AdTheorent, Inc. announced the completion and implementation of its comprehensive anti-fraud infrastructure, enabling the company to deliver the first truly clean mobile supply for advertisers. Tapping its robust data-driven predictive modeling platform and capabilities, AdTheorent uses machine learning and statistical analysis techniques to identify anomalies and aberrational behavior impacting publisher inventory. Simultaneously, the company conducts on-going qualitative reviews of mobile web sites and apps within its network, assigning qualitative scores to each property based on property-specific screenings.
AdTheorent’s qualitative analysis focuses on identifying low value ad impressions that might not be detectable through statistical algorithms, such as sites optimized to take advantage of “mis-clicks.” In all cases, properties identified as suspect and potentially fraudulent are filtered from AdThoerent’s inventory. AdTheorent’s solution is unprecedented because it was designed specifically for mobile, it identifies fraud not discoverable using prevalent third party anti-fraud solutions, and it works in both mobile web and mobile app environments.
Ad fraud is a well-known and persistent threat to the value of mobile advertising, yet currently available remedies are incomplete. Recently, a study conducted by the Association of National Advertisers reported that up to 11% of digital display and 23% of video impressions were fraudulent, potentially causing $6 billion of lost revenue to global advertisers in 2015. “Ad fraud undermines advertiser confidence in the efficacy of mobile advertising, thus we have made a significant infrastructural investment to offer our advertisers supplemental protections that simply don’t exist in the broader market,” said James Lawson, AdTheorent’s Managing Partner & Chief Legal Officer. “Fraud detection solutions are en vogue at the moment, but our strategy is not to re-skin existing third party verification offerings, which have limitations. Instead, AdTheorent data scientists and software developers continue to innovate towards both data-driven and qualitative solutions that will advance mobile advertising into its next phase.”
AdTheorent’s proprietary fraud-detection algorithms are driven by terabytes of data which identify artificial, non-human behaviors and traffic patterns likely driven or directed by bots, as well as traffic whose source of origin is masked using proxy servers. And most important, because ad fraud, like all fraud, is fluid and dynamic in its forms and shapes, the tools and methods needed to identify fraud as it evolves must be equally dynamic and fluid, and must be able to learn from ever-changing data sets.
Before devising and implementing its broad-ranging anti-fraud measures AdTheorent relied on customary third party verification tools and used its machine-learning capabilities to assist supply partners in keeping their inventory clean. Ultimately, the company decided that eradication of fraudulent supply was too important to be entrusted with any supply partner or third party verification provider, few of whom have access to the volume or type of data necessary to detect sophisticated fraud or the technologies capable of filtering fraudulent impressions at a pre-bid level in both the mobile web and mobile app environments.
“A fraud-detection algorithm is only as good as the data that fuels it, and AdTheorent has access to vast amounts of data that third party verification tools simply do not have, which we use to detect fraud in ways that others simply cannot,” said Chris Cagle, AdTheorent’s EVP of Technology. “Unlike most third-party verification tools, AdTheorent’s software can detect fraud within app environments, which accounts for a large percentage of mobile inventory.” According to recent comScore audience data, 88% of consumer mobile Internet time is spent in apps, while 12% is spent browsing the mobile web. Added Cagle, “when our systems continued to flag fraudulent supply despite our use of third party verification techniques, we knew we would need to develop our own solution to supplement existing methods, which we have and it works.”
Moreover, AdTheorent’s machine learning and data analytics capabilities make it uniquely suited to combat ad fraud. AdTheorent’s predictive models are regularly deployed to identify ad impressions most suitable to particular advertisers’ goals given robust first and third party data ingested and analyzed by AdTheorent’s machine learning platform. AdTheorent’s systems are superior to other “big data” mining systems because AdTheorent leverages a unique methodology and architecture that makes it significantly more efficient. Built on GPU and leveraging distributed computing, AdTheorent’s modeling platform employs an in-memory internal representation that is supported by a unique “Learner” component in its architecture. This setup enables AdTheorent to overcome data volume and dimensionality barriers to develop models that are more agile and can be updated in real time. Simply stated, AdTheorent is able to ingest more data, enrich it more nimbly and efficiently, and create and update more predictive models than any other solution. Such models can search for both impressions most likely to engage with an advertiser, as well as impressions whose attributes reveal indicia of fraudulent non-human activity.
“To ensure a clean supply on our network, we tapped our data mining platform to do what it does best: use machine learning to deliver results, and in this case, it is creating models to detect fraud,” said AdTheorent CEO Anthony Iacovone. “The measures we’ve taken to eradicate fraud on our network, coupled with AdTheorent’s data-driven predictive targeting capabilities, will deliver the highest return on ad spend for our ad partners. The prevalent multi-exchange-connected DSPs are in most cases ignoring the fraud problem or trivializing its importance simply because they offer their advertiser users no viable solution.”