Digital advertisers spend as much as 20% of their ad budget selling to penniless robots. That's according to ad verification company Adloox whose research indicates that fraud is on the rise and worsening.
The newest breed of bots armed with machine learning and natural language processing (NLP) are dissolving the final barriers to fraud-prevention. They're so adept at mimicking human behavior - from cursor patterns to misspelled words -that advertisers can't distinguish the two. And while these bots haven't yet cracked the Turing Test completely, a test of a machine's ability to exhibit intelligent behavior to that of a human's, it is an arms race that advertisers can't afford to lose.
Digital advertisers need to better differentiate the traffic to understand true user behaviors and optimize their ad spend. According to many, including brands like Experian and Visa, brands should begin measuring new indicators of human behavior that can't be faked. But alas, those are in short supply these days.
Will biometrics be enough?
Upon first glance, biometrics seem promising. Fingerprint identification and facial recognition could verify mobile ad recipients and newer smartphones include these features. Yet, it's unclear whether the technology or consumer tastes are quite there yet.
While Android allows third-party developers to tap its fingerprint API, reports that its facial recognition technology can be faked with a simple printed picture have emerged. And while Apple's FaceID technology is much more robust and there are indications that developers will be able to access some of its biometric data, history suggests that it will not be available to advertising providers. To date, Apple still prevents advertisers from gathering much data from even its iTunes app store.
Even if these technologies and policies were to suddenly change, there is something that won't: the nature of digital signals. While a biometric system can provide an external indicator of human behavior, once transmitted, it becomes a digital signal within a digital ecosystem. It is possible that eventually, fraudsters with machine learning might replicate these as well.
So what can advertisers do to sharpen their fraud detection?
Develop a holistic view of real world behaviors
To measure human behavior effectively, you must use external signals that are generated by an actual living, breathing person. All the better if those signals must pass through several networks and stages of verification which make fraud cost ineffective. One solution might lie in marrying several different types of data.
For example, because brands can now quantify and measure real world interactions just as they do online ones thanks to new metrics such as Pay Per Visit (PPV) and Cost Per Visit (CPV), advertisers might look to in-store visits as a method of verification. If advertisers can merge customer location data with offline data such an in-store purchases, they can demonstrate that a consumer traveled to a store and can retroactively validate the ad impression that was served to them.
This location-based verification is a type of two-factor authentication because location data comes from a separate, siloed partner. The complexity of faking both signals makes fraud far more difficult to pull off and potentially unprofitable.
Although fraudsters continue to improve their capabilities at rapid rates, they will never be able to truly, fully model real human behaviors. Aside from collecting data that would be impossible for a bot to generate, advertisers must embrace deeper analytics to better understand how humans act both online and on the ground. In the future, this will increasingly mean embracing data transparency and holding honest discussions with partners—and even competitors—about trends advertisers are seeing in behavior.
In the arms race that is ad-fraud, sharing and collating multiple sources of data, especially data that could only be generated by real people, such as location data, just might help brands put up a real fight.
Bhishma Savdharia is director of business development for GroundTruth’s Brand Safety Initiatives