Facebook’s head of ad tech, David Jakubowski, will participate in a panel discussion on sell-side impressions at the May 24 CLEAN ADS I/O conference in New York.
As Facebook expands its Audience Network, it faces a number of challenges.
Keeping out fraud. Achieving high viewability. Ensuring off-site placements still drive performance.
By directly integrating with publishers versus buying inventory through the open exchange, Facebook has given itself back some measure of control.
But it operates a $1 billion business that takes place off its owned-and-operated sites and in environments where it wields less control. And it added video advertising to Audience Network on Monday, an area known for high levels of fraud.
Ensuring Facebook Audience Network meets the same standards as its news feed is one of the jobs of David Jakubowski, Facebook’s ad tech head who oversees Atlas, LiveRail and Audience Network.
AdExchanger talked to Jakubowski about how Facebook is monitoring its publisher relationships, including on Audience Network.
AdExchanger: Atlas played a part in detecting fraud on LiveRail, right?
JAKUBOWSKI: We used our measurement tools, and Atlas is a measurement platform. LiveRail is one exchange. But there are lots of exchanges where we did our testing in programmatic buying. What we found is that very little of the inventory out there has any value at all.
On the low side, you have people saying it’s 17% to 20% bots. On the high side, 30%. That’s known bot fraud. And you have all this gray stuff in the middle, where there are techniques like ad stacking or changing the ads in tabs on top of the browser. We are able to clearly identify who the real people are, on the white end of the spectrum.
So Facebook killed its buying platform because there’s no point of having a buying tool if you can’t buy quality inventory on it?
That is the reason. It’s not that we didn’t want to do buying. But the open inventory in the ecosystem has very little value. We didn’t want to build a product and sell it to marketers knowing that there was no value on it. We have these unique measurement tools to determine those things. We couldn’t in good conscience move that forward. Now, because we know about it, we feel like we have a responsibility to tell marketers what to do next.
So where should marketers buy if not on the open exchange?
What I would say to them is make sure you are accurately measuring real people and if they saw an ad. Start there, and here is the crazy concept: Did they buy something? Did they do something that has actual value? One-third of marketers don’t measure squat. Start there.
Now, evaluate each of the places that you buy. Our research suggests that the people who are plugging into the publishers directly are having the higher-quality audiences first. And that makes sense. If you’re a publisher, you’re going to send the things to the exchange that you don’t want to sell to anything else. And that’s why we shifted the strategy on Audience Network as well. It’s curated, publisher-direct only.
What else are you doing to keep Audience Network clean?
The ads go to the white-hat end of the spectrum. Audience Network runs exclusively to the known people side. We started in-app because it is easier to control viewability and it’s a better, safer environment. We’ve now taken those learnings and expanded to mobile web, and on Monday expanded our video offerings. We know we have a base where we can identify those audiences when they are viewable. And we are curating it by going publisher-direct.
Facebook kicked out many of its LiveRail partners because of fraud issues. Why be so public about this?
We’ve been very transparent about how low-quality the inventory is, and it’s because we look at it through a value-based lens. We’ve also been public about the pieces of the business we aren’t going to continue to support because of it. And we shut down a lot of inventory. We pulled back on our buying initiative that was designed for open exchange buying. The biggest marketers in the world don’t want low-quality inventory. The minute you shift to value, the things that are white and black become much easier to see.
Google has acquired companies like Spider.IO and employs thousands of people all to root out fraud and police its network. Will you have to deploy resources like that for Audience Network?
The answer is no. But it raises an interesting question, which is why is our bar higher than others? What aren’t they seeing that we are, when we are not willing to put our name on certain pieces of inventory? And how does a marketer decide how much low-quality inventory they are willing to tolerate? Because if you are buying in those open exchanges, and you are price optimizing and not value optimizing, you are by definition playing in some garbage inventory. If you move to measurement goals like sales, you are more likely to see where the high-quality inventory sits.
I’ve spoken to one buyer that didn’t use Audience Network because it didn’t perform. Another said it used to have performance issues but that’s no longer the case. But that’s anecdotal. What do you see as the gap in value between the news feed and Audience Network, and what does fraud and viewability have to do with it?
The fraud problem with Audience Network is solved for the time being. It’s basically fraud-free. The pricing mechanics are such that Audience Network is benchmarked against the feed, and it shouldn’t deviate too far from performance in-feed. We have a thing called the advertiser outcome score, which was not available at the Audience Network launch, but we added because of this issue. The system calibrates according to whatever you give it – in-app conversions, for example – and predicts the value that an advertiser is going to get from going off Facebook. It then compares it to what you would be getting on feed, and adjusts the price to reflect higher or lower value in news feed.
How many publishers have you had to kick out of Audience Network?
We have a self-service tool to apply and a human-curated team that goes through and approves them. It starts upfront. From there, the tech takes every single ad request and looks at it and says, “Is this a quality user?” If it’s not or we don’t know, we send it back to the publisher.
This sounds like a pre-bid solution.
That’s a term for the open programmatic exchange world. We are plugging in our SDK directly to that publisher so we don’t have to do any of those gimmicky, complicated things. We simply return their request to their inventory management system.
So if there is a publisher with lots of bad inventory, they just get 10% fill rates?
At that rate, you’d get a phone call. If 90% of what you are sending us is crap, either something is mechanically wrong or something is wrong with you as a publisher.