Rocket Fuel Uses Hadoop Yarn and Apache Spark to Model and Score 85 Billion Bid Requests Per Day—a 90% Increase Year-Over-Year

Rocket Fuel (NASDAQ: FUEL), a leading programmatic marketing platform provider that uses artificial intelligence (AI) at Big Data scale to optimize marketing ROI for global agencies and enterprise marketers, today announced the next generation of its AI and modeling system. Marketers are already starting to realize the benefit of this upgrade with campaign performance improvements of 2x on average, compared to the previous system. The scalable architecture evaluates over 85 billion auction opportunities every day—an increase of 90% compared to the average auction opportunities served in the second quarter of 2014.

“Rocket Fuel’s next-generation technology system builds on a very strong foundation of innovation created by a world-class engineering team,” said Manu Thapar, SVP R&D at Rocket Fuel. “Over the past year, Rocket Fuel has devoted significant resources to R&D, resulting in the ability to very effectively learn from over 200 terabytes of data collected every day. The ability to automatically process more data faster, using further refined machine learning models, continues to give Rocket Fuel’s technology a tremendous advantage over competitive solutions that use simple algorithms and manual optimization. With these improvements, our customers win with better-performing campaigns that exceed their goals.”

More Data, Faster Learning, and Precise Targeting Equals 2x Better Campaign Performance

Rocket Fuel’s AI engine supports budgeting, prediction, and optimization in real time to deliver dramatically improved ROI for both direct response and brand-oriented advertiser goals. Over the past 12 months, its aggregate “composite index” of model performance indicates that the combined improvements to modeling, infrastructure, and supply availability have resulted in an average of 2x better performance for advertisers, adjusting for campaign size and mix. These AI updates enable marketers to see improved performance against their goals for direct response and brand campaigns, including return on ad spend (ROAS), lift rate, viewability, mobile lift, and video completion rates.

By the Numbers:

  • Average 2x performance improvement across all campaigns, adjusting for campaign size and mix
  • 85 billion bid requests evaluated each day, nearly 1 million per second
  • 200 terabytes of data collected per day to achieve impression-level targeting
  • 55000 CPU cores and 68 petabytes of data stored across 12 redundant data centers worldwide
  • Processing speed on 100% of bid requests of 10’s of milliseconds—well below industry threshold
  • On average, thousands of attributes and behaviors known about each profile

Cross-Device Clustering and New Machine Learning Models Deliver Highly Individualized Ad Targeting in Milliseconds

The next-generation system represents a significant breakthrough in developing machine learning models, and applies them to highly personalized ad targeting. Specifically, new models have been developed that leverage cross-device clustering, allowing marketers to identify and reach individuals on any device, as opposed to traditional audience segments. The system can then deliver the right message, at the right time, on the right device, and to the right individual, learning what works and doesn’t work through scoring and applying that insight to the next interaction. The AI system learns from 68 petabytes of data and accesses thousands of different attributes like demographic, behavioral, location, preference and device data for each of the nearly 4 billion+ unique and anonymous IDs in real-time to make these decisions.

The next-generation system’s ability to evaluate over 85 billion bid requests daily is attributable to improvements to the scoring speed on a per-bid basis which, at 10s of milliseconds using multiple machine learning models, is now well under the threshold generally set by ad exchanges. In addition to leveraging open-source Hadoop for its distributed computation needs, the next-generation system leverages cutting edge Yarn and HBase technologies to increase the amount and speed with which the new engine can analyze petabytes of data. Rocket Fuel is also leveraging Apache Spark’s open-source computing framework for faster and advanced machine learning. Some of these innovations were presented at the recently held Hadoop Summit and HBaseCon 2015.

The system now supports international exchanges and trafficking on Euros, Yen, and other major currencies.

About Rocket Fuel

Rocket Fuel combines the science of Artificial Intelligence with the scale of Big Data to improve the effectiveness of programmatic marketing. Customers trust Rocket Fuel's Marketing That Learns™ to achieve brand and direct-response objectives in diverse industries across North America, Latin America, Europe, and APAC. With the acquisition of marketing technology firm [x+1] in September of 2014, Rocket Fuel now offers a complete programmatic marketing platform for the world's most innovative, always-on marketers. The platform includes data management, programmatic media-buying, site optimization, and predictive analytics capabilities that extend across a marketer's paid and owned channels, and personalize every customer interaction. Rocket Fuel operates in more than 20 offices worldwide and trades on the NASDAQ Global Select Market under the ticker symbol "FUEL." For more information, please visit http://www.rocketfuel.com or call 1-888-717-8873.

Rocket FuelKenya Hayes, 650-481-6178pr@rocketfuel.com

ROCKET FUEL INC. (NASDAQ:FUEL)
Historical Stock Chart
From Mar 2024 to Apr 2024 Click Here for more ROCKET FUEL INC. Charts.
ROCKET FUEL INC. (NASDAQ:FUEL)
Historical Stock Chart
From Apr 2023 to Apr 2024 Click Here for more ROCKET FUEL INC. Charts.