This story originally appeared in the Nov. 11 issue of Marketing
A Marketing investigation shows many ads don’t end up where brands expect
Imagine you represent Boston Pizza, State Farm or Procter & Gamble, and while surfing the web you click a bad link and end up at BeachCreeps.com, an adult blog that posts photos of bikini-clad women. There at the top of the site is a shiny new banner ad from your latest media buy. Now, imagine you’re not imagining it. Beach Creeps is a real site, where Marketing discovered ads for those and many other household name brands like Target, Harley Davidson, Scene Card and Equifax. On CokeAndPopcorn.com, a site that collects links to illegal streams of popular television shows, we found advertising from the likes of ING Direct, Best Western and many others. And it wasn’t fleeting or random—the ads showed up multiple times in a variety of sizes and placements.
Brand safety in online advertising is a hot topic in the burgeoning ad tech world, where automated buying through programmatic and real-time bidding (RTB) strategies is gaining traction and is expected to grab half of the digital display market within a few years.
Despite dramatic growth globally, adoption of programmatic buying in Canada has been impeded by concerns over brand safety and a lack of so-called premium inventory for ad placement. The latter issue is being addressed by the market—premium inventory volumes have grown as leading publishers like TC Media, Torstar and Rogers set up or join private ad exchanges—but marketers can’t be faulted for breaking into a cold sweat over fears their ads have ended up next to raunchy bikini shots.
From a PR perspective, having your ad land on these sites is hardly scandalous because it’s so widespread. In fact, it’s clear from our investigation that many reputable marketers have a lot of good company in a lot of bad places. In partnership with Torstar-owned digital ad platform Olive Media, Marketing looked for big-name brands showing up on inappropriate content and “ghost sites” that sell millions of monthly impressions but are rarely, if ever, seen by consumers. It’s important to note that one of the challenges in defining the scope of the problem stems from the difficulty in differentiating between sites with unsavoury or just bad content and those created with outright fraud in mind. What ties them together is that marketers simply don’t want their ads appearing on any of them.
In addition to finding ads on Beach Creeps and Coke and Popcorn, we discovered ads for a corporate Canada who’s who from IKEA to TD, RBC, Tylenol, Reitmans, Chevrolet, Koodo and Sport Chek on sites that show erratic, high-volume traffic patterns characteristic of what experts identify as “non-human” (a.k.a. bot) visitors. Our investigation explored ad traffic on six such properties, currently blacklisted by Olive Media : SmartMomDeals.com, WomensHealth-Base.com, All-Allergies.com, QuickTechTips.net, VeggieMixer.com and Celebrity-Gossip.net.
We approached many of the brands whose ads turned up on the unsafe sites, and most declined to comment for this story. Those who did said they use whitelists (okay sites to deal with) and blacklists (avoid) to screen digital media buys for the right inventory. Evidently, some sites have managed to slip through these lists—even Beach Creeps, which has been previously red-flagged by Google and appears on exchanges under its real URL. It’s clearly an issue that needs more attention, says Andrew Zimakis, vice-president of marketing at ING Direct, whose ads turned up on several undesirable sites in our research. Though he is of the opinion the problem is not widespread for his company, he says protections for digital advertisers are not as rigorous as they should be and could be improved through more systematic controls. The spotlight is about to get brighter, and brand safety “will become a bigger issue as programmatic trading grows.”
André Leblanc, director of marketing communications for Yellow Pages Group, says his fi rm had no idea its ads had been served on WomensHealthBase.com until we shared that information. YPG looked deeper and says the ad was served as part of a programmatic campaign which was handled externally and that the site in question is actually ranked by comScore, lending it legitimacy.
Leblanc says YPG sees the added scrutiny required as part of “a constant tug-of-war that occurs between balancing cost and performance against environment and reporting transparency” and observes that “network buys and programmatic have enabled advertising costs to go down, but at times do present a challenge to ensure 100% quality.”
AD TECH’S CHALLENGE: REMOVE THE FEARS, UNLOCK THE BUDGETS
Rapidly rising demand for programmatic and RTB inventory means boom times for the advertising technology sector. Leading Canadian players like Casale Media, Olive Media, ExchangeLab and Chango are expanding their reach here and in the U.S. and British markets, while U.S. companies like Tremor Video and Millennial Media recently launched IPOs to take advantage of financial market interest in the sector’s rapid growth.
One of the giants in this space is New York-based AppNexus, a digital advertising technology company that provides solutions for buyers, sellers and everyone in between, not to mention one of the world’s biggest open exchanges. Michael Rubinstein, the firm’s Canadian president, says more focus on shortcomings in the current programmatic buying environment is needed to help speed reforms.
“The more stringent we all become as an industry and the more comfortable everyone can be that the system is working, then the better it is overall,” says Rubinstein.
In digital advertising more than any other medium, trust is essential, he says. Without that trust, it’s hard for buyers to do more than get their feet wet in RTB. “I still think there are people who hold back budget because they just don’t know,” he says. “The examples of ads running on those kinds of sites are probably the exceptions and not the rule. Different companies have gone to different lengths to address or ignore the problem.”
The ghost sites have a few obvious giveaways, such as high ad click-through rates coupled with low social engagement and low conversion rates (bots can click, but rarely sign up for newsletters with valid emails). In some cases it can be as easy as looking at the site’s publicly available profile on a service like Alexa to see how long the page has been registered, and where traffic is coming from. A new site that is growing its user base through audience discovery will usually receive most of its traffic from search or social media, whereas a ghost site will receive direct traffic or from unrecognizable referrers.
A lot of dodgy sites and any other irregularities are only caught by publishers after the fact, when they study post-broadcast reports. YPG’s Leblanc, for example, says a review determined that the site where Marketing found one of their ads, WomensHealthBase.com, “seems to be a ghost site.” These post-campaign evaluations allow a market participant to modify their black or white sites and, he says, “usually allow for the application of corrective measures,” including make-goods.
A site like Beach Creeps can evade some controls based on the fact it is listed on open exchanges under its real URL. Even so, it has been identified as a brand safety risk by major exchanges like Google’s Doubleclick. Ray Philipose, vice-president of performance media at Olive, says the fact the site passes muster for so many brands may be an example of the weakness of using algorithms to assess brand safety: “Beach Creeps passes a bunch of superficial tests; there’s no nudity, there’s no obscenity, there’s no torrent activity.” Showing women in swimwear is, in itself, not a disqualifier, he explains, using Sports Illustrated’s swimsuit issue as an example of premium ad-friendly content. “But I think anyone who’s doing their own subjective evaluation would conclude very quickly that Beach Creeps and Sports Illustrated are at different levels of brand safety.” Nathan Mekuz, founder and chief technology officer of demand-side platform AcuityAds, sums up the good in the digital ad revolution as well as its challenges. “The beauty of RTB is that it opens up to a lot of long-tail sites.
Anybody with a little blog can now join an exchange and sell inventory. It’s fantastic, you get great value, and sometimes very targeted audiences in these little sites—but also, unfortunately, the side effect of that is that it also makes it easier for bad players to join in,” he says. “Naturally advertisers are reluctant to move to RTB because the perception is that fraud is so prevalent, and partly it’s true.”
Mekuz says that problem isn’t specific to RTB—it’s a problem for search, mobile and direct sales as well. But exchanges are especially good targets for fraudsters because with so much volume, they’re hard to police.
In-banner video exchanges have become a notable problem recently. Integral Ad Science found that in the first half of 2013, 40% of video in-banner ads bought by clients could be classed as fraudulent. (Many of the ads that appeared on the sites Marketing investigated were in-banner videos.) While programmatic video buying is exploding in popularity, the technology to properly track video ads is clearly still nascent.
Premium networks and direct sales tend to operate with a lower (though not nonexistent) risk of fraud and brand safety issues than open exchanges like AppNexus or Google’s Doubleclick. Closed-door marketplaces like the Olive Private Exchange and the Canadian Premium Audience Exchange (CPAX) provide added safety by positioning themselves as “premium” environments where inventory comes only from well-known, reputable publishers. Some open exchanges, like Toronto-based Casale Media, have earned a reputation for safety by taking a similar white-glove approach, built on partnerships with known and trusted publishers. All new sellers on Casale’s exchange must undergo a vigorous human review, and the exchange has a strict one-strike system for anyone caught hawking bad inventory.
But while premium marketplaces provide more security—and someone to take responsibility if anything untoward occurs—they’re limited by the number of relationships they can maintain and the man hours it takes to vet inventory. Because of these restrictions, premium can’t always offer the scale, ultralow cost and potential for optimization that the big open marketplaces like AppNexus and spotExchange provide. For many buyers, increased reach and targeting capabilities are worth the higher risk of reduced security.
Some industry insiders believe that’s been a big reason why many global exchanges haven’t taken as strict a stance on questionable inventory as Casale or Google—because most buyers are looking for a large volume of ads at low prices, and screening for fraud is expensive. Moreover, most independent exchanges (i.e. not owned by Google, AOL or Yahoo) are funded by venture capital, and could potentially feel a lot of pressure to rapidly grow low-cost market share at the expense of quality.
Rubinstein says AppNexus practices a policy of zero tolerance on dodgy inventory. He says the firm has a team of dozens of people combing every site available to his network each day and manually auditing creative to make sure it’s associated with a real advertiser with no deceptions.
LISTENING FOR THAT SUCKING SOUND
Bosko Milekic, co-founder and CTO of AdGear, a Montreal-based demand-side platform that assists advertisers and agencies in buying digital inventory, first caught wind of suspicious activity on the exchanges six months ago, when reviewing reports of where clients’ programmatically purchased ads had landed. The reports showed that a number of bizarre URLs were sucking up ads—sometimes 30% to 40% of an entire campaign buy.
Upon visiting the sites, Milekic says it was obvious they were fraudulent. Despite generating a high volume of traffic, they were relatively small sites with low-quality content, hosted on free blogging networks. “You could very clearly see this was something automatically generated.”
AdGear found a large number of ad tags that were redirecting through one another—driving dozens of impressions from a single ad placement—and in some cases redirecting to the client’s own site, to make it appear as though users had clicked through the ads. The scam would be immediately obvious to any human who checked the site, but lots of buyers use algorithms for optimization—and to an optimizer combing the marketplace for high-conversion traffic, these sites looked like where the action was.
These kinds of finds convinced AdGear to start taking fraud more seriously. “Since then, we’ve developed a three-pronged approach,” says Milekic. First, AdGear uses past traffic data to maintain whitelists and blacklists of sellers and sites, and to train algorithms to sniff out new instances of fraud as they crop up on the exchange. Second is a proprietary “trust score” that AdGear maintains on all publishers, sellers and exchanges it deals with, to help buyers better target their bids based on how sensitive they are to price versus reliability. Third, AdGear factors those scores directly into algorithmic optimization for its clients so that when its optimizers are out hunting for high-conversion traffic, they’ll take into account how trustworthy the seller is before jumping on a tempting trove of clicks.
Non-human traffic usually leaves behind data signatures that fraud detectors can look for, such as spikes in site visitors at odd times of day, low viewability rates, oddly regular mouse movements and clicks, or a high proportion of visitors coming from a single browser type. Fraud detection services like Integral Ad Science, White Ops and Adometry estimate that anywhere from 10% to 40% of global display inventory is fraudulent.
Solve Media ventured an estimate earlier this year that 46% of international ad tra c was suspicious and 29% came from non-human bots, contributing to a projected annual cost of $9.5 billion for the global ad industry. There are no comparable estimates specific to Canada, but experts interviewed for this story said that actual fraud is as prevalent here as in other markets, since the organized criminal organizations that perpetrate fraud—many of them based overseas—don’t pay much attention to national borders.
Luckily, inventory on unsafe sites like Beach Creeps is much rarer than outright fraud. Integral Ad Science estimates that only around 6% to 6.5% of display inventory purchased in the first half of 2013 ended up in moderate- to high-risk inventory featuring adult content, illegal downloads, drugs and alcohol, etc. Vendors like comScore and Proximic offer brand safety protection that detects inappropriate placements by monitoring site content, tags and metadata for suspicious keywords or content a particular buyer doesn’t want to be associated with (like airlines and plane crashes, or diet plans and eating disorders). The same mechanism can be used to block “high clutter” pages that have a large number of ad placements.
One of the things that makes fraud so hard to deal with is how fast it evolves. Criminal organizations are pouring vast sums of money into fraud and constantly imagining new and exotic schemes to get past fraud detectors. Demand-side services and buyers have to be vigilant if they want to keep up. “As soon as you identify and blacklist the fraudsters, they will rapidly set up a new domain or a new IP address,” says David Hahn, senior vice-president of product management at Integral Ad Science, a media quality evaluation service that’s been developing anti-fraud solutions for three years. To get a sense of how fast fraudulent domains are being registered, Canadian DSP EyeDemand has blacklisted 1,341 domains in the last 15 months, or 22 per week.
Hahn explains that Integral no longer uses global blacklists because that’s a reactive approach. By the time the fraudsters are identified and blocked, the impressions have been served and the damage is done (and it’s only a short amount of time until they’ve got the site back up and running at a new URL). Instead, Integral attempts to detect fraud in real-time, on an impression-by-impression basis, to stop fraudsters in the act. Integral claims its predictive algorithms are able to detect fraud in “a thousandth of a second prior to the delivery of an ad,” and automatically stop it from being bid upon or served. It’s an impressive promise.
Hahn takes the unusual position that buyers have more power to resolve fraud than do sellers, since it’s up to buyers to remove the incentives that make fraud worthwhile. In particular, he’s concerned that current attribution models and performance metrics rely too heavily on clicks and views, which fraudsters are extremely good at generating. Hahn says buyers who focus on clicks are enabling fraudsters, pouring money into the already massive fraud industry.
Other experts agreed that cost-per-view and cost-per-click measurement should be avoided to minimize fraud. In the words of Casale Media founder and vice-president of strategy Andrew Casale: “If that’s the way you buy, you’re the bad actor’s best friend.”
IN SEARCH OF TRUST
As the industry matures, some of the larger exchanges are getting more serious about the problem. Eighteen months ago, OpenX, a global open exchange where 250 billion impressions are sold every month, put together a team of data scientists and statisticians and tasked them with developing an in-house fraud prevention solution for the exchange. In August, OpenX rolled out a proprietary detection and trust rating system which, according to the company, is expected to boot 30% of sellers and 10% of inventory o of its exchange—not a choice any marketplace can make lightly.
Qasim Saifee, senior vice-president of monetization platform for OpenX, said that the investment and technical work required to implement a publisher-by-publisher and site-by-site screening process, as well as ongoing exchange-wide detection, is a major barrier to most exchanges—but it’s also an enterprise investment. In the long run, fraud hurts exchanges and legitimate publishers as much as it hurts buyers, by slowing programmatic demand and dragging down CPMs across the board.
Many of those interviewed for this story weren’t sure whether the onus was on the buyer to watch out for fraud, the seller to clean up the inventory, or the industry to adopt cooperative standards to force out bad actors. But when asked about responsibility, Saifee didn’t hesitate to say OpenX and the other exchanges at the centre of the action need to step up.
“You can shift blame around, but ultimately, as the marketplace—as the conduit, as the thing that connects buyer and seller—we think that responsibility sits with us.”
There’s more! Pick up the Nov. 11 issue of Marketing for more on how the largest online publishers fare with bad inventory and a look at recent major ad scams. It’s available for the iPad Newsstand, or by delivery to our subscribers.