The Buy Side Doesn’t Want Impressions Counted Before They Hatch

chickensFor the buy side, the only fair way to measure mobile ad impressions is when they’ve actually rendered on a device.

Otherwise, advertisers might be getting taken for a ride by publishers using different methods of counting, said Richy Glassberg, CEO of Medialets, a mobile ad platform and mobile ad server that’s been under WPP since April 2015.

“If an advertiser pays a publisher to deliver 100 million ads, that publisher needs to make sure the advertiser gets 100 million ads rendered and not 100 million ad calls,” Glassberg said. “We need to hold publishers accountable and count on render.”

As the chief of a holding company subsidiary, it’s no surprise Glassberg feels that way. And, in fact, the numbers paint an unfortunate picture.

After looking at 2.7 billion impressions across its mobile ad server over 11 days in early May, Medialets found that roughly 20% of ad calls on the mobile web were “wasted,” aka they don’t ever fully render on a device.

In other words, one in five ads delivered to a device never actually shows up. That number drops to 10.9% for in-app ads, which is better, but not really all that great.

Speaking of not that great, certain creative types have a tendency not to fully render, with rich media as the top offender. Medialets found that 48% of rich media interstitials are delivered but never appear on the mobile web, while 30.2% of rich media expandable ads and 27.9% of rich banners are also no-shows.

That’s compared to 14.4% for static banners and 17% for static interstitials.

“If I were a client, I would look at these numbers and I wouldn’t even bother with rich media at all,” Glassberg said. “One of the issues holding back spending on mobile is that counts have not been correct. Publishers don’t want to lose inventory, but the truth is that this isn’t working for the advertiser because the ad isn’t getting to the device.”

Mobile is just trickier than desktop, said Rachel Pasqua, practice lead for mobile and emerging technology for North America at WPP agency MEC, which uses Medialets as its preferred mobile ad server.

“Network latency and vary data speeds make it much less certain that a mobile ad will reach its destination and load properly,” Pasqua said. “This wasn’t as big of an issue when the majority of mobile ads were simple banners, but now, as we’re raising the bar much higher for mobile creative experiences, it’s a serious concern – mobile by its very nature is more prone to error in the delivery of an ad unit.”

Counting on render helps allay that concern, said Pasqua, who noted MEC clients are seeing their percentage of mobile traffic steadily rise.

It’s that sort of progress that boosts client confidence in mobile, said David Stopforth, media director at Wieden+Kennedy.

“Delivery problems can also make clients nervous and reluctant to try new things creatively, which is tough at a time when we are all still testing and learning a lot in this space,” Stopforth said. “A more rigorous approach to measuring what is delivered successfully can only help offset our – and our clients’ – concerns. We still want to work toward a more accurate standard of viewability, so counting on the basis of completed loads/renders is just a sensible step.”

Because render isn’t the be-all and end-all – viewability is. The two are inextricably linked.

“Render is not a proxy for viewability, rather it’s a foundational layer of viewability,” said Joseph Galarneau, CEO and co-founder of Mezzobit, a startup that provides data leakage and viewability solutions for publishers. “It’s necessary, but not sufficient.”

Seeking to accommodate stringent viewability demands on the buy side, numerous publishers have bent over backward to redesign their websites, tweak their pages, try out new ad units and lazy-load content so that it doesn’t appear until a user scrolls to it.

All of that has admittedly led to better viewability scores overall, but it’s also reduced the amount of inventory that publishers have to sell and created a scenario in which publishers are optimizing for viewability – crappy ads count as long as they’re seen – rather than quality.

“Anecdotally, we see a lot of publishers get frustrated,” said Galarneau. “Viewability has become the onus of the publisher.”

mobilemediawasteFraud’s Eye View

Mobile viewability is still a work in progress. The final standard is expected in June. Considering the kinks that still exist around putting desktop viewability into practice, it’s going to be a long road.

Clearly, though, counting on render is where the puck is going. As the Media Rating Council noted in the mobile viewable ad impression measurement guidelines it put out for public comment at the end of March: “Each valid viewable impression originates from a valid rendered mobile served impression. In no case should viewable impressions exceed render mobile served impressions counted on a campaign.”

Basically, render is the egg and viewability is the hatched chicken.

Getting there means getting away from counting on ad call or ad request, Glassberg said.

“There is not a good technical standard for viewability in mobile – it’s still coming – but with viewability, it’s important to only deliver the pixels to someone who can see an ad and not to count if an ad never got there,” Glassberg said. “To do that, you need to count correctly at the beginning. If NASA sent a rover into space and it made it there, but then it crash landed, is that a success? It’s the same with an ad. Why would any advertiser pay if an ad didn’t land?”

It’s a question that also applies to another big agent of mobile media waste: fraud. Amit Joshi, director of product and data science at recently acquired anti-fraud firm Forensiq, estimates that about 20% of media is being scooped up by fraudsters.

“We’re much further along in the standardization of what counts as a valid impression than we are in the battle against fraud, and I believe what’s being lost to fraud is higher than what’s being lost to issues with viewability,” Joshi said. “Without eliminating fraud from buys and data sets, you’re optimizing on incorrect information, which makes the conclusions you’re drawing on the data false.”

And then there’s the added complication of bad actors actually gaming viewability to their own advantage. Forensiq has observed a growing number of instances in which sources of traffic are both riddled with bots and have high viewability numbers.

“If you take the total number of ads counted and viewed and remove the botnet traffic from that number, viewability in some cases might be as high as 90%, but the ‘human viewable metric’ is almost zero,” Joshi said.

The only logical conclusion, he said, is that “bots have figured out how to spoof viewability.”

Although Forensiq isn’t 100% sure yet how the fraud is being perpetrated, it seems like that an unauthorized party is loading an ad in an invisible browser and somehow counterfeiting the message that measurement vendors send back to the system to acknowledge that an impression was viewable.

Counting on render won’t protect the buy side from fraudsters intent on faking viewability to get a buck.

But that shouldn’t be something advertisers are responsible for fixing, said Glassberg, who noted that Medialets brought its mobile media waste numbers to the attention of a number of different exchanges in May.

Some, like AppNexus and The Trade Desk, responded with alacrity and took immediate action to address whatever problematic inventory sources were causing the low render numbers. A number of other exchanges Medialets contacted, though, didn’t seem to care, Glassberg said.

“When you get into the world of networks, DSPs, SSPs and exchanges, you get into unhealthy inventory sources, and it’s up to the inventory sources to handle fraud,” he said. “But a larger problem is that I don’t even know why there are so many exchanges, especially in mobile. There’s just too many people and not enough inventory.”


Popular On AdExchanger Right Now:

 

Add a comment

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>