In late October of 2015, a major insurance company ran a test campaign using two different viewability vendors: Integral Ad Science, a holistic media quality provider, and Vendor B, a publisher-focused point solution. The campaign ran across nearly 400,000 display impressions in the U.S. through a well-known agency trading desk. At first glance, there appeared to be a mistake. Vendor B reported back promising results: 54.4% viewability, 26% higher than the average*. However, Integral measured a different story: only 1 in 10 ads was in view. How could one vendor report 54.4% and another 10.3%? The difference is the result of different methods of fraud detection. For an ad to be viewable, it must be shown to an actual person. What’s the value of a viewable ad – for display, at least 50% of the ad in view for at least one second – if it’s ‘in view’ to a robot? Vendor B detected nearly 20,000 fraudulent impressions – 5% of the total. But Integral flagged more than eight times that – 43% – as fraudulent. If the insurance company had gone solely with Vendor B, they would have purchased more than 140,000 impressions that they thought were viewable, but were really served to robots.
Why was there such a large discrepancy in fraud numbers? How did fraudsters defraud a vendor purporting to cover fraud?
There are a few key reasons:
1. Scale & Big Data
Integral’s scale allows us to rate historical URL performance before an ad is ever served. For example, 149 sites in the campaign had fraud rates of greater than 80%. Half of those had historical fraud scores that were immediately flagged for fraud based on historical tracking. Integral’s team was able to validate suspicious actions with historical analysis, confirming that the sites were rife with fraud.
Without historical data, every user is new, and cannot be validated or flagged with previously collected information.
2. Real-time signals
Integral analyzes multiple real-time signals to detect fraud such as mismatched locations, softwar versions, and operating systems. But there are also more nuanced signals that our fraud detection technology pays attention to – like the likelihood for domains to be spoofed. A spoofed domain occur when an advertiser pays to bid on a specific URL– like cnn.com or espn. com – but instead has her impression served on a different fraudulent or low quality site.
Spoofed domains are frequently combined with bot activity. Why else would the ad be served to a different domain than what was sent to bid? Without looking at the domain in real-time and pairing with historical likelihood for the domain to be spoofed, you cannot detect what’s human and what’s not.
When it comes to fraud detection, a vendor that simply says they “do fraud” should not be enough to convince you. Serious fraud detection takes significant resources, both in talent and finances, to build a world-class solution and eliminate waste. When the insurance company saw the viewability discrepancy, they were shocked that vendors could report such different numbers, but now understand why. They now use Integral exclusively for fraud protection.
Fraud prevention is necessary to buying digital media. When deciding which partner to use, it’s important to remember that not all fraud solutions are equal.