What We Learned About the Future of Ad Tech at Optimize 2017

Comments Off on What We Learned About the Future of Ad Tech at Optimize 2017

At AppNexus, our mission is to create a better internet by driving innovation in the way online ads are bought, sold, delivered, and experienced. We believe that advertising is the lifeblood of the internet – marketers invest in digital advertising, which funds the creation of high-quality content; publishers provide consumers with access to that content; and consumers pay for it with their attention to engaging and relevant ads. It’s a virtuous cycle that benefits everybody, and it all starts with the engineers, product managers, and data scientists who build these tools.

That’s why last Thursday, we again hosted our annual Optimize conference. Optimize is our way of bringing together the technologists leading innovation in ad tech to share knowledge with one another, describe what they’ve built over the last year, and discuss what’s next for the industry.

If you weren’t able to attend, then don’t worry – we’ve got you covered. In this post, we’ll break down the three mainstage talks of Optimize 2017 and tell you what they mean for ad tech.

Revolutionizing Click Prediction with AppNexus Programmable Bidder V8

 

Key takeaway: Data science is helping advertisers bid more effectively

Our first speakers of the day were three of AppNexus’ own: CEO Brian O’Kelley, Director of Data Science Abe Greenstein, and Manager of Technical Services Uri Bushey. They got onstage to announce the newest iteration of the AppNexus Programmable Bidder (APB) and describe the evolution of the product.

The best ad tech providers differentiate themselves from competition is building value-added technology. On the buy-side, this has traditionally meant building a bidder to compete in RTB auctions. The more granular control a bidder has (denoted by “expressiveness” on the graph below) the greater value it provides. The goal is to maximize expressiveness and minimize the effort on the buyer’s part.

At the bottom of the scale, an out-of-the-box DSP makes it easy to create a bidding strategy, but users are beholden to the bidding technology the DSP provides. Building a bidder from scratch, on the other hand, would let buyers create a highly expressive bidder, but with a lot of time, effort and resources.

In 2014, AppNexus began alpha testing the first generation of the AppNexus Programmable Bidder (APB): a set of APIs that allow advertisers to build their own algorithm and run it on top of the AppNexus platform, without the need to build the commoditized technology that makes up a standalone bidder solution. AppNexus strikes a powerful balance between expressiveness and effort, which started with the release of the Bonsai decision tree language. Bonsai lets marketers simply push a decision tree representing their bidding strategy into the AppNexus Console Bidder, which is then evaluated in real time for each impression.

Alpha testers were wildly successful but wanted more. One common request was the ability to more directly utilize various components of the AppNexus optimization engine. Another common request was support for more advanced generalized linear modeling techniques. The latest version of APB moves in that direction, with a combination of conditional component models and logistic regression models.

Conditional component models allow clients to pick and choose which AppNexus algorithms to utilize and which to override. For instance, clients with specialized, proprietary user data might prefer to use AppNexus’ click prediction models in conjunction with its own models for, say, probability of post-click conversion. The logistic regression API will allow clients to take advantage of a popular technique for predicting the probability of a “yes or no” event, such as a click or conversion, without the need to first translate model output to Bonsai.

These new APIs will be available for clients in the second half of the year. In the meantime, we are already seeing great results – our data science team has been  “dogfooding” our newest version of APB to improve our click optimization algorithms. Early testing has shown a 10% drop in CPC in initial testing. We are looking forward to seeing what clients can do with this powerful toolset!

 

How Schibsted Media Group Balanced Scale and Real-time Performance

 

Key takeaway: The real innovation comes from our clients building on our platform

One of our core beliefs at AppNexus is that our clients can build powerful new tools that solve their most pressing problems if we just give them the right platform to build them on. No one drove that point home like our second speaker and client Eugene Stipp, VP of Engineering at Schibsted Media.

Schibsted was founded 172 years ago as a printing company, but has reinvented itself several times over the years to become the global internet giant it is today. Schibsted owns several sites geared toward readers across Europe, South America, and Asia. Combined, the total traffic of all its properties exceeds 200 million unique visitors per month. When Eugene joined, he decided to build a stack of ad tech tools for all of Schibsted’s properties atop the AppNexus platform. He discussed two of them during his presentation: self-serve ads and the audience targeting engine.

Self-serve ads. Eugene saw that most of Schibsted’s revenue came from the “long tail” of small advertisers who preferred to just book small ad campaigns for a set price rather than go through a lengthy buying process. Eugene wanted to give them a painless, quick way to do that. However, the AppNexus console wasn’t built to accommodate thousands of advertisers booking campaigns at once without adding latency to Schibsted’s sites. Eugene’s team addressed this by building a queuing system in Amazon Kinesis and connecting it to the AppNexus Console, and voilà. The bottom line? Schibsted’s self-serve ads queue can accommodate over 200 campaign requests per second with a response time under 45 milliseconds.

Audience targeting engine. When Eugene first arrived at Schibsted, all of its user data – location, gender, interests etc.—was spread across multiple systems, making it hard to manipulate and draw insights from. He wanted a solution that could capture user data and segment the user accordingly in real time. By integrating with AppNexus’ real time data provider, Eugene’s team was able to build an audience targeting engine that can process 500 million pageviews per day, record behavioral and user data, and move the user to the right segment in under five seconds.

 

Full Steam A<head>: Two Years in the Header and Still Innovating

 

Key Takeaway: Header bidding is only getting more sophisticated

In our last mainstage talk, AppNexus Engineering Manager Matt Kendall and Product Manager Matt Jacobson talked to the audience about one of the biggest ad tech trends of the last two years: header bidding.

They covered the spread of header bidding over the last year and discussed their work building Prebid.js, but the most interesting part of the presentation might have been Jacobson’s breakdown of what’s next for header bidding in 2017. He focused on four main trends:

 

  1. More fairness and transparency. We believe header bidding will continue to even the playing field for advertisers, with more providers ensuring all demand partners see the same ad slots, time out settings, and bid bucket granularity. There’s also room for more transparency in header bidding, not just through open source code, but perhaps also through an open analytics framework that gives all parties on the transaction chain access to a common set of data.
  2. New formats. In 2017, we expect header bidding to spread to video, native, and mobile as each of these formats becomes a more prominent part of the programmatic landscape.
  3. Buy-side response to header bidding. As we continue to move away from the waterfall model, very few demand partners are going to retain the first-look privileges they might have with some inventory sources. There will be a more direct connection between buyers and sellers, which will enable advertisers to build smarter, higher-performing campaigns.
  4. Increased wrapper performance. Publishers will continue to demand more from their header bidding wrappers. We’re already seeing this in the rise of server-to-server header bidding, which is expected to lower latency and increase auction capacity by making ad calls from a third party server rather than the user’s browser.

 

While it’s impossible to know what the future holds, we fully expect header bidding to build on the momentum it’s established over the last two years.

Optimize 2017 started Q2 off with a bang, and reminded us once again how bright the future is for ad tech. It’s one of our favorite AppNexus traditions, and we hope you’ll get to join us next year!

 

To learn more about great events like Optimize, visit the AppNexus Events hub and start booking tickets now!

Filed under AppNexus Events, AppNexus Updates, Industry Perspectives.

Comments Off on What We Learned About the Future of Ad Tech at Optimize 2017

Comments are closed.