Last year, Facebook tested software that would represent a bold expansion of its display ads business beyond its own inventory, a move potentially worth billions in revenue. Using a demand-side platform, or a DSP, marketers would be able to use Facebook users’ identity data to bid on ad slots across the mobile and desktop Internet in real time.
But Facebook recently yanked the bidding software from service because the tests showed that banner ads that were served attracted too many fraudulent impressions by bots trolling the Web, the company confirmed to The Information on Friday.
Facebook is defying expectations by not building out fully capable ad-bidding software for inventory across the web. Instead, the social network will focus on filling the kind of ad slots it’s most familiar with–native and video. It’s a bet that could hamstring its ad-tech efforts to compete with Google now, but it could pay off in the long run.
While Facebook’s current advertising business centers mostly on its own mobile inventory and apps plugged into its ad network, many in the industry have been awaiting its plunge into the mobile web’s programmatic ad marketplace. Its ad server, Atlas, which on Monday added capabilities to serve video ads and track offline purchases, can measure whether users saw ads across digital devices. But because of the pulled DSP tests, the ad server doesn’t yet have a bidding platform that would expand its pool of marketing clients wanting to tap this programmatic marketplace.
To some in the industry, it seems like Facebook bit off more than it could chew in trying to jump into the ad-serving business to take on Google’s DoubleClick, which dominates the market. DoubleClick Bid Manager, the company’s DSP, helped pull Google's ad revenue outside its own site past $15 billion last year.
Abandoning those DSP tests–at least for now–illustrates how Facebook can be “ruthless about chopping off products when they’re not working,” said one person who has worked closely with Facebook’s ad-tech team.
Brian Boland, Facebook’s vice president of advertising technology, described the decision to stop DSP tests as a way to keep control on the “value” of buying ads through Facebook. He said the ad-tech team “saw a ton of waste, a ton of fraud” from the banner ads it served. Banner ads, which make up the bulk of inventory on DSPs, even on mobile, “don’t drive the kind of value we see on other ad formats like native and video,” he added.
Facebook will still build a bidding platform that would be similar to a DSP, but it will likely have narrower capabilities than DoubleClick. After scrapping the DSP tests on banner ads, Facebook “shifted toward higher-value areas,” like native and video ads, Mr. Boland said. He declined to elaborate on what exactly its next attempt to build an ad-buying platform would look like. Two people with knowledge of the company’s plans said Facebook is building ad-buying software that enables bidding on inventory that exists on it own sites and its ad network, Audience Network.
Native ads now make up only a sliver of the mobile ad inventory across the web. Focusing on native and video is likely a bet on the future growth of those ad units on mobile as a way of catapulting Atlas ahead of DoubleClick, which was built out predominantly as a desktop-focused business. Just like Facebook has tied its fate to mobile as a publisher that supplies ads, it looks to be doing the same as it feeds demand for ads.
Michael Collins, CEO of the ad-buying platform Adelphic, said Facebook won’t be able to fulfill all the needs of advertisers by just offering native and video ad-buying. But he said Facebook seemed to be sticking with more lucrative ad units, which it knows best.
"Your average brand will need access to more than just Facebook. Will it be of values to brand? Yes, but it will be a piece of the solution," he said.
While marketers like Facebook’s powerful identity data on users—evidenced by the social network’s 81% year-over-year surge in ad revenue last quarter—its system to serve ads across the web is closed and fragmented. Without a bidding platform, Facebook’s ad server Atlas doesn’t have all the functions a marketer would want.
Ad servers Atlas and LiveRail, which Facebook bought for a total of about $600 million in 2013 and 2014, are not considered successful yet. In January, Facebook shut down the video ad serving part of its LiveRail unit because of similar ad fraud it saw in its DSP tests. In the meantime, publishers using LiveRail can use other Facebook ad-tech products to manage their inventory. “We are taking a principled approach focused on value and yield for publisher,” Mr. Boland said.
Publishers and Marketers
The ad-tech team “saw a ton of waste, a ton of fraud” from the banner ads it served.
Atlas has struggled to win over marketers from Google’s DoubleClick so far because the software isn’t fully built out. Facebook says the product is superior to other ad servers because it uses the social network’s anonymous user identifiers–rather than traditional cookies–to track ads.
Publishers have mostly held back from using Atlas to track how ads perform on their sites in part because they are wary of Facebook's control of ads data. While DoubleClick’s technology is detectable on two-thirds of the top million most trafficked websites, Atlas is only found on 3.5%, according to Datanyze. Mr. Boland declined to say how many marketers are Atlas clients, but they include Microsoft, Tommy Hilfiger and LiveNation.
“It has taken longer than expected for Atlas to get adoption,” said Adam Berke, president and chief marketing officer of AdRoll, a Facebook ad-tech partner. He added that he’s optimistic about Atlas’ eventual adoption but its sluggishness now “goes back to the reality that they’re going after a market that has a lot of inertia. Agencies are not famous for being extremely fast to move.”
Of course, Facebook’s ads business overall is far from struggling. The company’s bright spot in ad-tech has been the two-year-old Audience Network, an ad network that was built in house to place and track ads on apps and mobile websites. Audience Network generated a run rate of $1 billion in for advertising spend for the first time last quarter, the company has said.
Because apps can use people’s Facebook identities to enable them to log in, app developers can easily add Audience Network software to leverage the social network’s data on those users. The app maker gets ad revenue, and Facebook gets a hefty cut for using its software and data.
Those ad sellers “get the benefit of Facebook’s knowledge about the end user. The quality of that targeting data leads to higher prices,” said Ben Tregoe, senior vice president of business development at Nanigans, a Facebook ad-tech partner. “You get an avalanche of advertiser demand that is clamoring for more Facebook-like supply.”
This kind of system is preferable for Facebook, which prevents the social network’s closely guarded identity data getting out. The ad network doesn’t allow marketers to have much control over where their ads go. Plus, Facebook controls the price instead of an open bidding process.
If Audience Network continues its momentum, it’ll continue to grow Facebook’s already rapidly growing mobile advertising pot from its flagship app. Eighty percent of Facebook’s advertising revenue–$4.5 billion overall– came from mobile in the fourth quarter, when it also started running ads on its popular Instagram app. That total represents an 81% jump from the same time last year.
LiveRail's private programmatic marketplace continues to operate, while the video ad serving part has been shut down. An earlier version of this story, which said LiveRail had been shut down, didn't distinguish between the two parts of the unit