As part of research I’m doing both for the book and for my upcoming conference (the CM Summit, more on that soon), I’ve been in pretty extensive conversations lately with dozens of key players in the advertising technology industry. I find the ecosystem that has developed to be fascinating, complex, and ripe with opportunity (and deeply important to the future of our society, not just marketing). I’ll be writing about it quite a bit in coming months. But before I do, I wanted to call out a growing issue that our industry will have to tackle sooner rather than later.
Just as in the early, wild west days of search (1999-2004), the programmatic advertising business – a multi-billion dollar marketplace growing faster than search, video, or anything else for that matter – is riddled with fraud.
That’s what many very reputable sources have told me in great length over the past few months. It’s something of an open secret, and more and more people are speaking out against it. Here’s Federated Media’s Walter Knapp on the problem, back in March of last year:
The great thing about the Internet is that it is built on the foundation of openness — from the way the domain system works to the way content and publishing are increasingly democratic. The core technologies embrace openness, sharing, linking and the ability to consume content across devices and across wired or wireless connections. Unfortunately, the openness we depend on in the digital media business is also available to people who can (and will) take advantage of this openness and exploit it for their own selfish wants.
Knapp notes two forms of fraud – ad injectors, fraudulent browser plugins that take over ad calls; and the practice of inserting an entire site into a 1×1 pixel hidden on high traffic but low quality sites featuring porn or music lyrics. Both are examples I’ve heard about over and over in my reporting. A third involves “stacking” ads one behind the other, all playing video to completion, often playing in inactive tabs. A fourth features refreshing ad calls on accelerated schedules or in inactive tabs. Yet another involves running as many ads as possible out of view, simply to gain “view through attribution” on a closed loop success metric.
More people are starting to call these practices out. AppNexus CEO Brian O’Kelly prominently featured the issue of fraud in his blog post celebrating his company’s recent $75 million funding, and what he intends to use it for:
Quality We will continue to invest in cleaning up the advertising marketplace. We’re proud of our anti-piracy stance, and our 5x volume growth this year indicates that you don’t need to serve on BitTorrent sites to be an ad platform company. We are investing heavily in fighting fraud, porn, malvertising, and malicious toolbars, and we are actively working on viewability tools.
Programmatic industry watcher AdExchanger puts it this way:
AppNexus’s pledge to invest money in ad quality issues is worth calling out. The issue is becoming more pervasive as companies emerge to exploit the vulnerabilities of real-time traded inventory to data and impression fraud, malvertising, and other nefarious practices. Fraudulent activities aside, the emergence of robust ad verification and viewability tools means display ad marketplaces and buying platforms must keep a clean nose.
It’s true that many folks are working on addressing the issue, including the IAB. But the bad actors are currently far ahead of the good guys, and worse, many in our industry are turning a blind eye, hoping the problem goes away in time, without too much publicity. Why? Well, nearly everyone gets paid from fraud – the publishers, the exchanges, the data providers, and the agencies. Even the marketers,who are footing the bill, feel like they are getting value – because the success metrics they’ve set up are being met.
But fraud hurts the ecosystem in a massive way. It means that low quality, invisible, or purely fraudulent inventory is holding down the average value of the entire marketplace – hurting high quality, engaged publishers in the process, stunting investment in quality content.
Over and over, I hear that the reason CPMs (the amount of money a marketer is willing to pay for one thousand advertising impresssions) are so low is because “there’s infinite inventory.”
Hogwash. There’s only so much time in the day, and only so many pages where actual human beings are really paying attention, and the web (including mobile) is growing at a finite pace. There are even fewer places where marketers can be assured of quality, engagement, and appropriate context. It’s time we focus on identifying them, and ridding ourselves of the true source of “infinite inventory” – fraud.
There’s also a huge amount of traffic generated by bots but not scrubbed by the IAB Whitelist. Some of this is because bots identifying themselves as such will not get a “true” view of the page they are looking for and some is from real fraud perpetrated unwittingly by infected computers. Bots make up a very large % of impressions, some estimate 30% or more.
Thanks, where are you sourcing that number?
http://news.solvemedia.com/post/32457007008/solve-media-uncovers-1-5-billion-in-wasted-ad-spend
Thanks!
Are there any open source audit tools that can be used to detect fraud?
Hi Harry, we have a freemium version of our ad verification system, MetricScience.com (250k impressions per month).
Smart marketers don’t assign much value to view-thru attribution – that’s agency bullshit. Performance display based marketing to CPA, CPL and ROAS — the fastest (only) growing segment of display for 6 years running — is pretty immune to these “stuffing” issues. If it was something of major concern to these highly analytic and metrics based marketers who look at every dollar spent, I doubt they would keep pouring their money here.
You’d think. But “view to completion” and view through, as well as click fraud on CPC stuff, is real, I’ve heard.
No doubt it’s real. i’ve seen the bots Ari speaks about – keep in mind many time publishers themselves are guilty of these as they buy traffic – and I’ve seen click fraud since 2005 and levels of cookie stuffing by agencies you would not believe. However, in performance marketplaces, my opinion has long been that this “comes out in the wash” so to speak. If marketers are not hitting their “cost-per” numbers good agents on the supply side (Google has done an amazing job w click fraud) and marketers themselves through the line items they are buying mitigate the overall effect of bad agents. I’ve not heard one performance marketer in my life ever say “I’m not buying digital because of issues related to fraud.” All that said – for digital “brand” buys that operate under a different set of “impression” rules, these issues are more important.
And if we are to solve the brand issue – bringing those budgets to digital in a meaningful way – we have to fix this.
Real DR marketers don’t care about fraud as they tie media costs directly to sales or customer acquisition. Brand marketers are still trying to justify their buys with click or ‘engagement’ data. Which are easily gamed.
They are top of the funnel marketers trying to use end of funnel
metrics. They forget that their job is to raise awareness and get
impressions for their brand, not to move product. Can’t tell you how many brand buys get cancelled or not renewed due to poor click of engagement “performance”. What publishers don’t do is show brand lift, nor due digital brand marketers.
Agree 99% with you. View through attribution based on a control sample hold out is real (ie understanding the true impact of the media stimulus). In any case, if you have real, outcome based metrics it helps root this out in the end. If you are measuring CTR and “reach” as success metrics – good luck to you.
I work on the measurement/validation side of things (comScore), and this is definitely something our end of the industry has been keeping an eye on as well. Specifically, we have a product called Validated Campaign Essentials that monitors campaigns for fraud (among other things) and reports on only impressions that were displayed in a viewable portion of the browser window.
http://www.comscore.com/Products/Advertising_Analytics/validated_Campaign_Essentials
Curious if you think that product catches all fraud, if it bores through frames, etc…
Agree that fraud is a problem as it is with many industries. Disagree with the finite inventory assumption. Yes, there is a finite amount of “quality” inventory. But is quality really the bellwether for advertisers, rather than audience or demographic (ie programmatic buying)? If so, then why has your own company (FM) moved away from working with a small number of quality sites to working with 200,000+ sites of any quality and selling those programatically?
Not “of any quality.” I’ll let Walter talk about how we choose sites at FM.
wait, you don’t know?
Good point about quality. Quality is a subjective variable even with the same exact content. A “click is a click” no matter if it’s from a high quality page or “low” which is why it’s more effective and cheaper, to go for high volumes of lower quality content than small numbers of high quality content.
We run a small digital firm that serves 100 Mil to 250 Mil impressions per month and believe fraud is rampant. We now only use “white listed” sites that tie out to our tagging, and redundant analytic platforms. We believe that a lot of the fraud could be cleaned up on the DSP level, BUT ultimately why would they want to…? They get paid a piece on all impressions, as do the big ad firms. You should dig into how the Big Ad firms hide fees in impressions through the DSPs as well. We have had battles with quite a few ad networks and would be happy to share our findings and give specific instances. But again, it is the wild west, and the DSPs allow inventory from almost any network just so the “brain” of their platforms can help them turn a buck-
Why bother to ‘clean up the advertising marketplace” when we’re poised on the brink of inverting the model to user-centered ‘pull’ on a context-aware, AI-driven basis?
The current model will collapse under its own weight as a blunt instrument. Personal data ownership and the Personal Cloud can be self-managed to benefit each user, yet also enable marketing that’s actually welcome in the moment.
Vendor and buyer can share the former middleman’s cut. And you get improvements in user privacy. Plenty of work going on in this area right now, including my company, Kimera Systems.
I do believe in this model long term, but it is a shift that feels years away.
I think there can be a cascading shift in methodology among alert ad networks when a major non-ad company adopts. The AI that drives realtime salience analysis for the individual user doesn’t need to amass Big Data and run Watson, but instead prunes the Context Graph in several different kinds of ‘local’ and ‘personal’ dimensions, constantly.
In the meantime, this path is a rational one for underperforming companies like Y! and MS to try to catch up.
The hurdle is building out critical mass so enough individuals who don’t know each other’s intent have made their selected sets of personal datastreams available for the AIs to be of mutual benefit…to humans and vendors alike. Hence Y! et al.
Jargon Bingo!
Whoa!
Well, obviously.
Along with the usual forms of fraud – ad injectors, refreshes, etc., we see a lot of low quality pages that are designed to game view-ability and are from a page quality perspective, so horrible that they clearly were not designed for human consumption. Bad content environments should also be treated as fraud since they seek to game ad systems into treating those ad calls as legitimate. At some point, the industry has to realize that an audience only focus which deemphasizes the ad environment supports much of the waste that exists within current online ad spend. Inventory has to be cleaned before it enters the marketplace and that means the supply side and exchanges have to take ownership and set certain guidelines in their site/content acceptance policies and block or eliminate publishers and networks that don’t comply. Some of working on it like AppNexus. The tools exist and the processes exist to do this in real time, all we need now is the collective will to cut the crap out of these systems. Speaking for Proximic, consider the Challenge Accepted.
Part of the problem (which has allowed fraud and game-playing to flourish) is the disconnect between media buying and marketing. In “simpler times”, the two were more closely aligned (generally speaking, reaching a lot people equated to product moving or sentiment improving). But, the sophistication of the ecosystem as well as the consumer has resulted in a gross disconnect. Reach is no longer synonymous with actually “reaching” or connecting with a consumer. It translates to your ad was logged in a system. That includes digital and the golden child of brand marketing…. television (won’t even get started on the grossly inaccurate panel measurements of DVR statistics or skipped commercials). However, reach is the currency for buying media to EXECUTE a marketing objective…whether that is GRPs or impressions or completions/CPVC is irrelevant as they are all mechanisms for counting volumes. With a focus on buying reach efficiently, the conversation around measuring marketing effectively has been reduced to simplified “metrics” such as clicks, completes, view-thru, audience ratings or the latest which is viewability (the digital agency is always looking for that ONE golden number…and then, we open the flood gates for fraud once that “performance” goal is identified). The problem is that marketing is partially an art (even direct marketing to a degree), and we need to be looking at the holistic picture which is not the way most agencies are structured or incentivized. A completion rate of 80% on site A vs. a completion rate of 80% on site B can have many different marketing stories as can viewability and view-thru, etc.. In an age of data, we need to be better at incorporating ALL the facts, looking for outliers and applying an art to the numbers. We need to stop evaluating performance on what we purchased media against (ratings, completions, view-thru, viewability). And, start evaluating on the holistic picture of what actually happened. The technology exists to do this TODAY, but both sides of the value chain shy away from the reality of what happens with each impression or rating that is purchased/served/”measured”. The benefit to the enormous amount of Fraud is that it is causing marketers to step in, and the ecosystem to step back and re-evaluate not just digital, but the entire industry. And, some innovative brands and agencies are making interesting advancements. There are many tools developing that detect fraud, but there is always “another game”, the best tools are those that look across the funnel of whats happening that trigger abnormalities and positive insights on scenarios where marketing truly worked.
Very thoughtful! What’s your background?
Thanks John. Video Ad Tech and a few years on the agency side (digital)
A fascinating and enlightening read – thank you.
Great post John. I admit that my opinion is biased, as I manage the Click Forensics program for Adometry, but its the only product designed to handle most of these issues. Click Forensics assigns ‘scores’ to each click that occurs within Ad Networks (>8 Billion clicks in 2012). Beyond simply identifying ‘fast clicks’, Click Forensics detects complex anomalies and collusion across sites, networks, engines and sources. In addition, the Click Forensics product provides proactive block lists including a list of IP’s that have been proven to commit fraud across our community, a Bots and Spiders list (including, and beyond the IAB list), colocation lists and ad provider lists.
Isn’t it about time we stopped using the terms ‘marketing, advertising and sales?’ In the social economy where the consumer not the company defines the message, the relevance and effectiveness of advertising is diminished to such an extent that it has become a naive (and expensive) exercise in futility
Well, hmm. I’d not go that far. It’s commercial speech – and done right I think we can make it valuable to all parties.
Good metrics can help sniff out fraud, but it remains an issue. View through attribution where you are not using a control sample is total B.S. (please call me if your agency has sold you this bag of hammers!). Measuring through to specific actions can help, particularly if these require one to put down a CC. But, I agree that spending time and effort rooting this out is hurting the efficiency of the system overall and can scare away people based on bad experiences.
The old adage that only half your advertising was reaching the right people but you didn’t know which half seems to continue into the new age — but with a different percentage. The problem of click fraud, for example, and the practices you mention, are a problem but it’s not clear that it would dramatically alter CPMs as you suggest, but certainly it’s a factor. And I’m not sure that I would trust the ad networks to put a whole lot of energy into eradicating the problem since they make there money either way.
Out of interest, do you know anyone, anyone at all, who has deliberately clicked on an advert?
paying for CPC and CPM doesn’t seem viable due to high bid prices. Social media marketing could be the next step.
om
http://www.concentricsys.com
Thank you John 🙂
http://www.argplywood.com