What went wrong?
“We were able to deliver ads to real people with unprecedented accuracy, but came up against many bad ads and fraud (like bots),” Dave Jakubowski, head of ad technology at Facebook, notes in a new blog post.
“While we were fortunately able to root out the bad actors and only buy quality ads, we were amazed by the volume of valueless inventory,” according to Jakubowski.
In trials, Jakubowski and his team also discovered that native and video ads were the only two formats that delivered sufficient value. As a result, they removed over 75% of the volume coming from the exchange by turning off publishers circulating bad inventory into its LiveRail unit.
“We knew that in good conscience, we couldn’t sell what Atlas and our people-based measurement told us was valueless,” Jakubowski conceded. “Unfortunately, those ads were almost certainly dumped into another low-quality exchange where all of them were most likely purchased.”
The move follows Facebook’s decision to shut down the ad-server portion of LiveRail, in January. At the time, the company said it would continue to focus on LiveRail’s automated sales of publishers’ in-app mobile video sales and native display ads.
Going forward, Facebook plans to focus on native and video, in addition to mobile.
Under the Atlas banner, Facebook is also rolling out Offline Actions -- a new measurement tool that ties offline sales to online ad spend.
Advertisers that measure their ads with Atlas can now upload their point-of-sale (POS) data, and confirm whether or not their online ads are influencing offline purchases in near real-time.
The company is working with advertisers on Path to Conversion -- by device reporting -- to provide insight into all the ways that real people see ads across multiple devices before making a conversion. The offering does not rely on cookie-based reporting, which Facebook counts as its key differentiator.
Atlas is also rolling out Video Ad Serving this week. The new service should be broadly available by March, Jakubowski said on Monday.