traditional one-way relationship between advertisers and consumers that is inherent in standard advertising. Internet content providers take on both the roles of conduits and interaction enablers in a two-way interaction with prospective consumers.

2

Introduction Internet advertising is growing at an impressive rate. A recent report by PriceWaterhouse-Coopers estimates 1998 online ad revenues at $1.92 billion, more than twice those of 1997. By year 2000, it has been projected that ad revenues will grow to a level of $ 4.3 billion (IAB, 1998). This spectacular growth comes as no surprise if one considers that on the one hand, advertisers are always looking for new ways to reach their target audience, and on the other hand, web site operators are always looking for new sources of revenues to finance their ever-mounting operating costs. The willingness of both parties to work together does not mean however that the blending of Internet content and Internet advertising will be painless. For reasons that we will investigate in this paper, one cannot simply transpose current advertising practices from television, or newspaper, to the Internet. Indeed, in order to take full advantage of the potential of Internet advertising, and to avoid its pitfalls, one needs to rethink the consumermedium interaction as it applies to Internet advertising. In this article, we show how the underlying full-duplex networked organization of the Internet transforms the traditional one-way relationship between advertisers and consumers that is inherent in standard advertising. On the Internet, content providers take on roles of both conduits and interaction enablers in a two-way interaction with prospective consumers. This new model profoundly affects the relationship between each player. Consumers now interact with both content providers and advertisers. Content providers are not responsible for advertisement fulfillment anymore. It also affects how content providers are compensated for their services. As conduits, their compensation scheme was simple. As relationship creators, their value is harder to assess.

3 We focus on two key differences between advertising in traditional broadcast media (e.g., Radio or Television) and that online. We first look at how the two-way networked organization of the Internet contrasts sharply with traditional broadcast media. Second we look at how the advertising messages are imbedded on the Internet and other media. We examine some of the implications of the fact that banner ads typically share the screen real estate as opposed to take over it, in a linear fashion, during the broadcast of the advertising message, as television advertising does.

Contrast of Broadcast and Request-based Media Frameworks In traditional media settings, the flow of information between advertisers and consumers follows a unidirectional, and linear, path (see figure 1). Advertisers enter in an agreement with content providers whereby they purchase space on a provider’s medium where their promotional messages will be inserted. The content provider then merges all the advertising material he has contracted for with his own content and transmits the whole to prospective consumers. In turn, consumers elect to receive (i.e., view) or bypass the advertising message (e.g., they can zap by switching to another channel). For those who are exposed to the ad, the attention given to the ad message will depend on numerous behavioral factors (e.g., demographics, life style characteristics, product awareness, product need, use behavior, etc.). In this setting, there is no direct contact between advertisers and consumers. What advertisers know about their audience comes from industry reports such as those from AC Nielsen (www.acnielsen.com) or Arbitron (www.arbitron.com). These reports provide advertisers with information such as Reach (the number of unique individuals who have been exposed to an ad campaign at least once), Frequency (the average number of ad exposure for the individuals who have been exposed at

4 least once), Gross Rating Points (GRP = Reach x Frequency), and some demographic information about the reached audience. The information reported to advertisers is generated through surveys or panels of consumers that measure audience viewing, or listening, patterns. For example, in the TV industry, Peoplemeter devices are used to monitor TV viewing patterns (i.e., whether a TV set is on or off and what program, if any is tuned at a given point in time), and consequently audience viewing habits, for a representative sample of households. However, for the television medium, content providers have imperfect information about their audiences. This is because marketing research firms are merely able to monitor the TV set patterns of a sample subset of the population rather than the actual viewing patterns of their audiences. Thus, the monitoring of broadcasts does not provide completely accurate information concerning who is tuned in to particular programs. Print media, such as newspapers and magazines, fare a little better in that they can readily estimate the size of their audience from the number of copies they sold. However, they still experience uncertainty as to the characteristics of their audience as they do not readily possess information about consumers who chose to buy their publications at a newsstand rather than those who subscribe. The Advertiser->Content Provider->Consumer framework, which we call broadcast, is very efficient in reaching mass markets. It requires a minimum number of interactions and has been used for more than a century. It is very different from the request-based mode of information dissemination that occurs on the Internet (see figure 2). On the Internet, advertisers still contract with content providers to insert advertisements within the programming offered to consumers. But from this point on, everything is different. First, content providers do not push their content on consumers at pre-established points in time. Rather, they wait for consumers to

5 request the information when they are ready to consume it. Second, content providers do not broadcast the advertisements linearly along with their content. Rather they send their content to consumers with an instruction to retrieve the appropriate piece of advertisement directly from the advertiser. Upon reception of the location of the advertisement, consumers then contact the advertiser themselves to retrieve the ad. We review the key implications of this drastic change in the nature of the advertiser-consumer relationship in the next section. To facilitate our discussion, Table 1 provides a contrasting summary of the features of the Internet and Broadcast frameworks.

Advantages of Advertising on the Internet The multiplicity of interactions that come about by the modus operandi of the Internet creates a wealth of opportunities for marketers (some have called the Internet the Holy Grail of marketers). First of all, since consumers directly request the content they would like to access, it becomes easy (at least theoretically) to measure the size of the audience of a particular ad campaign. Second, since consumers who subscribe to several different content providers displaying the same ad have to request the ad from the advertisers themselves rather than from the content providers, it becomes potentially easy for the advertisers to measure the true frequency of exposure. Third, because the advertiser only serves the actual ad content at the time when the consumer is ready to consume it, advertisers don’t have to commit on the ad content until the last instant. This provides considerable opportunity for tailoring an advertising message to a particular prospective consumer. We now explore some of these opportunities.

6 Accurate measurement Now that consumers and advertisers interact directly on the Internet, great improvements in quality of marketing research can be made by shifting from a survey-based to a census-based method of assessing advertising effectiveness. Figures 1 and 2 suggest the benefits of a censusbased advertising tracking methodology. In figure 1, the advertiser reaches Consumer 1 through both Provider 1 and Provider 2. However, given the lack of consumer level data, all the advertiser knows is that Provider 1 and Provider 2 reach similar demographics. He will have no indication of the actual overlap between the magazines or television stations. In contrast, the Internet advertiser will see Consumer 1 come twice (figure 2), once send by Provider 1 and a second time send by Provider 2. Hence, the advertiser now has an accurate measurement of the number of consumers who see its ad (Reach) and of the number of times each consumer sees the ad (Frequency). The interactive nature of the Internet medium has lead to new criteria for the measurement of advertising effectiveness: "page views" (or impressions) and "click-throughs." Impression represents an opportunity for a surfer to see and click on a banner ad within a publisher's Web page, and click-through represents the committed action of a surfer who actually clicks on a banner ad in response to its message. This immediate measure of advertising response is a new concept that can only be measured on the Internet. Many see it as the ultimate measure of advertising effectiveness.

Potential integration of advertising and marketing research functions The standard broadcast medium environment only permits the dissemination of advertising messages to customer prospects. However, because of the bi-directional nature of

7 the request-based Internet advertising environment, both the dissemination of information (in the form of advertising messages) and the gathering of information (in the form of market research data) are made possible. Because the dissemination and gathering of information can be done simultaneously at a given Web site, the Internet medium offers considerable advantages relative to the integration of the advertising and marketing research functions. First, the Internet medium provides for timely and speedy gathering of research information (e.g., information regarding page views, visit duration, and even on-line purchase statistics can be gathered as a customer navigates through the pages of a Web site). The cost of gathering research information in this manner is less than that of standard research methods such as off-line surveys (i.e., information can be gathered very cost effectively on the Internet by means of available tracking software that may be installed on a Web site's server). As was mentioned before, the information gathered on the Internet is census-based and thus does not incorporate any sampling error. Further, the data can be collected in an unobtrusive manner and thus does not produce response biases1. The ability to instantaneously measure the response of consumers to a particular banner ad is a powerful tool to optimize advertising effectiveness. Drèze and Zufryden (1997) showed how the market research technique of conjoint analysis could be used to develop a Web site design configuration that would maximize the average number of pages viewed by visitors as well as the average amount of time spent by visitors on the Web site. A similar methodology can be used to test and develop advertising message design configurations (e.g., in terms of type of appeal, colors, font types and sizes, location on page) that would maximize an advertiser's objective (e.g., maximizing awareness, product interest, or even purchase response).

1

When consumers know they are being surveyed or monitored, they often alter their behavior to conform to what they believe the researcher thinks is an appropriate behavior, thus invalidating the research. This response bias is

8 Potential integration of advertising and transaction functions With the growth of e-commerce, the Internet medium provides the additional facility for consumers to conduct product transactions directly on the Internet. Thus, the bi-directional nature of the Internet environment provides a way to not only direct advertising appeals and stimuli to target customers in an effort to encourage customer response but also provides for the potential realization and tracking of the purchase response on the medium itself. This provides considerable opportunities for the development and fine-tuning of advertising and marketing strategies that best stimulate prospective customer's purchase response behavior.

Adaptive Advertising First, it should be noted that ads from standard media come to prospective customers whereas prospective customers on the Internet medium come to ads. Furthermore, in contrast to the mass, or segmented, marketing strategies that are afforded by standard media (e.g., a TV advertisers can focus their efforts on a given demographic or geographic segments), Internet advertising creates the potential opportunity of micro-level marketing. This means that an Internet advertiser can specifically design and tailor a given advertising message to target a particular individual consumer. For example, an advertiser can chose to deliver a specific ad to a given individual only after the individual has exhibited a particular type of behavior (e.g., has expressed a given level of interest in a particular product in view of his origin or previous clickstream history). As such an individual consumer who comes to an advertiser's site for the first time might be given advertising in the form of general product information. In contrast, a repeat visitor might be given more detailed product information in an effort to move him/her towards a product purchase decision. If one refers back to Figure 2, Customer 1 is in a situation the marketing equivalent of the Heisenberg uncertainty principle.

9 where he retrieves the same ad twice. In this web-based environment, the advertiser can decide whether he wants the consumer to see the same ad twice, or see two different ads. Further, by limiting the number of repetitions a consumer is exposed to, advertisers can limit burnout and maximize their return on marketing expenditures. An advertiser also has the ability to tailor the ad campaign to the environment the consumer is involved in at a particular moment. For instance, specific advertising messages can be tailored to specific search engine keywords. Thus, a particular key word (e.g., PC hardware) may be used to generate ads and for PC brands whereas a combination of key words (PC hardware and storage media) might be used to generate ads for CD-ROM, removable hard drives, or DVD drives. Challenges in Maximizing Ad Effectiveness on the Internet The previous section is very upbeat about the potential of online advertising. This potential is very real and has been demonstrated in case studies in numerous occasions. Unfortunately, there is a big gap from case studies to large-scale working production-grade tools. When attempting to unleash this potential on a large scale, marketers have encountered many implementation issues. The problems generally fall in two broad categories. First, the ability to track individuals is significantly impaired by the physical implementation of the Internet (e.g., caching or IP sharing). Second, although we have information about a lot of users, we know very little about each one of them. We will now review some critical roadblocks one encounters when trying to implement the techniques we have just developed.

10 Difficulties in identifying unique visitors As noted in Drèze and Zufryden (1998), in order to develop accurate measures of unduplicated audience reach for Internet-based advertising, it is necessary to identify unique visitors' page requests. However, on the Internet, visitor request patterns, Web site traffic, and flow patterns are generally established on the basis of visitors' (or requestors') Internet Protocol (IP) addresses. Unfortunately, as Drèze and Zufryden point out, IP addresses are not necessarily uniquely assigned by a visitor's Internet Service Provider (ISP). In fact, it can be shown that several Internet users can be assigned the same IP address. To make matter worse, visitors may be assigned different addresses even within a single Internet session. These idiosyncrasies make it very difficult to identify unique visitor requests on the basis of IP addresses alone. In fact, Drèze and Zufryden report significant errors in the measurement of Reach and Frequency of advertising when IP addresses are used for this purpose.

Problems in measuring multiple ad exposures A key determinant of the measurement of banner advertising effectiveness is the number of pages requested by a particular surfer on the Internet. The accurate determination of this measure is essential for an accurate measurement of the average frequency of advertising exposure on the Internet. In this regard, one must remember that a Web page is often a complex composite document that may contain text as well as graphics, sounds and video files. In order to speed up the retrieval process of the Web page when a surfer uses the back button to return to a page, a browser will often store (or "cache") the Web page, at the time of the surfer's first request, on the surfer's hard drive. If, and when, the Web page is requested again, the browser will draw the Web page from the hard drive (cache) rather than go through a more time-consuming page

11 request from the server. Unfortunately, this means that the server may not record subsequent page views by the surfer. Therefore this leads to a potentially significant underestimation of the magnitude of advertising frequency on the Internet.

Absence of standardization of ad effectiveness measures Several third party market research tracking companies (e.g., MediaMetrix, RelevantKnowledge, Nielsen) have attempted to focus on the development of standard advertising effectiveness measures (e.g., Reach and Frequency) in an effort to promote the standardization of advertising effectiveness and permit the comparability of Internet advertising with that in standard media. Unfortunately, some of the measurement problems discussed above have often led to dramatic disparities in reported advertising effectiveness statistics from alternative third party companies (e.g., see Peter Kafa, 1999). Aside from the discrepancies due to differences in measuring unique visitors and accounting for caching, disparities in reported statistics from different Internet tracking organizations are also due to methodological differences in sampling procedures. For examples, MediaMetrix recruits its panels by buying population lists that includes PC users (some of whom do not currently use the Web) and conducting random mailings and phone calls. In contrast, RelevantKnowledge relies on random phone calls to recruit people who already use the Web. Furthermore, some have questioned the relevance of the measures reported by these companies. Briggs and Hollis (1997), and Drèze and Hussherr (1999) both question the relevance of click-through as an advertising-effectiveness measure. They show empirically that banner advertising has a significant impact on consumers even in the absence of click-throughs. Constructs such as brand awareness or advertising recall are influenced by banner ads even after

12 one or two exposures. This means that click-through measures underestimate the effectiveness of online advertising and that more traditional ways of measuring this effectiveness should be used.

Errors in current measurement A recent study by Drèze and Zufryden (1998) investigated the very significant errors that may be expected in current advertising measurement methods as a result of the problems described with respect to identifying unique page requestors as well from page caching. They found for example that current measurement methods, relying on IP addresses to identify unique visitors, led to an overestimation of Web pages seen by a visitor by about 64%. In terms of standard measures of advertising effectiveness, they found that the use of IP addresses led to an overestimation of Reach (within the range of 12% to 42%), an underestimation of Frequency (within the range of -13% to 9%), and an overestimation in GRPs (Gross Rating Points) of about 23%. In examining the effect of caching on the recording of multiple exposures, it was found that about 38% of page requests were cached and thus not recorded by servers.

Absence of demographics In standard media, demographic characteristics are typically gathered from survey information and reported to advertisers. For example, AC Nielsen reports audience measures for the TV medium by for various age and geographic groups. Unfortunately, comparable information is not readily available on the Internet. Although certain Web sites collect this information on the basis of registration data, the accuracy of the information is unknown. One

13 problem is that Web surfers often try to maintain their anonymity. Therefore, the reliability of information that is supplied via registration procedures is subject to question. One possibility is to develop demographic data on the basis of geographic information. That is, it may be possible to determine a visitor's geographic location in view of his IP address and the ISP providing his/her service. From this information, it may be possible to develop demographic distribution statistics such as those that are available from geodemographic sources of data (e.g., Claritas Prizm).

Content Multiplexing In addition to delivering content through a different model framework, the Internet differs from broadcast media in the way it integrates the advertising message into the content that consumers are accessing. When delivering an advertising message, Radio or Television operators need to interrupt the broadcast of their message, in a linear fashion, in order to broadcast the ad (see figure 3). That is, the broadcaster alternates between sending content (e.g., music or news) and sending advertisements. At all times, one hundred percent of the available bandwidth is allocated to one of the two alternatives. By contrast, Internet content providers embed advertising within their content, allocating only a fraction of their bandwidth to the advertisement (see figure 4). This results in spatial multiplexing rather than temporal multiplexing. This spatial versus temporal multiplexing has far reaching consequences on how consumers process ads. This, in turn, has significant implications on the design and placement of banner ads on a page. In terms of information processing, online banner ads are at a

14 disadvantage. Two factors are contributing to the difficulty for banners to influence web surfers: "involvement" and "size". Television is a low involvement medium. Viewers generally choose a program to watch and sit back to enjoy the show. When commercials interrupt the broadcast, some might switch to another channel, but they constitute a small minority (less than 5% per Siddarth 1999). Most viewers simply watch the commercial waiting for the show to resume. By contrast, the Internet is a high involvement medium. Surfers are actively choosing what material to see. They are often searching for information (e.g., Stock Quotes, Football results, or the latest MP3 files) and are processing the content of the pages they are accessing in order to find that content in the least amount of time. During this search, banner ads are irrelevant objects that tend to slow down the process. Hence, by default, surfers will avoid processing banners. Drèze and Hussherr (1999) have shown using an eye-tracking study that surfers tend to avoid looking at banner ads and that only 50% of banners are attended to. Size of the advertising stimulus is another problem for online advertising. Banners typically take less than 10% of a screen’s real estate. This is a far cry from television ads that take up the whole screen for 30 seconds or one minute. This means that while television ads can be akin to small movies, with an introduction, a plot development, and a climax (in full color and surround sound), banner ads are limited to simple one-line messages with maybe some primitive animation. In short, banner ads are at a disadvantage because, by default, surfers are looking at something else they are interested in, and due to their size limitation, banners have only limited ability to grab the attention of surfers and convince them to take action. As such, banners are closer to highway billboards than anything else. This means that Internet advertisers are faced

15 with a formidable task and will need to engage in a lot of concept testing in order to find a banner that works. Fortunately, the ease of implementing only marketing research procedures (e.g., split-cable ad testing in order to examine which ads are most effective (Drèze and Zufryden 1997)) may help to mediate this problem.

Conclusion To benefit from the unique Internet medium environment, advertisers and managers need to consider the idiosyncrasies of Internet advertising. A unique advantage of the Internet Medium is its bi-directional framework that incorporates two-way links between advertiser and prospective consumers. A beneficial byproduct of this linkage is that customers can be targeted individually at the micro-consumer level. In addition, because of the interactivity afforded by the medium, the advertiser can tailor individual advertising and formulate advertising messages adaptively. Because of the two way flow of information that characterizes the medium, the latter can deliver advertising messages to individual consumers and, at the same time, gather consumer information on a continuous basis as prospective consumers surf the publisher or advertiser's Web sites. The integration of the transaction function within the medium provides for a closing of the marketing-response loop which permits the potential development of appropriate advertising and marketing strategies that will best influence a prospective consumer's purchase behavior. Despite the promises of the Internet as an advertising medium, there are continuing challenges that make the effective implementation of advertising strategy difficult. Measurement problems exist due to lack of standardized advertising unit of measures, measurements, as well as difference in the tracking methodologies that are used by third party companies. Current

16 measurement methods typically have limited accuracy. It is generally not possible to provide measures of ad effectiveness on the Internet that may be appropriately compared with those of standard media. Thus, relative media effectiveness across media, including the Internet, is difficult to assess. The different methods of pricing advertising on the Internet (on the basis of impressions or click-throughs) also make it difficult to assess the cost/effectiveness of advertising in this new medium and thus hamper the implementation of advertising budgeting. Because Internet advertising involves a partial allocation of bandwidth to advertising messages, it is more difficult to draw a prospective consumer's attention to an ad on the Internet as it is shared with other contents including potentially competing ads on the same Web page. This suggests that ad design and placement aspects may be of great importance to insure that a consumer will be drawn to the banner ad. This means that it is important for advertisers to gain a greater understanding about how prospective consumers process information, including ads, on Web pages. In sum, there are great opportunities for advertisers on the Internet. However, as was pointed out, considerable challenges still remain. Nevertheless, it is expected that the opportunities and promises of the Web will be realized as advertisers and researchers continue to gain a better understanding of the unique characteristics of the Web as a medium as well as the nature of consumer search and response behavior on the Web.

17

References Briggs, R. and N. Hollis (1997), “Advertising on the Web: Is There Response before ClickThrough?,” Journal of Advertising Research, 37 (2), 33-45. Drèze, Xavier and François-Xavier Hussherr (1999), “Internet Advertising: Is Anybody Watching?,” USC Working Paper. Drèze, Xavier and Fred Zufryden (1997), “Testing Web Site Design and Promotional Content,” Journal of Advertising Research, Vol. 37, No. 2. Drèze, Xavier and Fred Zufryden (1998), “Is Internet Advertising Ready for Prime Time?,” Journal of Advertising Research, Vol 38, No. 3, 7-18. Internet Advertising Bureau (1998), "Internet Advertising Approaches $ 1 billion," April 6, [URL:http://www.iab.net] Kafa, Peter, (1999), "Unaccountable," Forbes, Technology the Internet, (May 3), 194-95. Siddarth, S. (1999), “Describing the Dynamics of Attention to TV Commercials: A Proportional Hazards Model of the Time to Zap an Ad,” USC Working paper.

Content Provider 2

Advertiser

Figure 1: Content Flow -- Traditional Media

Content Provider 1

Consumer 1

Content Provider 3

Consumer 2

Figure 2: Content Flow -- Internet

Consumer 1

Content Provider 1

Requests

Replies

Advertiser

Content Provider 2

Content Provider 3

Consumer 2

Content Advertising

Output

Figure 3: Temporal Multiplexing – Broadcast Media

Input

Temporal Multiplexing

100%

Bandwidth

t

Content Advertising

Output

Figure 4: Spatial Multiplexing -- Internet

Input

Spatial Multiplexing

100%

Bandwidth

t

22 Table 1: Contrast of Internet vs. Broadcast frameworks Internet Two-way communication Passive (impressions) and Active ad viewing (click-through) Micro-level audience targeting Census-based audience information No sampling error Advertising message can be tailored "on the fly" to the individual Shared bandwidth Potential concurrent integration of advertising and marketing research functions Potential integration of advertising and purchase transaction functions

Broadcast One-way communication Passive ad viewing Macro-level or segmented audience targeting Sample-based audience Sampling error Fixed advertising message is prescheduled in space and time Utilizes 100% bandwidth Advertising and marketing research functions conducted separately Advertising and transaction functions occur through different channels