GABRIEL WEINTRAUB: When you look at the current practices with marketplace design and rules in the digital ad marketplaces, you get the sense that a lot of them grew out of necessity, out of needing a quick solution. So I think there is an opportunity over time, as these markets become more mature, to try to get this auction right. This is what AppNexus would like to do: How do we build a marketplace that is on one hand transparent for buyers, but on the other hand achieves high revenue for sellers?
Publishers face low bid density and huge gaps between the bid price and the clearing price. To fix this, publishers fiddle with floor prices, sometimes in ways viewed as unfair. What’s your take on this issue?
The fact some of the markets are so thin [i.e., have low bid density] is in some sense a side effect of how efficient they are. Advertisers are targeting at such a detailed level that it naturally creates thin markets. The advertisers that won the impression targeted the exact user [they wanted]. It’s a symptom of all the efficiencies the market is creating. The alternative would be you don’t target. You would have more competition, prices would be higher, but you wouldn’t be creating the value through targeting.
What’s another factor causing low bid density?
If you think systemically, buyers are not presenting all the bids. If I’m a DSP, I will only submit one bid, my highest bid from my client. They are withholding bids. Sellers react with floors, to try to extract some of the value being created. That’s another cause of this low bid density.
I have some colleagues at Columbia University that are suggesting that in a world where there is no bid withholding, and all bids are submitted to the exchange, that decreases the incentive of publishers using aggressive floors. As a result, you can show that in many scenarios everyone is better off, both buyers and sellers.
How does the rise of header bidding impact the auction?
We have been thinking about header bidding auctions intensely. On a high level, there is a benefit in the ecosystem, which is to penetrate a closed system, which was DFP [DoubleClick for Publishers], and to create a thicker market. That could have bid benefits. What we are thinking about at AppNexus is what the right auction logic should be for header bidding. Should it be the same as RTB auctions, or should it be different?
What are the alternative auction options that could work for header bidding?
We are right in the process of figuring it out – it would be too speculative to say. But you want to come up with an auction logic that allows you to compete head-to-head with [Google] AdX, so that our buyers from AppNexus are not disadvantaged when competing head-to-head with AdX. But at the same time it should be an effective tool for publishers in monetizing their inventory. Balancing these objectives requires thinking carefully about the auction rules and logic.
Do you think buyers bidding against themselves is an issue with header bidding? Many disagree on whether this is an issue.
Maybe in theory it could happen, but in practice we still don’t know how relevant this is going to be.
Some theorize that the presence of ad fraud or bad-quality publishers impacts prices for everyone. What’s your take on that?
A [successful] market has three components: thickness without being congested, the ability for buyers to interact without friction and safety. Safety is a broad concept, but one meaning in this context is that you want to reveal your value. If you trust the marketplace, you are willing to reveal your willingness to pay for an impression. If markets aren’t transparent and don’t feel safe, then maybe they are willing to pay $1 but only bid 80 cents. Transparency could lead to more trust and safety in a way [where] everyone benefits.
Facebook doesn’t use an RTB second-price auction, instead using a Vickrey auction. What’s the difference?
[Vickrey is] truthful. Everyone is always incentivized to reveal their true preferences and willingness to pay, which is what people call a dominant strategy. It doesn’t do gaming. For buyers, it’s very simple. They don’t need to think about strategic behavior. In other scenarios, the buyer has to strategize – “How much should I shade my bid to make it optimal?” – and create algorithms for optimal shading parameters. In a truthful auction, you need none of that. You just need your true value. In the context of display advertising, truthful auctions are appealing.
When will the industry start seeing the outputs of your auction research starting tests?
I hope we start seeing them later this year, but they are complex and involve many parties. We are hopeful that we will start seeing them in the medium term.
Is there precedent for changing auction rules in advertising marketplaces?
The search marketplace had a huge change fairly early on. It started as a first-price auction, and that can actually be very unstable. So they came up with the generalized second-price auction. I think [with RTB] there will be small, incremental innovations that will slowly improve the marketplace. It will be hard from one day to another to transform the entire marketplace, and that will be too disruptive.
Would changing the auction require buy-in from industry organizations or other players, or could AppNexus just roll it out by itself?
We are a multi-sided platform, so we need buy-in and to convince multiple participants that it’s a good change. Some of the changes I think we could roll out just in AppNexus, and if they work well I wouldn’t be surprised if others follow.
This interview has been condensed and edited.