By: Eva Keller
In 2025, marketers spent $750 billion on digital advertising. An estimated $165 billion of it was stolen.
Not misdirected. Not wasted on low-intent audiences. Stolen, by bots, malware, and human fraud farms designed to consume ad budgets without ever exposing a single ad to a real person.
That number comes from Anura Solutions, an ad fraud detection company that has analyzed billions of traffic events across more than two million domains. But the more consequential problem isn’t the money. It’s what happens to the data that’s left behind.
“You’re making good decisions based on bad data,” says Rich Kahn, CEO and co-founder of Anura. “That’s how it gets out of control.”
Rich has spent more than two decades studying how digital ad fraud operates, first inside his own ad network, eZanga, where he built fraud detection tools in-house after discovering in 2004 that no commercial solutions existed, and later through Anura, the standalone fraud detection company he and his wife Beth launched in 2017. The core insight that drove both businesses is one the broader marketing industry has been slow to absorb: fraudulent traffic doesn’t just drain budgets. It actively corrupts campaign optimization.
How Does Fraudulent Traffic Corrupt Campaign Optimization?
Here’s how. When a fraudster’s bot clicks on a paid ad and lands on a website, it doesn’t simply sit there. Sophisticated bots know which signals platforms reward. They trigger time-on-site metrics. They engage with page elements. Some fill out lead generation forms with stolen personal data. In affiliate marketing environments, where average fraud rates run at 45%, according to Anura’s data, they mimic the conversion behavior that marketers use to decide which traffic sources to scale.
The marketing team sees the numbers and draws the only logical conclusion: this source is performing. Buy more of it.
“The marketing department is saying, these sources of traffic are driving lots of leads, let’s buy more of it,” Rich says. “And they start inching their way up until they realize, we’re buying bad leads.”
By that point, the campaign has been optimized around fraud. Budgets have been reallocated toward the worst-performing sources. The data model is broken at the foundation. And the signals that platforms use to improve targeting, the conversion pixels, the engagement metrics, the audience feedback loops, have been trained on phantom behavior.
This is the data poisoning problem. And it doesn’t announce itself.
Why AI Has Lowered the Barrier to Ad Fraud
The driver accelerating it is AI. The same tools that have lowered the barrier to building software have lowered the barrier to building fraud. A fraudster who once needed technical expertise to operate a bot network can now prompt an AI model to write the code. The attack surface hasn’t just grown. It’s been democratized.
“The barrier to entry is lessened because you don’t need to be technical,” Rich says. “You’ve got people right now that can write apps that don’t have to run a single line of code.”
In January 2026, Anura identified a form of AI-assisted invalid traffic built to exploit weaknesses in JavaScript-based fraud detection, an attack that had slipped past multiple detection vendors before it was isolated. The episode illustrates the asymmetry at the center of this problem. Fraudsters iterate fast. Most marketers aren’t watching closely enough to know when something has changed.
Why Platform-Level Fraud Controls Have Limits
The standard industry response, relying on the platforms themselves to clean the traffic, has structural limits. Google maintains a large fraud detection team. Facebook works to remove fake profiles. What neither can offer is visibility beyond its own inventory, and fraud rarely stays inside a single channel. Anura monitors traffic across channels simultaneously, a vantage point designed to surface emerging attack patterns while they are still small.
“We have such a vast amount of traffic that when new types of fraud attacks happen, they start really small, and we can almost see what they’re thinking as they’re building them out,” Rich says.
What Anura Looks For in Traffic Data
Anura analyzes more than 800 data points per visitor to distinguish fraudulent from legitimate traffic. The signals aren’t what most marketers would guess. Real users connect from real devices with standard browser configurations. Fraudsters connect through a number of ways, some of which could include residential proxy networks, spoofed device signatures, and data center IPs. Rather than assigning a probability score to each visitor, the platform returns a definitive determination. The point, Rich says, is not just removing fraudulent clicks. It’s cleaner data from which to make real decisions.
Anura works primarily with organizations spending $50,000 or more per month across digital channels, where invalid traffic is spread across enough sources to be difficult to spot manually. At that scale, the first thing a detection layer changes isn’t the media plan. It’s the quality of the data the media plan is being managed against.
Anura is one of only six companies currently certified by TAG (Trustworthy Accountability Group) for meeting its fraud detection standards, out of approximately 100 operating in the market.
The company also runs a traffic quality audit for advertisers assessing their own exposure. It covers a two-week window and reports both fraud volume and source, which gives marketers a view of which channels are delivering real audiences and which are delivering numbers that look good until someone calls the leads.
Optimization is only ever as good as the traffic the data was built on. For most digital advertisers, that’s a problem worth measuring.
