Share on LinkedIn
Share on XAd fraud is sophisticated and purpose-built to avoid detection. Bad actors attempt to identify new vulnerabilities, gaps, and attack vectors every day. The constant evolution of ad fraud requires a connected, multi-stage approach to detect and avoid.
At IAS, we use a combined, AI-powered approach to detect ad fraud and protect our advertiser and publisher customers:
Pre-Bid
IAS advertisers are able to apply our fraud pre-bid avoidance segment to their campaigns within their DSPs. Our DSP partners decide whether or not to bid on a given bid request based on their engagement framework with IAS. Our DSP partners share information from the bid request such as URL, app bundle ID, and/or IP address. Based on our proprietary in-house fraud detection algorithms, we let the DSP know whether or not we would consider the impression to be invalid traffic or “IVT” (i.e., the impression is non-human).
Some of our DSP partners prefer not to share IP addresses. For these partners, we allow the DSP to check the IP address from the bid request against our list of known IVT IP addresses, and, if the IP address is on the list, advertisers can choose not to bid on the impression.
In either case, our pre-bid integrations benefit from the large scale of data we process post-bid and allow us to inform DSPs of bad IP addresses or sites.
Our DSP partners can leverage the IAB/ABC International Spiders & Bots List to detect and filter non-human traffic based on user agent pre-bid. IAS uses this information in post-bid processing.
Publishers can also prevent direct campaigns from displaying on fraudulent IPs by leveraging IAS’s Publisher Optimization product, which allows them to target ads away from impressions labeled as IVT.
Post-Bid (prior to ad render)
Our JavaScript tag is enabled prior to the ad rendering on the page. The tag inspects the page for URL, IP address, user agent, and many more signals not available pre-bid. We then assess if the impression is fraudulent or demonstrates any potential profile of ad fraud. If we determine the impression is fraudulent, we block the ad from rendering or report on the presence of IVT, depending on advertisers’ settings.
Publishers also use our measurement so they can see the same data we report to our buy-side customers. The difference with our primary publisher tag, however, is that, by design, it does not block ads from rendering during post-bid processing. When publisher monitoring tags are in use, we label impressions as IVT where appropriate so publishers and advertisers can quantify IVT rates, but the creatives are still rendered. Publishers can use our previously mentioned Publisher Optimization service to target ads away from IPs labeled as IVT.
Post-Bid (post ad render)
We have developed sophisticated machine learning models that continuously learn across different environments of operation (e.g., open web, mobile, CTV). These models analyze hundreds of signals to detect ad fraud, thereby providing us proactive fraud detection and mitigation capabilities.
Our extensive machine learning modeling capabilities allow us to detect IVT by analyzing data and activity patterns over time to delineate human behavior vs. non-human behavior mimicking human behavior. Detecting IVT requires analyzing data over time to identify patterns of activity not associated with real people (e.g., devices running 24×7; devices generating excessive volumes of ad requests). Most data that our technology processes comes post-bid via macros and our Firewall, which processes signals from multiple HTTP headers.
In addition, our Threat Lab, a team of reconnaissance engineers, analyzes the data output by our models to identify emerging fraud schemes. The team researches schemes to determine how they are working and then collaborates with our platform partners to combat those schemes. The recent Vapor Threat scheme identified by our Threat Lab is a prime example of this collaboration, which resulted in the removal of more than a hundred malicious apps from the Google Play Store.
The Ad Fraud Detection Flywheel
Our AI-powered approach identifies fraudulent sites and apps by looking at a multitude of signals, including IP addresses, and this information is used to power our pre-bid avoidance technology. We detect fraud at three different stages. Everything we do is connected in a continuous “flywheel” effect. This holistic approach allows us to deliver the industry’s most robust fraud detection and avoidance solution to our clients. We work collaboratively and closely with industry participants to reduce fraud and the proliferation of bad actors by applying our solutions across the 280 billion digital interactions we see daily, and continuously evaluate and innovate our offerings to respond to today’s rapidly changing fraud threats.
