Ad tech company Kubient has sought to transform the digital advertising industry into audience-based marketing, making advertising more transparent while mitigating ad fraud.
The company in December received a patent for its proprietary ad fraud identification and prevention technology, Kubient Artificial Intelligence (KAI). According to Kubient CEO and Founder Paul Roberts, machine learning and AI is the main way to identify fraud on a large scale.
Companies are already using AI and ML for that purpose, but the issue lies with how they’re using it.
“A programmatic auction happens within a 300-millisecond window of time,” Roberts told Fierce Video in an interview. “If you work with [DoubleVerify], IAS, White Ops (now known as HUMAN)…these companies are using machine learning after you spend your dollar. Then they give you a block list, or a known bot list, to basically load into your SSP, DSP, ad exchange, wherever you want.”
Fraudsters, however, are well aware of how that works, as they create botnets to ensure they’re not on those lists. Essentially, the bots mimic an individual’s digital fingerprint and can quickly switch from one fingerprint to another, Roberts explained, “from the NFL dad to the millennial mom, to the world traveler.”
“The challenge we realized was that nobody was actually identifying fraud, they were basically figuring out after the fact that fraud happens, and then they give you this list that doesn’t really work,” he said.
What KAI does differently is it uses machine learning within that 300-millisecond window where media is bought and sold. Thus, it’s able to detect fraud in real time.
On any given day, “23% of impressions we [see], we don’t think they’re real,” Roberts stated.
Pitfalls of CTV
As for CTV, he went on to say fraud criminals try to game the system through two main methods. They either create false users who keep the devices on even when a person isn’t watching (DV coined this as the “TV off” issue), or they conduct app spoofing.
“You have criminals that are creating bots that can keep these apps on 24 hours a day, constantly engaging, constantly pausing,” said Roberts. Whereas app spoofing tricks advertisers into buying a spot on a fraudulent version of a popular app.
“The big joke we have is, ‘how many different versions of Roku do you want to advertise on – the real one or the 15 other ones we see?’” he said.
These spoofed apps populate the ad exchanges, including “some that don’t even sell into programmatic exchanges.” For the most part, advertisers are using machines to decide what to bid on and which prices to bid. Roberts pointed out many companies aren’t very stringent on what they allow into their networks, which fraudulent apps can readily bypass.
“It’s not inflammatory content, it’s not adult content…it’s safe in their eyes,” he said. “It passes the sniff test of the filters they have, but the reality is that nobody checked it to allow it into the market in the first place.”
CTV is understandably a hotbed for fraud activity, as it’s a relatively new technology. Roberts explained fraudsters don’t want to waste time creating botnets for desktop video at a $5 CPM, “when they can do it in CTV for a $20-30 CPM.”
“I watched this happen with the rise of mobile, where everybody rushed into mobile, CPMs skyrocketed and the criminals followed,” he said. “Interestingly enough, desktop is actually becoming a safer place to advertise, because a lot of the criminals right now are so focused on CTV.”
IAB has developed a technical standard, called ads.txt, which allows media owners to create a public record of Authorized Digital Sellers, thus reducing the risk of spoofing and other fraud methods. The framework was originally developed for web-based transactions, but IAB since then came out with app-ads.txt to help combat fraud for mobile and CTV environments.
Although that updated framework works for CTV, implementing it can be tricky, as the CTV landscape can have multiple entities that share inventory. Pixalate has noted app spoofing rates on mobile devices are typically lower than rates on CTV, due to more difficult end-to-end implementation of app-ads.txt.
“So I think there’s going to be something similar coming along for CTV, I’ve heard folks on the IAB side talking about it,” Roberts said. “I think that will help tremendously.”
Lack of education
The prevalence of ad fraud also comes down to companies being unaware of how exactly fraud works, Roberts noted.
“We’ve sat in the offices of some of the largest DSPs in the industry and they all tell us the same thing, ‘we basically pick ad fraud vendors that are approved by the MRC, what they do – you have to pay them – we’re not really sure,” he said, adding those companies “don’t want to see 20% of their revenue go out the door.”
Basically, the DSPs and SSPs don’t want to be involved in the fraud prevention process, but that comes at the cost of advertisers losing revenue. Furthermore, advertisers often have a “surface level” understanding of the issue.
“If you sat and talked to most media buyers and said, ‘explain in detail how digital ad fraud happens,’ they’ll just say ‘it’s criminals,’” Roberts said. “We’ve sat with media teams that control a quarter of a billion dollars in budget a year, and they kind of scratch their heads when we started explaining how fraud actually works.”
Roberts went on to say he has spent significant time on the dark web buying fraudulent traffic to see how fraudsters operate. What he’s found is they are creating bots “that can sniff a page before it loads, whether it be mobile, desktop or an app, because most brands don’t scan 100% of their inventory.”
So, the bots can detect if DV, WideOps or any other ad fraud protection vendor has code on the page. If they do, the bots “only send 5% of bot traffic there, so they stay right within the industry parameters.”
“The other impressions that are not being scanned, not being looked at – it’s a feeding frenzy,” he said.