Mobile Ad Revenue Lags

Publishers seeking to create more competition and access for programmatic buyers have embraced header bidding over the past year.

AppNexus first brought header bidding technology to market in 2009, back when RTB first started, but not many of its customers used it. Now it’s seeing increased demand from publishers adding its header bidding (or pre-bid) technology to drive up yield.

“We believe the reason pre-bid is taking off is because of DFP [DoubleClick for Publishers] dynamic allocation,” said Ryan Christensen, GM of AppNexus’ global publisher business.

With dynamic allocation enabled, DFP allows Google’s own demand, AdX, to see and bid on every impression, but not other exchanges. That hurts overall publisher yield and partners who want to compete with AdX. “It’s closed. No one else can participate,” Christensen said.

Header bidding is the only way for other demand sources, like AppNexus, to get equal footing in the auction with Google AdX for DFP publishers.

But there are costs: Header bidding code slows load times, affecting the user experience, Christensen said.

Header bidding technologies come in different flavors to mitigate the downsides of the tech and improve yield. AppNexus touts features that allow for reduced latency in international markets and better price granularity.

“We’ve heard some horror stories of publisher integrating partners and having slower load times, especially internationally,” AppNexus product VP Arel Lidow said. “Latency is a major issue for pre-bid ad tech,” he added.

Because AppNexus has localized data centers, publishers in Europe and Asia will likely notice AppNexus returns bids faster than other partners. But AppNexus said there’s still great variation in latencies between header bidding partners, and its response times can rise to the top in the US too.

The product reviews, tech and science publisher Purch has worked with AppNexus for the past two months and with a couple of ad network partners using AppNexus pre-bid technology for eight months. It also works with about a dozen header bidding partners.

Marc Ropelato, Purch’s director of programmatic revenue, has observed that speed varies depending on the integration. It uses the slower but easier to implement client-to-server integration with AppNexus, which adds 200 to 300 milliseconds compared to the more complex server-to-server integration it uses with the other partners using AppNexus’ technology.

The publisher is pushing all its partners to integrate with Purch in the faster server-to-server setup, and will likely explore that option with AppNexus too.

AppNexus said it found a workaround to another header bidding issue: returning pricing information, down to the penny, without adding latency.

For header bidding, data precision matters. Some tech partners only pass general information into the ad server – a “yes/no” or wide price tiers – in an attempt to speed up the process and make it easier for ad ops.

But the trade-off with using estimates is that publishers lose pricing accuracy and, with it, overall yield. Publishers can end up accepting an impression worth less than they think (a $1.05 impression in a price bracket of $1 to $1.50, for example).

“We believe there’s actually no excuse not to return price,” Lidow said. “Partners that can’t return an actual price return the risk back to the publisher from ad tech vendor.”

According to Ropelato, about half of the header bidding partners Purch works with create penny-specific price granularity, including AppNexus. It set up AppNexus’ technology so it passes bids into the ad server in penny increments all the way up to a $20 bid. Then the ad server can decide who wins based on the actual price of the bid, not an estimate.

While some header bidding partners pass on minimal data, Ropelato appreciates that AppNexus shares more complete data about bid requests. Purch can quickly transmit that data via APIs to its data warehouse.

Although Purch is still only a couple of months into its AppNexus integration, it’s seen strong results from moving to header bidding setups with all of its partners.

A key idea behind header bidding is to increase pricing pressure on Google AdX. Since it’s started adding header bidding partners, Google AdX buys roughly the same amount of impressions, but for a higher price.

Plus, by bringing in more demand, header bidding raised clearing prices above what ad networks could pay. “We have very little remnant inventory now. Most of our ad networks we’ve shut off, because they can no longer compete with RTB,” Ropelato said.

Purch’s success running outside the ad server with little downside is the exception, not the rule. It created a proprietary system so it could run multiple header bidding partners without dramatically slowing down page load times.

The best way to address this problem is to rework the ad server, but Google has little incentive to remove the advantage it gives AdX.

AppNexus may be in a position to spur such an evolution.

With its acquisition of the ad server OAS from WPP-owed Xaxis in September, AppNexus now has the ability to create an alternative to DFP by unifying demand within the ad sever, not outside it.

AppNexus wouldn’t say how it’s set up OAS to receive demand from its own sources. But it claimed that unlike Google, which keeps its algorithms and data to itself, it’s “completely transparent with publishers on the platform about how the system works, and we are also transparent with buyers,” Lidow said.

AppNexus is delivering its own full-stack ad server later this year, Christensen said.

Will it unify demand from other sources in its ad server, or give AppNexus demand an edge, as Google has done with DFP?

“We will approach things in a more open way by DNA,” Christensen said, while declining to give more details.