January 29, 2013
For Immediate Release – New York, NY - AdTheorent, Inc., the first intelligent Real Time Bidding (RTB)-enabled mobile ad network, today announced the introduction of the industry's first real-time learning and predictive modeling platform. The AdTheorent Real-Time Learning Machine (RTLM) learns in real time, generates data-driven predictive models "on the fly" and predicts faster than any other data mining technology, yielding demonstrable results for AdTheorent's mobile advertisers. During the 2012 holiday shopping season, the AdTheorent RTLM delivered an average engagement level of 200-300% above industry average for some leading retail brands.
Based on the data mining technology developed by AdTheorent's Chief Data Scientist, Dr. Saed Sayad, AdTheorent's RTLM application analyzes 50,000 bid requests per second on a single server, filtering out bids with a low probability of click, conversion or awareness lift. The RTLM system "learns" from incoming bid requests and builds and modifies predictive models as it learns, applying such models in live campaigns to match each mobile advertisement with the optimum mobile impression. As a result of AdTheorent's RTLM system and its unprecedented ability to filter-out undesirable targets, participating advertisers have enjoyed improved engagement levels, such as an uplift in click through rates (CTR) and awareness, and reductions in cost per acquisition (CPA) rates. In some campaigns the uplift has been as high as 500%.
For agencies, AdTheorent's RTLM creates a new paradigm for mobile media planning.
"The value of data in media is immense and intelligent technology companies are now using that data across mobile devices. This is enabling real-time learning and prediction, which is presenting a great opportunity to move marketers toward data-driven results," said Sal Candela, mobile director at PHD Media. "These new developments offer real promise in enhanced performance and brand lift, a combination that will spark interest across the mobile marketplace."
AdTheorent's RTLM application is a foundation of its second-generation mobile ad network, facilitating powerful modeling for click and post-click analytics and providing real-time intelligence for more effective ad targeting. AdTheorent's RTLM leverages AdTheorent's recently introduced Traktion product, which affords mobile advertisers a seamless way to track post-click behavior across the entire spectrum of mobile advertising channels. The RTLM's power to learn in real time, model "on the fly" and predict faster than any other data mining technology distinguishes AdTheorent from other mobile ad networks.
Said AdTheorent CEO Anthony Iacovone: "RTB companies face a big data problem in modeling terabytes worth of data every month. RTLM is the only solution that can scale to meet the big data demands to produce result driven models in the RTB world. Our RTLM system is designed for mobile advertisers, and we anticipate a greater willingness from advertisers seeking to deploy real-time bidding in the coming year."
Using the RTLM, advertisers can increase CTR (and decrease CPA) by dividing the population and filtering various segments, and then more definitively targeting appropriate audiences. In a recent CPA engagement, AdTheorent's RTLM was able to run 2000 model variations in under five seconds to extract the most efficient model for prediction of conversion events.
"The massive amounts of data that marketers need to filter in order to gain traction with target audiences require systems that can develop greater intelligence within actual campaigns, and we are now seeing products that seem to do this in a precise manner," said Joseph Rose, Vice President, Associate Analytics Director, MediaVest WorldWide. "Our clients are just learning about the real-time bidding on mobile technology, and now we have some compelling field data with which to articulate reasons for deploying the next generation in ad network technology."
The mobile marketing industry is beginning to take notice of how RTB can be more effectively applied to mobile advertising.
"Mobile gets you closer to the consumer than any other media. Marketers that understand this are increasingly leveraging mobile's ability to provide real-time context. This context, when leveraged appropriately, gives the marketer the ability to instantly adjust their programs and how their customers view and engage with a campaign's media," said Michael Becker, North American Managing Director of the Mobile Marketing Association. "MMA members, like AdTheorent, are providing the marketplace with cutting edge, real-time, capability that generates results and we applaud them and thank them for their innovations and industry leadership."
Real-Time Learning Machine: How it Works
The AdTheorent RTLM leverages technology that processes Big Data for real-time analysis and scoring, based on criteria including advertiser's demographic data, geographic data, publishers' data and other information. The RTLM enables variables to be added or removed from the analysis as data evolves so that, for example, if one demographic data point were removed (e.g., "women age 25-49"), the RTLM would almost instantly re-calibrate the predictive model without that data.
Conventional data mining algorithms operate in a batch mode, where having all of the relevant data at once is a requirement. Due to large increases in the rate of generation of data, the quantity of data and the number of attributes (variables) to be processed, the data situation is, increasingly, now beyond the capabilities of conventional data mining methods. AdTheorent's RTLM uses a three-stage data analysis approach that refines billions of bid requests, removes extraneous data, and delivers the most accurate and efficient targeting for advertisers' mobile campaigns. The RTLM provides the only viable predictive modeling platform to process Big Data with zero-latency. The power and efficiency of AdTheorent's RTLM allows one data scientist to perform the work of 10.
AdTheorent's RTLM is unprecedented in mobile advertising, featuring:
- Incremental learning (Learn): immediately updating a model with each new observation without the necessity of pooling new data with old data
- Decremental learning (Forget): immediately updating a model by excluding observations identified as adversely affecting model performance without forming a new dataset omitting this data and returning to the model formulation step
- Attribute addition (Grow): Adding a new attribute (variable) on the fly, without the necessity of pooling new data with old data
- Attribute deletion (Shrink): immediately discontinuing use of an attribute identified as adversely affecting model performance
- Scenario testing: rapid formulation and testing of multiple and diverse models to optimize prediction
- Real Time operation: Instantaneous data exploration, modeling and model evaluation
- In-Line operation: processing that can be carried out in-situ (e.g.: in a mobile device, in a satellite, etc.)
- Distributed processing: separately processing distributed data or segments of large data (that may be located in diverse geographic locations) and re-combining the results to obtain a single model
- Parallel processing: carrying out parallel processing extremely rapidly from multiple conventional processing units (multi-threads, multi-processors or a specialized chip).