Scalarr announces the results of its new AI-powered service called DeepViewTM that analyzes millions of data points to detect even the slightest anomalies in CTV, mobile web, in-app, desktop, and identify patterns that signal fraud among them.
DeepViewTM has been deployed for several months now and its initial findings are remarkable. Taking an example only of one AdExchange, the anti-fraud solution found that up to 40 million video impressions served for ad campaigns (CPM~$3.7) were invalid.
DeepViewTM has detected more than 4,000 fake botnet apps, over 2 million fake device IDs, more than 500,000 IP addresses, and over 10,000 fake user agents. The data was emulated in a refined manner, and up to $7 million dollars in yearly losses were estimated for the client.
In this case, cybercriminals imitated the impressions of supposedly real users who watched ads in trusted, popular sources*, including:
Fraudsters were utilizing the Deepfake approach, which is based on Generative Adversarial Networks (GANs) and reinforcement learning. These algorithms helped them to fake video, voice, and news articles.
Scalarr’s new most advanced AI-powered anti-fraud solution harnesses the power of data accumulated within the last 4 years - more than 500K samples of smart bots and 100M+ samples of other fraud types. In our case, we utilize transfer learning techniques to protect our clients from fraud.
“From our strategic vantage point, we were able to see how ad techs were dodging increasing cybercrime and fraudulent activity,” said Inna Ushakova, Scalarr’s CEO and co-founder, “our endgame was to build a robust solution that protected that part of the funnel. We’re extremely proud of the results DeepView has yielded and always strive to help our clients grow their revenue based on fair results and transparency that is not compromised by invalid traffic.”
To learn more nuanced details about this massive botnet that was harming our client’s traffic quality and generating severe losses, read our case study where we provide all relevant information.