If an Advertisement Runs Online And No One Sees It, Is It Still an Ad? Empirical Generalizations in Digital Advertising Stephanie Flosi comScore [email protected] Gian Fulgoni comScore [email protected] Andrea Vollman comScore This study presents findings from three charter studies involving leading global advertisers in three key geographical regions: the United States, Europe, and Canada. The goal of the research was to identify and better understand the incidence of suboptimal digital campaign delivery as it pertains to viewability, audience delivery, geographic targeting, and brand safety. Through an evaluation of the study findings, several significant empirical generalizations emerged, and this article highlights these generalizations and discusses their implications for the digital advertising ecosystem. [email protected] INTRODUCTION and to uncover opportunities for improvement, According to the Interactive Advertising Bureau comScore has conducted research in the United (IAB), online advertising spend in the United States, Europe, and Canada. The current study dis- States exceeded $36 billion in 2012—a 15-percent cusses empirical generalizations drawn from the increase over the prior year that represented a research involving dozens of the world’s leading growth rate five times faster than all other media marketers. combined. Similar growth trends also were seen in Canada (up 16 percent in 2011) and Europe (up BACKGROUND 15 percent in 2011). Across the globe, digital media have become an This strong migration of advertising spend- important component of many advertisers’ mar- ing to digital across the globe is likely the result keting mix. Just as there has been exceptionally of increased confidence in the medium, which has tremendous growth in terms of the volume of been enabled by technological advancements that digital advertising, the digital ecosystem itself also make the buying and selling of online media more has experienced a profound evolution. From new efficient, and measurement tools that reveal the advertising formats and placement strategies to effectiveness of digital advertising. new delivery systems and advertisement-targeting Today, it is largely accepted that digital is an effective advertising medium for both direct 192 JOURNAL technology, it has become challenging for marketers and researchers to stay up to speed. response and branding advertising and that it can Historically, the measurement of digital adver- lift both online and in-store sales. Still, given the tising has not kept pace with these kinds of complex nature of the online advertising ecosys- complex changes. The transactional focus has tem, questions remain about potential sources of been on measurement of gross impressions deliv- online advertising waste and opportunities to lev- ered, as opposed to those advertisements that actu- erage digital in an even more efficient manner by ally were seen by consumers in a particular target improving the accuracy of ad delivery. segment. Until these questions are addressed, it is likely In fact, the primary metric used to buy and sell that digital will not be able to achieve its full poten- online advertising has been the number of delivered tial as an advertising medium. To better understand (or served) advertising impressions. As advertis- issues associated with digital-advertising delivery ing platforms, formats, and delivery technologies OF ADVERTISING RESEARCH June 2013 DOI: 10.2501/JAR-53-2-192-199 empirical generalizations in digital advertising DATA SOURCES AND RESEARCH DESIGN Empirical Generalization To draw the empirical generalizations, The broad market adoption of holistic campaign validation will allow the online advertis- comScore leveraged validated Campaign ing ecosystem to reach its full potential by helping to eliminate impressions that are EssentialsTM (vCE), an integrated comScore paid for but never have an opportunity to be seen and affect behavior. A viewability solution for complete campaign-delivery metric will provide the ecosystem with a currency by which to measure campaign deliv- validation and in-flight optimization. ery and to provide the first truly cross-media-comparable metric for use in effectively Unlike existing single-point solutions, building multi-platform campaigns. vCE provides an unduplicated accounting of impressions delivered across a variety have evolved, however, it has become The empirical generalizations discussed of dimensions, such as advertisements obvious that not all online advertisements in this study focus on the following core delivered in-view, in the right geography, delivered actually have the opportunity to dimensions of advertising delivery: in a brand-safe environment, and with be seen. Moreover, in some instances, they NHT removed. It also evaluates the degree were not accurately delivered based on the • reach and frequency: the percent of the to which validated impressions reached buying specifications (e.g., they are deliv- population that was reached with at the correct campaign target audience. It ered to the wrong demographic segment, least one ad impression, and the aver- should be noted that while vCE typically in the wrong geography, delivered near age number of advertising impressions is used for in-flight optimization, for the brand-unsafe content). delivered per individual; purposes of the current study, optimiza- Until recently, it was impossible to properly validate delivery across the relevant dimensions because the appropriate measurement technology did not exist. • target audience: the extent to which validated impressions were delivered to the correct demographic target audience; tions were not implemented because the intent of the research was to better understand sources of sub-optimal delivery and identify opportunities for improvement. now • in-view: defined according to IAB The validation module of the vCE tool exists to validate whether an advertise- guidelines as an advertising impression has received accreditation from the Media ment was delivered as intended based on with at least 50 percent of the advertise- Rating Council. The accreditation includes a variety of key dimensions, including ment’s pixels in the user’s viewport for the measurement of viewability, brand one second or longer; safety, in-country geographic delivery, Advanced technology, however, • viewability, • target audience, • target geography, • brand safety, and • non-human traffic (NHT). The use of “validated” impressions—specifically, impressions delivered in-view, in the right geography, in brand-safe content and with NHT removed—provides a more accurate picture of advertising delivery compared to gross impressions. Furthermore, when validated impressions are used to calculate validated gross rating • brand safety: advertising delivery on sites deemed not appropriate for brand advertising due to objectionable content considered to be in violation of brand safety; engagement, and the removal of NHT. This measurement includes certain advertisements delivered via cross-domain iframes, which account for a significant percentage of U.S. display advertising impressions but also have proved particu- • target geography: measured by country; larly challenging to the research industry from a measurement standpoint. and • NHT: impressions served to non-human agents, as per the IAB spiders and bots list and advertisements that were served to users via illegitimate methods or content. METHODOLOGY Census Measurement For every campaign under measurement, a single advertising tag was placed on each creative execution to provide a com- points (vGRPs), the metric provides a more plete view of campaign delivery. This accurate comparison to traditional GRPs The current study sought to uncover tag provides detailed information at the than the broadly used gross digital GRP, sources of waste, identify opportunities for publisher-, creative-, and placement-level which can include wasted advertisements improvement, and ultimately enable digi- about impression delivery, enabling the (e.g., impressions that were delivered but tal to reach its true potential as an effective reporting of viewability, geographic deliv- never had a chance to make an impact). and efficient advertising medium. ery, brand safety, and NHT (See Figure 1). June 2013 JOURNAL OF ADVERTISING RESEARCH 193 what we know about advertising ii Duplication and inconsistency typi- accuracy of delivery of digital media plans “cross-domain” iframes that historically cally arise when disparate data-collection for more than 120 advertisers and agencies have been nearly impossible to measure. sources are merged, which can negatively in 28 countries using its Campaign Essen- impact the quality of the analyzed data. tials service. These studies measured the EMPIRICAL GENERALIZATIONS OF Because the methodology in the current reach, frequency, and demographic deliv- DIGITAL ADVERTISING study relies on a single tag, issues associ- ery of thousands of digital campaigns. EG1 ated with duplicate impression counting More recently, comScore also has con- Because cookies are deleted regularly by are eliminated, offering a more accurate ducted studies using its vCE service that a substantial proportion of Internet users, view of campaign delivery. involved dozens of leading global adver- their use leads to large errors in the meas- tisers in the United States, Europe, and urement of the number of unique visitors Unified Digital Measurement Canada to better understand the inci- to a Web site and in the delivery and meas- To understand the extent to which adver- dence of sub-optimal digital campaign urement of planned reach and frequency tisements are delivered to the right target delivery in terms of viewability, accuracy for digital ad campaigns. Because of cookie audience, the authors used comScore’s of targeting, geography, brand safety, and deletion, the use of Web-site server data global 2-million-person panel in conjunc- NHT. that leverages cookies to estimate unique tion with census server data. The use of Key details of the vCE studies included the panel enabled a fuller understanding visitors to a Web site will overstate the true number of unique visitors. In the United of the audience that was exposed to each • 40 advertising campaigns for a group States, this overstatement was found to be campaign, including demographic infor- of advertisers that included Kellogg’s, by a factor of 2.5. The measurement of the mation, such as age, gender, household Kraft, Ford, Roger’s Wireless, Volks delivery of advertising campaigns based income, household size, and ethnicity and wagen, Unilever, and American Express; on the use of cookies will underestimate information regarding different behavioral • measurement of approximately 3 billion actual delivered frequency by about 2.5 segments (See Figure 1). The use of panel in conjunction with cookie data (as opposed to ad impressions; times and overestimate actual reach by the • measurement across more than 600,000 a solely cookie-based approach) eliminated same amount. Web sites; and High rates of cookie deletion occurred errors associated with cookie deletion and • 100 percent of the ad impressions deliv- in all countries included in the current multiple-user machines—practices that can ered in iframes, including a majority of study. On a global basis, comScore studies skew results dramatically. The panel used in the current study is the same comScore panel that powers comScore’s audience measurement capabili- A single ad tag is appended to every creative to provide a complete view of campaign delivery ties, including its Media Metrix service. All panelists have opted in to panel participation and have given comScore explicit per- CENSUS PANEL mission to measure their online behavior. A comScore’s global two million person panel is used in conjunction with census data to validate target audience delivery unique and critical component of the panel is Unified Digital Measurement, a measurement methodology that merges comScore person-level panel data with raw census- in-view level server and advertising-delivery data Geography (collected through tagging) for a complete Brand safety 360-degree view of consumer behavior at the Non-human traffic person-level across markets (See Figure 1). RESEARCH DESIGN Over the last 5 years, comScore has conducted more than 4,000 studies of the 194 JOURNAL Audience – – – – – – Age Gender HHI HH size Ethnicity Behavioral Segments Figure 1 vCE Methodology Through Single Ad Tag and Unified Digital Measurement OF ADVERTISING RESEARCH June 2013 empirical generalizations in digital advertising High rates of cookie Because of Multiple Users on a Computer, Cookies Alone Can’t Always Accurately Identify who is Using a Computer at any Given Point in Time deletion occurred in Percent of people Percent of computers all countries included 3+ Users 15% in the current study. 3+ Users 32% have found that Web site server (i.e., “first- 2 Users 26% party”) cookies were deleted from an 1 User 36% 1 User 59% average of 28.5 percent of computers in 2 Users 32% a month while ad server (“third-party”) cookies were deleted by 32.5 percent of computers. The rate of deletion was higher for ad-server cookies (average of nine times per month) than for Web site cookies (five times per month). In the United States, 28.5 percent of first- Figure 2 Use of the Same Home Computer by Multiple Individuals party cookies and 35.8 percent of thirdparty cookies, on average, were deleted compounded cookie-related behavioral segments typically resulted in in a month, with first-party cookies being problems: the same person using multiple inaccurate advertisement-delivery infor- deleted six times per month and third- devices (e.g., a person using both a work mation due to targeting errors caused party cookies being deleted nine times and a home computer and, therefore, hav- by cookie deletion, multiple devices per per month. The lower deletion metrics for ing more than one cookie per person for a user (which leads to multiple cookies per first-party cookies likely is because of their given Web site or advertising campaign), individual), and multiple users on a higher utility to the user (e.g., they elimi- and different people using the same com- given computer (which makes it difficult nate the need to re-enter “sign-in” infor- puter (resulting in more than one person for an ad server to know which person is mation at some Web sites). per cookie). using a “cookied” machine at any point in by other Every time a cookie is deleted from a Trying to tie a cookie to a specific indi- computer following a Web site visit or vidual is problematic. To begin, comScore When targeting to only one variable the receipt of an advertisement, when the data showed that the average U.S. com- (e.g., women), cookie-based delivery accu- Web site server or the ad server “sees” puter user utilized 1.29 different comput- rately reached its target 70 percent of the that same computer again, the server ers across work and home locations and, time. When two variables were used (e.g., will “think” it represents a new unique as a result, an individual had a 30-percent women ages 25–54), the targeting accu- visitor or a new recipient for an advertise- likelihood of having more than one cookie racy decreased to an average of 48 percent. ment and will count it as a new unique vis- assigned to him or her. And when three variables were used (e.g., time). itor or deliver it an advertisement, thereby comScore panel data also showed that women ages 25–54 with children younger leading to over-statement of the true 64 percent of U.S. Internet users utilized than 18) the targeting accuracy fell to only number of unique Web site visitors and multiple user machines, which means 11 percent (See Figure 3). an over-delivery of advertising frequency that a given cookie potentially could be Using cookies to target behavioral seg- and a corresponding under-delivery of assigned to multiple individuals on that ments (e.g., visitors to travel sites, viewers reach. computer (See Figure 2). of food preparation content, etc.) also is EG2 EG3 campaigns, the current study showed Measurement errors caused by Inter- Using cookies to target digital adver- that the average campaign accurately net users deleting their cookies are tisements to specific demographic and reached the correct targeted behavioral subject to significant error. Across all U.S. June 2013 JOURNAL OF ADVERTISING RESEARCH 195 what we know about advertising ii Low % of Ads Delivered to Demographic Target 100 Average In-view rates varied High dramatically by site, ranging 80 from a low of less than 70% 60 1 percent to a high 48% 40 of 100 percent. 20 11% 0 1 Variable 2 Variables 3 Variables Importantly, pricing for advertising Figure 3 Percentage of Advertisements Accurately Reaching Demographic Target Based on Number of Targeting Variables* most often is not derived based on view- *Demographic variables can include: age, gender, household income, and/or number of children in the household. Due to sample size, a meaningful range could not be calculated for campaigns with 3 demographic variables. placements with a correlation of only 0.19 ability rates. According to the current study, in-view rates varied across sites and between viewability and price paid (cost per thousand impressions—CPM) for the advertisements (See Figure 4). segment only 36 percent of the time, with rates of 48 percent to 96 percent; “below a wide range of accuracy from 23 percent the fold” messages (images viewable EG5 to 67 percent. to computer users only after scrolling) The number of advertisements delivered resulted in viewability rates of 3 percent to next to unsafe content is not substan- 67 percent. tial, but the authors of the current study EG4 Approximately one in three delivered digi- Lack of viewability above the fold typi- tal advertising impressions never have the cally occurs when a user scrolls past the opportunity to be seen (i.e., are never vis- advertisement (or moves to another page) In the current study, the majority of digi- ible to the end user), with the viewability before the message was fully rendered. tal campaigns in the United States (72 per- rate varying dramatically by site. Lack of visibility below the fold generally cent) and Europe (67 percent) saw some occurs because a user does not scroll down impressions delivered alongside unsafe far enough to see the advertisement. content, which is defined as adult content, For the campaigns measured in the current study, comScore evaluated view- believe that the absolute number of consumers impacted is significant. ability rates on an individual campaign Viewability rates tended to be higher hate sites, or other types of digital content basis and found that, on average, 30 per- on large sites when compared to mid-size deemed unsuitable by leading advertisers cent to 37 percent of all served advertising or “long-tail” sites. Even within the larg- (e.g., spam, malware, child abuse, copy- impressions in the United States, Europe, est sites, however, the viewability rate right theft). and Canada never were actually view- averaged only 74 percent across the top Fortunately, because of the use of third- able by the end user. In other words, the 100 U.S. sites and 63 percent across the party advertising-blocking technologies, advertising impressions did not deliver an top 100 European sites. Within the long only a small percentage of overall adver- “opportunity to see.” In-view rates varied dramatically by site, ranging from a low of less than tail, viewability rates averaged 61 per- tising impressions (0.01 percent) are deliv- cent in the United States and 55 percent ered in brand-unsafe environments. The in Europe. absolute number of consumers exposed 1 percent to a high of 100 percent. Pla- Across all studied advertising cam- to advertisements in these unsafe envi- cing an advertisement “above the fold” (a paigns, the average in-view rate hovered ronments, however, is substantial enough term borrowed from the newspaper at around two of three impressions, mean- (5,000 per U.S. campaign and 1,400 per industry to describe images that appear on ing one in three impressions were never European campaign) to concern many a computer screen) generated viewability seen and, therefore, were wasted. marketers (See Figure 5). 196 JOURNAL OF ADVERTISING RESEARCH June 2013 empirical generalizations in digital advertising EG6 an individual campaign basis, how- the inability for an advertisement to be On average, geo-targeting of digital ever, error rates can be high, reaching delivered in its intended geography was campaigns at a country level using IP 27 percent in Europe and 15 percent in related to limitations in the ability of IP addresses is quite accurate, with only the United States. The current study addresses to accurately designate geo- 4 percent of advertisements in the United showed that geographic-targeting accu- graphic location and errors in complex States and 7 percent in Europe falling racy diminished as the target market advertisement buying and selling pro- outside of the intended geography. On became smaller in area. In such cases, cesses (See Figure 6). EG7 An equally weak correlation was also observed between CPM and ability to hit a primary demographic target NHT, including fraud, is a significant challenge for accurate digital advertising delivery, and it is not adequately elimi- Correlation of In-View Rates & CPM 300 nated by industry blacklists of known R2 = 0.0373 robots or fraudulent operators. 250 In the U.S. comScore study, known industry black lists identified between CPM Index 200 US 150 100 50 0 0 20 40 60 80 100 4% of ads were served outside geography. Up to 15% of ads on a given campaign % of Ads In-view Figure 4 Viewability Does Not Relate to Price Charged for Digital Ads EU 7% of ads were served outside geography. Up to 27% of ads on a given campaign 28% 72% Some ads in inappropriate content No ads in inappropriate content 92,000 people exposed to adult content and/or hate sites Figure 5 Percentage of U.S. Advertising Campaigns with Impressions Delivered Next to Content Deemed “Not Brand Safe” Figure 6 Percentage of Advertisements Served Outside of Intended Geography in the United States and Europe June 2013 JOURNAL OF ADVERTISING RESEARCH 197 what we know about advertising ii 0.03 percent and 0.49 percent of campaign • Need for Validation Advertisers need to understand adver- impressions as being delivered to NHT. The recent comScore studies demon- tisement delivery along each of these However, when using comScore’s more strated that each dimension of adver- core dimensions. And they also should sophisticated methodology to detect tising delivery—viewability, audience require a holistic, unduplicated view and filter out NHT, a higher 4-percent to targeting, geographic targeting, brand of total campaign reach and fre- 11-percent of impressions are identified safety, and NHT—had a substantial quency. To achieve this un-duplicated as being delivered to NHT. The comScore impact on whether an advertisement accounting of delivered impressions, methodology goes well beyond industry achieved its intended objective. advertisers must find a way to elimi- blacklists of known robots. By its very Each dimension, therefore, should nate all the wasted time and error asso- nature, this type of non-human activity be a central component of advertising ciated with merging disparate data is highly adaptable to the countermeas- delivery measurement and validation. sources. ures applied against it, and it is no surprise that published industry blacklists There have been some clear top performers. In total, this campaign effectively delivered to the target and efficiently balanced the exposures. capture only a marginal amount of this activity. Impressions Impressions Avg. Frequency Targeting Index % on Target Total Campaign KEY IMPLICATIONS AND CONCLUSIONS Given the empirical generalizations derived from this research, there are a varCampaign Total 216 58% 3.2 advertising ecosystem. Among them: Publisher 1 163 44% 4.0 • In-Flight Optimization of Media Plan Publisher 2 225 60% 3.1 Publisher 3 140 37% 3.5 iety of key implications that the industry should consider as it relates to the online Increases Return on Investment (ROI) For advertisers and agencies, holistic inflight validation of campaign delivery provides an opportunity to understand delivery across all dimensions. It also There have been some opportunities. In this case many impressions are virtually being thrown away as the wrong audience is reached over and over again. allows time to efficiently make changes to the media plan (e.g., spending allocations across individual publishers) that can improve overall campaign results (See Figure 7). Kellogg’s used comScore’s Campaign Campaign Total Essentials service to obtain daily feedback of actual campaign delivery to adjust its spending by individual pub- Kellogg’s digital media plans based on market mix models revealed that the Impressions Avg. Frequency Targeting Index % on Target Total Campaign 105 53% 12.0 Publisher 1 97 49% 8.8 Publisher 2 25 13% 21.7 Publisher 3 92 47% 11.4 Publisher 4 143 72% 15.6 lisher in order to optimize frequency and improve targeting accuracy. Analysis of Impressions company realized a dramatic improvement—five- and six-fold—in ROI for two of its brands as a result of in-flight optimization of its digital media plans (See Figure 8). 198 JOURNAL Figure 7 Kellogg’s Experience in Optimizing its Delivery of Digital Media Plans OF ADVERTISING RESEARCH June 2013 empirical generalizations in digital advertising effectiveness, providing the first truly Brand 1 ROI cross-media comparable metric for use Brand 2 ROI in effectively building multi-platform 6x campaigns. 5X ROI Year 1 2X Year 2 3X ROI 1H 2011 Year 1 Stephanie Flosi is a senior analyst at comScore Year 2 in the marketing communications group, where 1H 2011 she manages product marketing for comScore’s Brand Survey Lift suite of advertising effectiveness Figure 8 Kellogg’s ROI Impact after Optimizing Digital Media Plan In-flight in 1H2011 solutions. She supports various thought-leadership initiatives at the company and has co-authored several comScore whitepapers including NextGeneration Strategies for Advertising to Millennials • Viewability Is Necessary but Not models, resulting in better proof of (2012), Surviving the Upfronts in a Cross-Media Sufficient digital advertising effectiveness—ulti- World (2012), and comScore’s annual Digital Future Viewability is a critical component of mately bringing more branding dollars in Focus reports. campaign validation. Given the vari- to the Internet. ance in in-view rates across sites and Gian Fulgoni is the co-founder and executive chairman campaigns and the importance of • Broad Adoption Will Allow the Digi- of comScore, Inc. (NASDAQ:SCOR). Previously, he “opportunity to see” in developing a tal Marketing Industry to Reach Full was president and CEO of Information Resources, cross-media comparable GRP, campaign Potential Inc. During a 40-year career at the c-level of corporate delivery metrics must include viewabil- The broad market adoption of holis- management, he has overseen the development of ity. Such a treatment would generate a tic campaign validation will allow many innovative technological methods of measuring service analogous to television audience the online advertising ecosystem to consumer behavior and advertising effectiveness. He guarantees. reach its full potential. It will help is a previous contributor to the Journal of Advertising Research. In addition to viewability measure- eliminate impressions that are paid ment, it is also necessary to validate for but never have an opportunity to campaign delivery based on geography, be seen and affect behavior. It will offer Andrea Vollman is senior marketing director of brand safety, and NHT. Eliminating more accurate valuing of impressions advertising effectiveness at comScore. Her research inventory that does not have a chance that deliver on their promise across interests span all facets of online and cross-media to make an impact results in more a broad spectrum of campaign attrib- advertising effectiveness and measurement and effective/efficient campaigns and less utes. And a viewability metric will evaluation of brand communications and creative waste for advertisers. It also provides provide the ecosystem a standard messaging. Her work has appeared in the Journal of more accurate metrics for market-mix set of currency by which to value Advertising Research. June 2013 JOURNAL OF ADVERTISING RESEARCH 199
© Copyright 2026 Paperzz