1、1 Bot Baseline 2018-2019 May 2019 2018-2019 Bot Baseline Fraud in Digital Advertising 2 Bot Baseline 2018-2019 Table of Contents 2 Table of Contents 2 A Special Thank You to the Following ANA Member Company Participants 4 A Letter from the Authors 5 About the Study 6 The Size and Scope of the Study

2、7 Major Findings 8 Projected Fraud Losses in 2019 by Category 9 How the Industry Has Made These Gains Against Fraud 10 The Best Buyers Are Doing Better than Ever 10 Its Still Too Hard to Know What Youre Buying 11 The Good News 12 Fraud Volumes Are Growing in Newer Frontiers, Such as Mobile and OTT,

3、12 But Reductions in Desktop Fraud Prove This Fight Is Winnable Structural Improvements to the Fight Against Ad Fraud 13 More Money Is Now Being Spent Through Ad Tech Platforms with 15 Built-In Fraud Prevention Measures TrafficVendorsGoUnderground16 CybersecurityMeasuresLikeAds.txtHaveReducedDesktop

4、Spoofing18 The Formation of Industry Coalitions 19 The Bad News 20 Less than Half of All Impressions Are Fully, Transparently Validatable 21 Most Video Ads Still Dont Support VAST4, the Gold Standard in Transparency 22 The Quality of Real-Time Independent Monitoring Is Still Inconsistent 23 Outdated

5、 Technology 23 Technical Limitations 23 Walled Gardens 24 3 Bot Baseline 2018-2019 3 Human Error 25 Intentional Evasion 25 In Practice, High-Fidelity Validation Is Surprisingly Uneven 25 The Frontiers in Fraud 26 FraudontheMobileWeb:EasytoDo,HardtoProfit26 In-AppFraud:GrowingandInnovative26 Server-S

6、ide Ad Insertion (SSAI) 27 Connected TV (CTV) 27 Recommendations 28 Steps to Take to Reduce Fraud 28 Steps to Take to Increase Measurability 30 Ad Fraud Glossary: An Exhaustive List 31 About the Study Partners 38 About the ANA 38 About White Ops 38 4 Bot Baseline 2018-2019 A Special Thank You to the

7、 Following ANA Member Company Participants 4 5 Bot Baseline 2018-2019 A Letter from the Authors This is our fourth Bot Baseline report, and in some important ways, things are better than they haveeverbeen.Illegitimatetrafficsourcingisdeclining.Lesssophisticatedcybercriminalshave abandoned their frau

8、d schemes. Industry coalitions have formed to combat more sophisticated threats. Advertisers are spending more on channels with robust fraud protection measures. We are seeing our founding faith turn into validated fact: the battle against fraud is winnable. However, the battle is not won yet. Ad fr

9、aud persists in new, creative forms. Every year, cybercrime becomes more advanced, evolving in response to the selection pressure of fraud detection practices. We saw this clearly with 3ve, the fraud operations that implemented sophisticated tag evasionaswellasIPhijacking-wheregroupsofIPaddresseswer

10、etakenover-tobypasssecurity measures. Though transparency issues have always plagued digital advertising, they have become more pressingthanever.Inthisreport,werevealtheimportance-andpervasiveness-ofthe limitations in complete third-party auditability of ad impressions. Now is the time for advertise

11、rs to push for the ability to hold all ad impressions to the same high standard of validatability. Imagine if every CAPTCHA on the internet was the same. It wouldnt work very well. Validation with only a pixel is like serving a CAPTCHA that never changes. Only a dynamic challenge can be used to catc

12、h a dynamic adversary.1 Despitethesechallenges,weareoptimisticaboutthefuture.Ads.txt,andthereductioninspoofing it achieved, has proven the power of industry-wide cooperation. Federal indictments such as the ones handed down for the 3ve and Methbot operations demonstrate that cybercriminals can be he

13、ld accountable, and face real consequences, for their actions. Asanindustry,wenowhavethemomentum,ifwepersevere,tofixthefraudprobleminadeeply meaningful way. We believe that marketers can help bring about a future where ad fraud is not profitableenoughtobeworththerisk. 1 A CAPTCHA is a program or sys

14、tem intended to distinguish human entities from those that are automated; typically this is a method to prevent spam and automated extraction of data from websites. 5 Michael Tiffany Co-Founder and President, White Ops 6 Bot Baseline 2018-2019 About the Study For the fourth time, White Ops and the A

15、NA have partnered to measure bot fraud in the digital advertising ecosystem. Previous studies measured bot fraud in the digital advertising ecosystem in August/September 2014, August/September 2015, and November/December 2016. In the latest study, 50 ANA member companies participated. White Ops work

16、ed with brand advertisers and their agencies to analyze digital advertising activity data between August 1, 2018 and September 30, 2018. Measurements of fraud found in the global marketplace are derived from White Ops ANA study participant data. In this years Bot Baseline, we share: Baseline measure

17、ments of Sophisticated Invalid Traffic (SIVT),2 excluding SIVT that was thwarted or otherwise not paid for. Initial metrics of the measurability and auditability of all the digital media paid for by study participants, illustrating the uphill battle toward full transparency. Practices related to the

18、 detection and prevention of digital ad fraud. Recommendations on best practices that will help advertisers protect budgets by ensuring their advertising dollars are not stolen by fraudsters. 6 2GeneralInvalidTraffic(GIVT)includesknownnon-humanorfraudulentsourcesthatcanbeidentifiedwithindustrylistsl

19、iketheIABBotsandSpiders Listorparameter-baseddetectiontechniques.SophisticatedInvalidTraffic(SIVT)includesinvalidtrafficthatispurpose-builttoevadedetection. 7 Bot Baseline 2018-2019 The Size and Scope of the Study Of the 50 ANA member participants in the current study, 26 contributed data to previou

20、s Bot Baseline reports. Eleven have participated in all four studies, nine participated in three studies, and six participated in two studies; the remaining 24 participated for thefirsttime. Our study examines advertising by brand marketers. It does not include search or paid social media campaign d

21、ata. 7 Domains 606K Participants 50 Campaigns 2,400 Placements 130K Impressions 27B 8 Bot Baseline 2018-2019 Major Findings For the First Time, More Fraud Will Be Stopped This Year than Will Succeed Today, fraud attempts amount to 20 to 35 percent of all ad impressions throughout the year, but the f

22、raud that gets through and gets paid for now is now much smaller. We project losses to fraud to reach $5.8 billion globally3 in 2019. In our prior study, we projected losses of $6.5 billion for 2017. That 11 percent decline in two years is particularly impressive considering that digital ad spending

23、 increased by 25.4 percent between 2017 and 2019. A detailed breakdown by device and media type is on the next page. Forthefirsttime,themajorityoffraudattemptsaregettingstymiedbeforetheyare paidfor,byDSPsandSSPsfilteringfraudulentbidrequests,byclawbacks,orbyother preventative measures. Absent those

24、measures, losses to fraud would have grown to at least $14 billion annually. Fraud is an evolving threat. Fraud rates have been growing in new formats, and continued vigilance is needed. But there is no denying the structural gains theindustryhasmadeinthisfight. 8 3 Loss estimates are based strictly

25、 on digital spending in the categories included in the study: video, display, and other CPM formats for desktop and mobile devices. 9 Bot Baseline 2018-2019 Projected Fraud Losses in 2019 by Category 4 Desktop Study Sample Size: 13.6 Billion Impressions Mobile Study Sample Size: 13.5 Billion Impress

26、ions 4 Fraud rates exclude social media and search. 5 As in prior years, video still shows more fraud. 8% Display down from 9% in 2017 14% Video5 down from 22% in 2017 12% Other Rich Media, Takeovers, etc. 3% Display8% In-App Video 14% Web Video7% Other 9 10 Bot Baseline 2018-2019 How the Industry H

27、as Made Gains Against Fraud Anti-fraudmeasureshavemadetrafficsourcingmoredifficultandexpensive. ThankstotrafficsourcingtransparencyeffortsledbytheTrustworthyAccountability Group(TAG),trafficvendorshavegonefurtherunderground,reducing“retail”botbuying on the open web. Ads.txthasreduceddomainspoofing.6

28、 Additionally, TAG requires publishers to have completedads.txtfilesiftheywanttobeCertifiedAgainstFraud. Far more dollars are now being spent through programmatic platforms with built-in fraud prevention measures. Arrests, like those of the alleged masterminds behind the 3ve and Methbot operations,

29、have brought real consequences to botnets operating overseas. The Best Buyers Are Doing Better than Ever Inourfirststudyin2014,fraudaffectedeveryone,thosewithstraightforwardmedia plans and very sophisticated ones alike. This year, every buyer in the study knew about fraud risk; 90 percent had MRC- a

30、ccreditedfraudverificationmeasuresinplacetodealwiththisrisk. Performance was loosely correlated with study participation; long-time participants performed better than they did in previous years. For the top quintile of buyers, fraud was nearly nonexistent. However, discrepancies betweenadserverimpre

31、ssionsnumbersandverificationimpressionsnumberswere present for even the most sophisticated buyers. 6 Ads.txt, created by the IAB, stands for Authorized Digital Sellers. The mission of ads.txt is to increase transparency in the programmatic advertising ecosystem.Itisasimple,flexible,andsecuremethodth

32、atpublishersanddistributorscanusetopubliclydeclarethecompaniestheyauthorizetosell their digital inventory. 10 11 Bot Baseline 2018-2019 Its Still Too Hard to Know What Youre Buying The ability to hold all ad spending to the same high level of validatability should be one of marketers top concerns in

33、 2019. The time has come for marketers to stop tolerating and stop paying for outdated media formats like VAST 2 that cannot support the highest levels of third-party validation. By selling media under conditions that do not support high third-party validation, even trustworthy media companies are e

34、ssentially part of the problem, since the vast sea of good but low-transparency inventory provides cover for the fake inventory sold under the same formats. Fraud detection and prevention are only as good as their implementation. Low Invalid Traffic(IVT)7 measurements do not provide a full picture w

35、hen half of a media plan ishighlyvalidatablewithaccreditedthird-party,dynamicJavaScriptverification, but the rest is covered by a mix of half-measures like 1x1 pixels, list-based protections, and limited, static integrations. 11 7InvalidTrafficisoftenreferredtoasNon-HumanTraffic(NHT).ItisthetotalofG

36、eneralInvalidTraffic(GIVT)andSophisticatedInvalidTraffic(SIVT). Case Study: Trading Desk Video One CPG participant purchased 44 million video impressions from an agency trading desk. Of the 44 million impressions, only 48 percent were validatable at the highest standard, with a dynamic fraud test se

37、rved via JavaScript. 12 Bot Baseline 2018-2019 The Good News Fraud Volumes Are Growing in Newer Frontiers, Such as Mobile and OTT, But Reductions in Desktop Fraud Prove This Fight Is Winnable The rates for desktop ad fraud were the lowest in the history of the Bot Baseline study. Strideshavebeenmade

38、indesktopdisplay,wherebuyingundetectablebottrafficis becomingtooexpensivetobeprofitable.Fraudindesktopdisplaywas11percentin2014, 9 percent in 201617, and 8 percent in 201819. This is how we win: if the industry makes cybercrime unprofitable, cybercriminals will lose the incentive to invest in ad fra

39、ud. 12 Case Study: Nonprofit 5.8 million display impressions were purchased from a century-old nonprofit organization, with just 22 percent validatable at the highest level. 13 Bot Baseline 2018-2019 Structural Improvements to the Fight Against Ad Fraud Inthefiveyearssinceourfirststudy,adspendingint

40、hecategoriesmostvulnerableto fraud video and display advertising across desktop and mobile devices has more than doubled,8 but losses to ad fraud have not. Increased awareness of the problem among marketers raised the priority of dealing with it, and leaders throughout the industry have risentothech

41、allenge.First,thisslowedthegrowthofadfraud.Now,forthefirsttime, weareprojectinglossesthisyeartobesmallerthaninthefirstyearofourstudy. 8 PWC Global Entertainment they were all in-house, with limited coordinationbetweenplatforms,varyingwidelyinresourcesandstaff.Ultimately,they were grading their own h

42、omework. Today,incontrast,therearethird-partysecurityfirmstreatingfraud-fightingasa dedicated function. When security is separated from audience and media intelligence, collaboration across platforms becomes substantially easier to scale. These measures helpmarketersavoidbiddingoninvalidtrafficbydef

43、ault,withoutextraeffort. Breakout of Bot Baseline participants who employ fraud verification measures: Use fraud verification services Do not use fraud verification services 90% 10% 15 16 Bot Baseline 2018-2019 Traffic Vendors Go Underground Itusedtobeeasytobuyhigh-qualitybottraffic.Inpastyears,weev

44、endiscovered trafficvendorssellingtrafficbyverification“filter,”tunedtolookgoodtodifferent measurementvendors.Trafficaggregatorskepttrackofthebotvarietiesflaggedas suspiciousorlowqualitybycertainadvertisingmeasurementorverificationvendors, ensuringtheycouldonlyselltrafficthatwouldgoundetectedbybuyer

45、s. Notorious threat actor group KovCoreG exploited this to monetize the Kovter malware, infecting millions of machines in the aggregate while making the majority of their earningsoffthefreshestmillioninfectionsatanygiventime.Trafficfromfreshly infectedcomputersfetchedthehighestprice.Olderinfectionsw

46、eresoldfor“tonnage” onlow-CPMadcampaignswithlittleornoadverification.Thispattern,adoptedbymost of the top-tier fraud operators, created the appearance of fraud protection without tippingovertheapplecartofwidespreadtrafficsourcing. But this year, we are pleased to report the beginning of a turnaround

47、. It is still possible tobuysophisticated,realisticbottraffic,butithasbecomefarmoreexpensiveand harder to source. 16 The extraordinary effort led by TAG toshinealightontrafficsourcing has severely reduced demand for low-qualitytrafficvendorsthat arecaughtsellingbottraffic. As with 3ve, major botnets

48、 have been dismantled. When a botnet getsdismantled,itstraffic,nomatter the quality, cannot be sold anymore. Two forces have upended this model: 17 Bot Baseline 2018-2019 Retailvendingofbottrafficontheopenwebhassubstantiallydiminished. Realisticbottrafficusedtobeaccessibletoanyonewithacreditcard.Now

49、,more of the sophisticated bot-buying has been forced underground, to invitation-only forums or transactions made over chat apps. That requires a more sophisticated buyer.Youcanstillbuysophisticated,nearlyundetectablebottraffic,butitisharder, which diminishes its scale. OverthefourBotBaselinestudies,thevolumeofpaidtrafficacquisitionhas increasedbyasmallpercentageoftotaladtraffic;howe