Presented By O’Reilly and Cloudera
Make Data Work
September 11, 2018: Training & Tutorials
September 12–13, 2018: Keynotes & Sessions
New York, NY

AppNexus's stream-based control system for automated buying of digital ads

Brian Wu (AppNexus)
4:35pm–5:15pm Wednesday, 09/12/2018
Average rating: *****
(5.00, 1 rating)

Who is this presentation for?

  • Software engineers, data scientists, and Ad-tech professionals

Prerequisite knowledge

  • A high-level understanding of streaming technology, such as Kafka and the lambda architecture (useful but not required)

What you'll learn

  • Understand how the lambda architecture contributes to a clean and modular application design
  • Explore a real-world use case of streaming technology and PID controllers


Automating the success of digital ad campaigns is complicated and comes with the risk of wasting the advertiser’s budget or a trader’s margin and time. Brian Wu describes the evolution of Inventory Discovery, a streaming control system of eligibility, prioritization, and real-time evaluation that helps digital advertisers hit their performance goals with AppNexus.

AppNexus has many tools that assist traders with the buying of digital ads. However, before Inventory Discovery, customers kept complaining about two big problems: achieving performance took too long, and manual optimization took too much effort. Given the scale of, well, the entire internet, the pool of inventory that an ad campaign can serve on is huge, and traders don’t know which inventory is going to perform. Even with savvy traders and informed targeting, advertisers run the risk of “boiling the ocean” (i.e., spending money on everything but never enough on a particular slice of inventory to collect significant data). Boiling the ocean wastes money and time. It may take a while to accrue enough meaningful data to call inventory “good” or “bad.”

When a trader identifies that inventory is bad, she may spend several hours a day running reports and excluding the inventory that doesn’t work. Because this takes time, most traders will implement changes once per day at most, which means they cannot respond quickly or dynamically to market changes. If cutting bad inventory is taking too much time, the trader might implement a whitelist of the best domains and placements. This might help performance in the short term, but it also prevents the ad campaign from discovering any new performant inventory sources.

From case studies and research, AppNexus knew that the answer to these problems lay in finding the right inventory (placements and domains) for each advertiser. Three years ago, the company set out to create an entirely automated product for advertisers to solve these problems—to take an advertiser’s campaign from no data to fully optimized without wasting a trader’s time or an advertiser’s budget.

Brian describes the dynamic evolution of Inventory Discovery, AppNexus’s stream-based real-time control system for choosing the right inventory for an ad campaign. The journey included many wild ideas, many MVPs, and many mistakes as well as a persistent and patient team.

Topics include:

  • How a simple concept became complicated, and then simple again
  • How an ugly eligibility problem turned into an elaborate control system
  • Ad campaigns it worked well for and ad campaigns it didn’t
  • A PID controller that solves the explore/exploit problem
  • A rigid ranking process become streamlined
  • The final result: a completely automated solution deeply integrated with lambda architecture and streaming technology
Photo of Brian Wu

Brian Wu


Brian Wu is an engineer on the AppNexus optimization team, where he has worked closely with budgeting, valuation, and allocation systems and has seen great changes and great mistakes. Coming from a pure mathematics background, Brian enjoys working on algorithm, logic, and streaming data problems with his team. In addition to control systems, data technologies, and real-time applications, Brian loves talking about process, team work, management, sequencers, synthesizers, and the NYC music scene.