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.

Read the full article on TheDrum.