On Tuesday June 9th, we held our first-ever developer conference called AppNexus Optimize. This inaugural event brought together over 100 of the brightest minds in advertising technology to learn, teach, and discuss the biggest challenges and opportunities facing not only our industry, but the internet at large. More than 20 CTOs, 10 company founders, and dozens more engineers, data scientists, and product managers descended on King’s Place in London where the AppNexus team led main stage presentations and closed-door workshops focused on the latest innovations taking ad tech into the future.
The importance of real-time insights for successful ad bidding
Consumers see a smooth rotation of ads on their favorite websites, but what happens in order for those ads to be displayed takes a huge amount of infrastructure. In the time it takes for a page to load, an ad request is made, the ad auction is created, responses are taken, and the winning bid logged. This information is provided to AppNexus clients as a log-level feed. Up till now, it has taken on average two hours for the stats on a particular transaction to be available, making real-time decisions on changing ad spend impossible to do with limited — and delayed — transparency into key decisioning data. Enter streaming log-level data, an AppNexus innovation made possible with a set of new solutions including Ratsoda, our homegrown complex event processor that deals with log-level data with a latency of 10-30 seconds. This real-time information on impressions and clicks makes real-time optimization possible, allowing publishers to divert spending after a certain level of conversions when additional spend loses effect, or changing spend to improve an impressions/clicks ratio.
A totally customizable bidding platform
One of the ongoing challenges in ad tech is automated campaign optimization. While streaming log-level data helps determine real-time insights, the issue is how to capitalize on them in a timely, visible fashion. What’s more, every company brings different needs and targets to the table – so a truly effective bidding platform needs to be customized in order to make smart automated decisions on ad spend. But, with 4.3 million ads served and 1.4 million active campaigns stored across 1500 servers, supported by 16,500 cores to execute real-time bids, it’s neither safe nor computationally feasible to allow developers to write their own code, when a single piece of bad logic could crash the entire AppNexus exchange. So, how did the AppNexus team set out to solve this problem? We created the AppNexus Programmable Bidder, a platform that accepts our customers’ own algorithms and executes their custom logic in real-time on every ad. The AppNexus Programmable Bidder is based on a new, domain-specific language called Bonsai that any publisher or agent can use to write their own decision trees for varying bidding based on features such as region, time or how many ads a particular user has seen over the lifetime of the advertiser, all in a drive to push performance and drop CPAs for advertisers. Let’s see how this works in action: Say you’re an advertiser who’s determined through A/B testing that people in California are more likely to respond to ads in the wee hours; while over in New York, buying impressions cheaply results in conversions. Bonsai lets you write a tree that varies how much you spend per bid based on a series of true/false branches. With real-time trafficking, changes go live within 5-10 minutes. At the same time, this language avoids the messiness of native code that’s too slow to analyze or validate for errors, protecting our clients from other people’s rogue code. Bonsai is ready to use now, with over 20 features available to set as yes/no nodes, including browser type, day of the week and OS, with many more to launch soon.
Closed door workshops
Along with the main stage announcements, our engineers and data scientists ran four workshops on engineering challenges with the new technologies, including building large-scale real-time systems, as well as how to use data science for writing algorithms to solve real-world problems, identify bot patterns, and tamp down fraudulent ads and sites.
The first of many
The inaugural AppNexus Optimize event closed with a evening of cocktails, sushi, and mingling. Since our launch as a scrappy but ambitious start-up eight years ago, we have always worked closely with clients and partners to empower them through developing the tools they need to get more out of our ad tech infrastructure. Today, as our company approaches 1,000 employees, we are providing a real-time bidding platform which any developer can customize to reach a better targeted audience and make savvier ad spend decisions. We can’t wait to see how our community leverages it to invent, improve and of course,optimize.