Understanding Digital Advertising Attribution


Published on

In the life sciences/pharmaceuticals industry, targeting digital advertising spends is a sophisticated undertaking. Our methodology - developed in conjunction with Adstrategist - takes into account the impact of influencing health care providers, combining analytics for techniques such as clickthrough, last click, weighted, regression, and panel testing, plus buying cycles and page location.

1 Like
  • Be the first to comment

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Understanding Digital Advertising Attribution

  1. 1. • Cognizant 20-20 InsightsUnderstanding DigitalAdvertising AttributionSubjective metrics used to understand digital advertising effectivenessare typically insufficient to correlate spend with results. A new analytics-driven approach is emerging to help life sciences companies measure adimpact by correlating data from various online advertising modalities. Executive Summary reality of measuring online behavior became better understood. Individuals are wary of clicking Spending on digital advertising has risen each on ads because of concerns about pop-ups, year even as TV and print advertising declines.1 security and tracking. Facing a dramatic prolifera- Search, the darling of the digital world, is becoming tion of ads, most people have become desensi- saturated as advertisers compete for position in tized to them. the finite number of searches performed each month. It is digital advertising and the rising tide Most important, there is no “best-practice” of social media that will be driving business in the method of assigning value to digital advertising future. with a solid analytical foundation. As a result, companies that operate ad serving networks are Digital advertising deployment creates obstacles open to accusations that their systems can be in accurate measurement of results, and vendors “gamed” to make their services appear to perform have a stake in promoting the technique that best better. Current attribution methods provide the favors their service. Understanding the value of most value to the last click or the last ad a person digital advertising requires understanding of how viewed. They aggregate or sample data, and miss value is attributed to each activity. Understand- the important sequencing and timing of each ing the true worth of these channels is key to any person’s exposure to advertising. company’s success. What an advertiser wants to know is how digital Opportunity and Challenge advertising really affects the bottom line. If you Digital advertising had seemed to be the answer spend more, do you get more? How do you know to a marketer’s dream, providing concrete data what works and what does not? Do combinations on the number of eyeballs viewing ads and the of advertising work best? How does advertising number of people acting on them with a click. work with search and social media? The final Unfortunately, the dream was short-lived, as the analysis must take into account every ad, search, cognizant 20-20 insights | june 2012
  2. 2. tweet, etc. that a person has seen and assign a Digital Display Advertising value based on the actual influence that it had More than 80% of all digital ads are served by ad on the desired outcome. Subjective measures, servers such as DoubleClick and Atlas.2 Content such as awareness, brand favorability and brand Web sites (publishers) such as CNN rent out ad attributes, are typically insufficient to correlate slots on their pages to the ad servers. When the spend with results. page is displayed, the ad server is asked for an ad to fill the slot. Advertisers buy inventory from the Influencing Pharmaceutical Decisions ad servers, which then display the ad when the Online advertising of pharmaceuticals has its own page is requested. challenges beyond regulatory compliance. While doctors prescribe medications, advertising to The price can be based on impressions served, their patients can influence the brand chosen or clicks or even by acquisition — only paying when make them aware that they may have a treatable the viewer buys something. Typically, however, problem in the first place. When “Googling” large volume advertising or complex sales use or searching a disease or the impression model commonly known as CPM So, much like print symptom,to a brand site or directed patients are often (cost per 1000 impressions) or simply a flat rate to buy that ad space for a period of time. If the ad and TV, there is no to an information site such servers know that the ad location will be viewed definitive proof that as WebMD where there is an more often than the number of impressions that an ad was effective, opportunity for product ads to create brand awareness. the advertiser has bought, the advertiser’s ads will be shown on a random basis a percentage of but with digital ads The process for a patient to the time.there is a digital trail reach a medication decision In addition, ad networks can negotiate with other that can provide may take weeks orare many which means there months, ad networks to find ads to display. Advertisers some insight. opportunities for advertising can arrange for their own ad rotation within the to affect the outcome. Since inventory that they have purchased. So, much like the prescription is often written in the doctor’s print and TV, there is no definitive proof that an office rather than online, pharmaceuticals adver- ad was effective, but with digital ads there is a tising is usually impression based (as opposed to digital trail that can provide some insight. per click or per acquisition). The effectiveness of Search Advertising each impression becomes even more critical and the analysis even more complex. Only sophisticat- There is tremendous focus on search because ed attribution can determine which combinations current Web analytics show that much online of advertising really provide a high ROI. traffic originates from Google, Bing and Yahoo. However, a closer look at search terms reveals Doctors and health care providers (HCPs) search that many searches are for a brand name — the Web with different goals, and their susceptibil- meaning the visitor has already been influenced ity to advertising varies. Health care providers are some other way, possibly by an online advertise- highly valuable as specifiers and recommenders ment. The most convenient way for a person to of pharmaceutical products, so targeting them follow the ad without clicking (and hence avoiding effectively has a positive impact on the business. security concerns and unwanted pop-ups) is to Detailed analysis without sampling can improve “Google” the brand name and click on the link. insight into how to influence the decisions of this Consequently, search is given far too much small but critical audience. Digital campaigns credit for driving traffic to the site, especially in for top-selling pharmaceuticals often involve attribution schemes that favor the last action. displaying billions of impressions per month. Big Search is ubiquitous, so the fight over relevance data techniques using a server farm or cloud ranking in Google has become a zero-sum game. computing enable every detail to be processed Understanding how ads influence search is key to in a timely fashion. These techniques offer the increasing online traffic. precision of real-world observations without the need to estimate. They also scale across large Measurement: It’s All Connected data sets to provide results in a timely manner. No doubt, the bulk of your site traffic comes from Google or specialized sites such as WebMD. And cognizant 20-20 insights 2
  3. 3. much of it appears to — if you consider only the Studies have shown that 99% of users never clicklast action. But analysis shows that the process on an ad but that those that do are very likely tois much more involved. Banner ads on portals click again. It is estimated that as many as 20% ofand ad networks drive brand awareness and clicks are accidental, especially among the olderdrive traffic to specialized sites. Specialized sites age range, and Click Forensics estimated that assuch as WebMD provide education on diseases much as 23% of clicks were fraudulent in Octoberand direct patients to further information on the 2010.3brand site. An ad that appears while a visitor isresearching diseases and treatments triggers Last Clicka branded search via Google. Each one of these Last click only counts clicks immediately beforesteps is important and plays a part in the decision a conversion event, which eliminates credit forprocess. Each one involves a different cost and accidental and repeat clicks. But this method stilleffort to manage. Attribution helps discover the assigns all the credit to a click and ignores allmost effective mix for your ad and search spend. the preceding advertising. People use search for convenient navigation and click on paid searchAttribution Techniques advertisements when they would otherwise haveThere is a wide range of attribution methods clicked on an organic link. Last click attributionin use today. Many rely on the judgment of the overvalues clicks in general and paid search inadvertiser as to what matters most, resulting in a particular. An Atlas study showed that betweenpredetermined answer of dubious value. Even the 93% and 95% of audience engagement withmore sophisticated methods often use sampling online advertising receives no credit at all whentechniques that make their results difficult to advertisers review the ROI on their campaignscompare against the actual numbers. Advertis- because of the use of last click attribution.4ing activities work together, so attribution must Last Impressionunderstand the relationships between the variousforms of advertising employed (see Figure 1). Last Impression is often called Viewthrough. Dou- bleclick discussed this method in a 2008 paperClick-through Rate suggesting that 100% credit should be given toCTR, or click-through rate, has long been the the last ad seen before the conversion event. Thestandard for Web advertising because it is easy suggested window was one month, which tendsto calculate. CTR is simply the number of clicks to increase the importance of digital advertisingon an ad divided by the number of impressions impressions. This is in the interest of ad servingserved. As clicks decline and click fraud increases, companies that are paid by impression. Thethis measure has become increasingly irrelevant. method still gives all the credit to a single event, which is unlikely to be the true case. There is noDigital Ads: “Different Strokes” of Measurement Attribution Method Description CTR Click-through rate is the number of clicks per impressions seen. Last Click The last ad clicked before the conversion event is given the full credit. Last Impression or The last ad that was seen before the conversion is given the credit. Viewthrough Weighted All ads that were seen prior to the conversion are given some credit. Typically, the first and last ads seen are given most of the credit while the ones in between get less. Often referred to as “U” weighting. Regression Credit is assigned based on an algorithm usually using some type of regression that considers some or all of the possible factors. Panel Outcomes are measured in experiments in advertising against a known audience and control group.Figure 1 cognizant 20-20 insights 3
  4. 4. way to know if the person actually saw the ad Both weighted and regression methods look atbecause it was on a part of the page not on the the activity of each individual cookie — and therescreen (i.e., below the fold) or whether he or she can be many to account for. This often leads tonoticed it if it was visible. Recently, retargeting sampling, which may pass statistical significancehas become a popular technique that attempts measures but miss the important key events.to deliver ads to people that are known to have Schemes that do not use sampling avoid thisvisited the site already, in the assumption that problem and allow the results to be more easilythey are more likely to convert than the general compared and matched to other data.population. Panel TestingViewthrough tends to increase the importance of Panel testing is potentially the most preciseretargeting because those ads are served to the method of determining the effectiveness ofpeople most likely to visit and convert. In general, online advertising, search and social media. Thelast impression attribution can encourage ad experiment involves signing up millions of peopleagencies to target cheap media buys at people who will allow themselves to be tracked. Then A/Balready likely to convert in order to get credit for testing is performed, using a control group thatas many conversions as possible rather than influ- does not receive ads and is compared to a testencing new people to buy. group that does. If the experiment is conducted correctly, the differenc-Weighted es between the controlMost weighting schemes attempt to look at the group and test group can Different types ofactivities of individual cookies as they progress be tested for statistical purchases have verytowards a conversion (or not). Weighting schemes significance and the likely different decisionassign values to certain events and add them up reasons identified.to provide a relative measure of that item’s value. patterns, and a singleA popular weighting scheme is really an extension The problems with panel weighting schemeof “last view” and is known as “U” weighting. testing are mainly the cannot possibly be logistics and cost ofFor each cookie that converts, it assigns a weight doing it. Typically, a study right for all.to the first ad that the viewer sees, a larger such as this takes manyweight to the last ad seen and lesser weights months and a huge amount of effort and cost toto the ones in between. This seems like a very enlist the participants, construct the experimentrational approach, and it tends to deliver answers and analyze the results. Many advertisers areyou would expect. Unfortunately, the results reluctant to do it because it implies that theusually provide very little real insight because control group will not receive ads and possiblythere is no analytical basis for choosing the value buy less, directly affecting their bottom line whileof the weights or even what should be weighted. the experiment is being run. These studies tendDifferent types of purchases have very different to be carried out on a one-time basis to try todecision patterns, and a single weighting scheme understand how people purchase a particularcannot possibly be right for all. product or service and then be used as insight into daily operations rather than to evaluateRegression specific campaigns or ad activities.Regression attribution uses an algorithm tomodel the behavior of cookies. This can then be Rich Data to Measure What Matterstested against the outcome of a campaign that The best way to understand how digital ads,has already been run to test the validity of the search and social media (touchpoints) affect indi-method. Any number of aspects may be used in viduals is to follow each individual’s experiencethe model. The advantage of this approach is that with them. Without the person’s consent to installit actually attempts to determine the weights or special software on his computer, this is not reallyimportance of events analytically rather than possible today. However, each ad or search leavesby subjective weights assigned by the user. The a fingerprint that can be pieced together to build aalgorithm does not require a control group, and it picture of how he or she sees the advertiser’s adscan often use the changes in campaign strategy and arrives at the target site (or not). Broad-basedas the basis for comparison. The drawback is that brand advertising can actually be an advantage,there are numerous dimensions to be considered because promotions that span numerous sitesand decisions on which ones to use tend to be and networks increase the likelihood that a visitorsubjective. cognizant 20-20 insights 4
  5. 5. will see them and hence leave a trail. This is par- Sites and Sequences ticularly interesting for HCP visitors. The number Most serious digital attribution methods concen- tends to be small so that each individual’s activity trate on the sequence of sites where a cookie can be analyzed while retaining anonymity. sees ads or social media along with searches and Web site visits. The ad site is important Adstrategist, a strategic partner of ours, provides because different sites attract different demo- a solution that builds a complete history of graphic profiles, each with their own propensi- everything known about a person’s interaction ties. However, large portal sites such as Yahoo do with ads, search and the target site. Analytics not provide this granularity and ad networks may utilize the third-party cookies that are already obfuscate the picture still further because it may being set by ad servers rather than requiring be difficult to tell exactly where an ad originated. new tagging or other forms of tracking. Every Some attribution analytics measure creative dimension of an ad impression relating to the treatments as well as, or instead of, the site where publisher site, ad network, page, ad location, the ad was seen. The “creative” served will have creative and time of display is recorded and a specific appeal and goal, which can be used as used in the analysis. The approximate location a variable in the attribution algorithm. Different of the person down to the metrocode level is creatives will appeal differently to different also known and can be used demographic groups, and hence will be related toMost serious digital to match online activity with the site where they are shown.attribution methods off-line sales and other forms of advertising such as print Page Location concentrate on the and TV. Adstrategist’s proprietary research and analytical sequence of sites method have found that the specific page where The resulting data set revealswhere a cookie sees everything known about every an ad is seen is relatively unimportant except for broad areas of the site (see Figure 2). Analysis ads or social media person that saw an ad, made indicates that the ad location on the page isalong with searches a search or visited the target mainly important for its size and position, partic- site. This data set is analyzed and Web site visits. to discover the journey of each ularly whether it is below the fold or a banner ad. successful cookie to compare Reach and Frequency with all of the paths that lead nowhere. This is a The number of impressions, or touchpoints, highly effective use of big data, not just an exercise experienced by a cookie is important. There are in capturing and managing data in the vague hope conflicting trends at work. Reach and frequency of finding useful insights at some point. (R&F) is an accepted principle of off-line adver- tising that indicates there is an optimal number Digital Ad Placement: What You See Is Not What You Get Campaign Goals For example, branding, information, action, conversion. Where Publisher site, ad network. Location Page, placement. Ad size, ad location. What Ad type (rich media, etc.). Creative Message/Tactic How Much Reach, frequency. When Recency, time spacing. Attribution window. Interactions Sequencing, combinations, assistors. Figure 2 cognizant 20-20 insights 5
  6. 6. of times people must see an advertisement treatments can work very differently in differentbefore they are “converted.” The chance of them parts of the country.converting rises as this number is approached.Once a person has reached the optimal number of Pathsad views, the chances of them converting do not Essentially, attribution is about looking atincrease much with more viewings — so further converting and nonconverting visitor paths andads are wasted cost. On the other hand, people deciding what made the difference.become bored with ads, especially on Web sites.Several studies suggest that users may not even Just looking at simplified representations of thebe aware of ads after a time. However, this effect cookie paths can offer significant insights intotends to wear off if ads are not seen for a while, so how cookies see ads and which sequences areless is more, at least for a time. What most adver- most effective (see Figure 4). It is easy enough totisers don’t understand, or cannot calculate, is condense the sequences by publisher sites wherethat when they buy a percentage of the inventory ads were seen but it gets more complicated whenof an ad location, the probability of a visitor other dimensions such as creative treatments,seeing an ad “N” times changes dramatically — so timing, etc. are included.they usually buy too few or too many impressions Nevertheless, graphical rep- Attribution analytics resentations of the paths areto achieve the optimum R&F (see Figure 3). the essential starting point can reveal the relevantTime Frame and Geography for meaningful attribution. window for yourThe time frame over which attribution is One Size Fits None business and yourconsidered is also important. Ad servers want Buying cycles vary widely advertising goals.advertisers to believe the window is large,because that means that ads will be valued more. depending on the productThe reality is that it varies with product type from being purchased. The goal of the digital advertis-days to months. Attribution analytics can reveal ing may be to inform and persuade rather thanthe relevant window for your business and your to trigger a transaction. Pharmaceuticals adver-advertising goals. tising is frequently used for information and brand-building to help people recognize treatableThe geographic location of the cookie may also conditions and understand treatment options.play a role. TV and print advertising on local This type of digital advertising can have a signifi-stations and papers usually direct people to a cant impact on the business but it is more difficultspecific URL but it is common for responders to measure than direct response ads whereto simply enter the brand into Google instead. conversion is immediate.We have found that search terms and creativeReach Frequency Impressions vs. Pageviews at 50% Inventory 50% % People Viewing N Times 45% 40% 35% 30% 25% Page View % Ad View % 20% 15% 10% 5% 0% 0 1 2 3 4 5 6 7 8 9 10 Number of Page ViewsSource: AdStrategist Reach-Frequency Report.Figure 3 cognizant 20-20 insights 6
  7. 7. Assessing Page SequencesSource: AdStrategist Solution Conversion Report.Figure 4By contrast, booking a cheap hotel room is a Results are only meaningful when compared withfairly quick decision because (a) the person needs the goals of the campaign. Is the purpose to drivea room and (b) price and location are the main people to the site for action or a brand-building andfactors. A few visits to Hotels.com and Kayak.com awareness program? If the purpose of a creativeand a few visits to the hotel site over a few is to inform, why evaluate the campaign’s click-hours or days is all that is needed to make a through rate? There are many external factorspurchase decision. The goal is to sell your room that can affect the campaign. Some are availableto the shopper before he or she buys from your in additional data feeds (e.g., TV campaigns) butcompetitor. factors such as a patent expiring, a production shortage or a related article in The New YorkContrast this with the choice of a diabetes drug, Times may dramatically affect the traffic viewingwhich may be a lifelong decision and involve ads and visiting a site.weeks or months of research on numerous Websites. The goal of the advertising is not a sale but Needle in a Haystacka visit to a doctor who has also been influenced to Attribution is about discovering clear relation-favor your medication through directed advertis- ships between digital advertising and conversionsing activities. Because the goals and time frames that will lead to improvement in marketing effec-are very different, the attribution algorithm must tiveness. But it is not that simple. Advertisingbe flexible if it is going to deliver meaningful agencies usually run the campaign and achieveresults. varying levels of success through experience, advice and trial and error without any analyticalCommon Sense basis for understanding their success. Attribu-Simply analyzing data is not enough without an tion algorithms that are introduced into matureunderstanding of the environment and context. campaign environments have to be looking for cognizant 20-20 insights 7
  8. 8. more subtle yet valuable improvements and tion picture that becomes more accurate withenhance rather than compete with what is being greater variances in campaigns and time.done already. Collecting the DataIf the goal is to bring people to the target site, Many solutions use tagging to track ad impres-it is clear that digital advertising works. If it sions, clicks and site activity. This involves costlyis to perform an event such as downloading a effort and maintenance and typically slows thecoupon to cover co-pay, the result is less clear. page load times. Ad agencies that are respon-Over the appropriate time period, more impres- sible for creatives are already overwhelmed bysions produce more visits at a declining rate. But the number of tracking tags they must implementeven that lift is small relative to the increase in and maintain. In fact, the same data is alreadyimpressions. Even a modest lift in prescription being collected by the ad servers and it can besales can pay for a lot of advertising, but dis- obtained and analyzed to see every impression,covering which additional impressions contrib- click and on-site action that a cookie experienced.uted most and in what combinations is the real There is no additional effort for the advertiser tochallenge. Drawing meaningful conclusions when implement the AdStrategistthere are thousands of similar nonconverting solution. No risk, no cost — justpaths for every converting path is both essential Drawing meaningful benefit.and nontrivial. Advertisers know that while spe- conclusionscialized sites such as WebMD seem to drive the What to Do? when there aremost traffic, that flow dries up without extensiveadvertising on ad networks and portals. Finding Even if there is a perfect attri- thousands of similar bution algorithm, there is stillthe correct balance is the key to maximizing the the problem of what to do with nonconverting pathsreturn on advertising investment. the results. Cookies arrive at for every convertingIn many campaigns, a large number of visitors the conversion via a series path is both essential of ad exposures, searches,arrive at the site and convert without seeing an visits and social media in and nontrivial.ad or performing a search other than a brandedsearch, which implies they were coming to the a particular sequence andsite anyway. Many attribution models go to great timing. Attribution may correctly model thatlengths to carefully model the cookies that see sequencing but there is little or no way to controlads. In other words, they are modeling the noise the sequence or timing of how a cookie will seewith a lot of precision. This turns out to be the ads in the future. So attribution must not onlybiggest challenge for many attribution models. have a way to measure its accuracy but provide useable recommendations for improvement.A simple segmentation of the results looks at thevisitors/convertors that see ads vs. those that Solid attribution analysis builds a picture ofdo not. As the number of impressions increase, how advertising campaigns deliver results andthe number of people that see ads will increase provides a platform for continuous improvement.and some of these will visit the site and convert. The next step is to use the predictive analyticsSo an increase in impressions will increase the capability to predict the outcome of digitalnumber of visitors that see an ad even if the ads campaigns.have no effect whatsoever. Subtracting this group Understanding the dynamics of digital advertis-provides a clearer view of the lift due to advertis- ing is an essential first step in understanding howing. A/B testing may not be an option but there effective it can be in your marketing mix. Thereare often significant variances in ad campaigns are plenty of ways to guess or assign value, but aover a few months. This is an implicit multivari- rigorous and flexible approach to attribution willate experiment that can be used to compare dif- enable you to find the patterns in the data thatferences in lift with different impression volumes, can inform smarter decisions and higher return.sites and creative treatments to build an attribu- cognizant 20-20 insights 8
  9. 9. Footnotes1 eMarketer, March 29, 2011, http://www.emarketer.com/PressRelease.aspx?R2 JP Morgan Management Presentation for DGIT, May 2012, http://go.dgit.com/rs/dgfastchannel/images/ JP%20Morgan%20Management%20Presentation%205%2016%2012.pdf3 Pellman, Paul, Adometry Press Release 2010, http://www.adometry.com/news/release.php?id=34 Microsoft Atlas Institute, 2008, Alltel Wireless Case Study.About the AuthorAlex Gochtovtt is a Principal with Cognizant’s Enterprise Analytics Practice, working within its DigitalAnalytics Center of Excellence. He can be reached at Alexander.Gochtovtt@cognizant.com.AcknowledgmentThe author would like to acknowledge the contributions of Douglas Watt, CTO with AdStrategist, aCognizant strategic partner that develops attribution and prediction technology for digital advertising,search and social media.About CognizantCognizant (NASDAQ: CTSH) is a leading provider of information technology, consulting, and business process out-sourcing services, dedicated to helping the world’s leading companies build stronger businesses. Headquartered inTeaneck, New Jersey (U.S.), Cognizant combines a passion for client satisfaction, technology innovation, deep industryand business process expertise, and a global, collaborative workforce that embodies the future of work. With over 50delivery centers worldwide and approximately 140,500 employees as of March 31, 2012, Cognizant is a member of theNASDAQ-100, the S&P 500, the Forbes Global 2000, and the Fortune 500 and is ranked among the top performingand fastest growing companies in the world. Visit us online at www.cognizant.com or follow us on Twitter: Cognizant. World Headquarters European Headquarters India Operations Headquarters 500 Frank W. Burr Blvd. 1 Kingdom Street #5/535, Old Mahabalipuram Road Teaneck, NJ 07666 USA Paddington Central Okkiyam Pettai, Thoraipakkam Phone: +1 201 801 0233 London W2 6BD Chennai, 600 096 India Fax: +1 201 801 0243 Phone: +44 (0) 20 7297 7600 Phone: +91 (0) 44 4209 6000 Toll Free: +1 888 937 3277 Fax: +44 (0) 20 7121 0102 Fax: +91 (0) 44 4209 6060 Email: inquiry@cognizant.com Email: infouk@cognizant.com Email: inquiryindia@cognizant.com©­­ Copyright 2012, Cognizant. All rights reserved. No part of this document may be reproduced, stored in a retrieval system, transmitted in any form or by anymeans, electronic, mechanical, photocopying, recording, or otherwise, without the express written permission from Cognizant. The information contained herein issubject to change without notice. All other trademarks mentioned herein are the property of their respective owners.