Pi Week: A Recap of AppNexus' Data Science Extravaganza

This post comes to us from Stephanie Tzeng. Stephanie is a Senior Data Scientist at AppNexus, primarily designing and prototyping algorithms to improve bid optimization for buyers in our marketplace

In the technology age, people and organizations have been able to generate an unprecedented amount of data, tracking different transactions, connections, images, text – anything you can think of. Now rich with data, businesses were hungry for insights from all this new information! It was only logical that the data science field would rapidly expand to fill this appetite.

So, what is data science? Data science is a field that combines mathematics, modeling, analytics, and computer science to find information from the exponential increase of data.

Why is this important at AppNexus?


At AppNexus, we use this data to empower our customers. Our products make automated, intelligent decisions on behalf of our clients. We also use data science to build visualization and analytics tools to inform our clients on how to use our product and how our marketplaces are performing. Data science is also crucial for keeping our ecosystem clean – we have a team dedicated to researching and preventing invalid traffic on our platform.

By helping our customers find the optimal interactions between advertisers and websites, we make it easier for them to serve their advertisements to the best audiences and improve engagement for Internet users. For example, our newest decision-tree based algorithm modifies bids depending on how recently and frequently particular users have seen an ad and how they are expected to react. It is built using our in-house Programmable Bidder.

We also strive to spend buyers’ money efficiently while maximizing revenue for our sellers. To do this, we have built a reserve price optimization system as part of the AppNexus Publisher Suite, which utilizes an online learning approach. In fact, this algorithm was published in an award winning paper! In conjunction, by patrolling the depths of the web, the data science team strives to be a huge slice in contributing to the AppNexus mission – improving the Internet.

But while it’s incredibly valuable, data science works in the background, so it can end up flying under the radar not only for our customers, but even for some of our own employees. After AppNexus made a big effort to invest in data science, we knew that it was important to clear up any mysteries about the team -- especially internally.

The main course


We wanted to capitalize on one of our core values, learning and teaching. We were able to do this by launching with a good old-fashioned Razzle Dazzle.

What is a razzle dazzle? It’s an internal, AppNexus event focused on knowledge-sharing and cross-team collaboration. Fortuitously, we planned the Data Science Razzle Dazzle for the most appropriate date possible: March 14, 2017, also known as Pi Day.

We were so excited about data science that we expanded / leavened Pi Day to an entire week! Think of it as Pie-themed Razzle Dazzle a la mode x 7, each extra scoop of ice cream represented as a separate breakout session or event. We picked out the data science topics most people find confusing and set out to shed some light on them.

Questions we wanted to answer included: What are Big Data technologies, and which ones do we like to use? What is the fundamental idea behind Machine Learning? What role does AI play at AppNexus? What is invalid traffic in the advertising ecosystem? How do analytics play into empowering our customers? How do AppNexus data scientists build products?

We tackled those questions with events like:

  • 5 Quick and Easy Tips to Sound Like the LeBron James of Machine Learning at Your Next Cocktail Party
  • How Data Science is Changing the World and How You Too Can Be Part of This Revolution
  • These 10 Facts About The Data Science Technologies Will Shock You


We ended with a project themed cocktail event in our headquarters, featuring V8/tomato juice; v8 also happens to be the name of our logistic regression click prediction algorithm.

Our goal for the week was to enlighten AppNexians on data science technologies and techniques, educate people on how data science empowers us to build sophisticated automated products, and show how our algorithms can help our clients roll in dough.

Based on those goals, I deem this first internal data science conference a great success! Each of our global breakout sessions reached 40 to 60 AppNexians, and our kick-off session reached more than 150 employees. With somewhere between 15-50% of the company attending the events, my best estimate is that we engaged 1/3 of the company.

I apologize for all the saccharine pie puns, but I was really on a roll from last week. Just a few more, I promise.

I leave you with one last tasty treat: a special video our data science team has created to demystify our team structure and how we work on projects here.


If this post hasn’t filled you enough with data science, come to Optimize where we’ll have not just data scientists, but also the engineers and product managers driving innovation in adtech. Don’t flake out!

  • AppNexus Updates
  • LifeAtAppNexus