The fraud rates for online ads are scary, with advertisers losing billions of dollars each year.
Last week, Pixalate released a report showing that 35 percent of U.S. desktop ad impressions sold through programmatic advertising networks are fraudulent. Programmatic ads are those purchased through online networks, often through automatic bidding systems, instead of directly from individual publishers.
U.S. advertisers spent more $25 billion on programmatic online ads last year, meaning that about $8 billion is lost to fraud.
The problem is that the advertising networks that connect advertisers and publishers don't have adequate screening in place to ensure that the advertisers are legitimate, that the publishers are legitimate, and that the ads are seen by real people and not bots.
But that's starting to change.
Take, for example, Xertive Media. Two years ago, the company was rated as one of the three worst ad networks when it comes to fraud, based on a Pixalate study of more than 400 networks.
Now, it's in the top 30.
The company turned things around by investing in technology to screen its advertisers, its publishers, and its ad viewers.
"About a year and half ago, we took a financial decision to invest in this," said Raeut Ben-Asher, VP of sales at Tel-Aviv, Israel-based Xertive Media Inc.
The company decided to use technology from two vendors, Forensiq and GeoEdge.
Forensic scans the publisher websites to ensure that the websites and views are legitimate. GeoEdge checks that the advertisers are on the up-and-up and that the ads themselves don't have malicious code or spammy redirects.
"We are covered on both sides," she said.
Both vendors also check that nothing changes after the ad has been placed, to catch crooks who look legitimate at first, but then switch over to fraud after the ad is already running.
"If you change the content on your site, we will know that it has changed," she said. Xertive's contracts with publishers require websites to give a heads-up if the content changes -- say, if it changes its editorial focus to, say, reviews of cocktails.
Alcohol is one of the screens that the company uses, allowing advertisers to avoid websites with types of content that may be incompatible with their brands.
In addition to checking that the publishers are authentic, Forensiq also looks at the users who view the ads.
"We place a JavaScript tag and capture different device characteristics from the browser and make sure that the traffic is being shown to a human, not to some botnet," said Antoni Kolev, VP of product at New York-based Forensiq LLC.
For example, Forensiq checks if the viewer's IP address belongs to a residential Internet provider or to a data center.
The fraudsters do try to find new ways to imitate human users, he said, but having a big picture view can help spot patterns of suspicious activity.
"We can look at the entire ecosystem and analyze traffic from billions of requests and impressions daily," he said. "For bots to be profitable they need to generate a lot of activity, and in general, it becomes hard for them to hide."
Clients, who include advertisers, advertising networks, and even the publishers themselves, get notified about problems.
Advertisers can then see which publishing platforms and ad networks have high fraud rates, and switch their ad buys, he said.
Similarly, Xertive is selective about the advertisers it works with. For example, it no longer accepts performance-based ads that are popular with smaller -- and riskier -- advertisers. Today, its advertisers buy ads based on the total number of views, a more expensive type of advertising that is typically used by larger brands.
Xertive also screens the actual ads themselves.
"GeoEdge blocks the ad from uploading if there's malware in the chain somewhere," said Xertive's Ben-Asher. "No drive-by downloads, no redirects to a malware page."
GeoEdge will also catch it if a legitimate ad image is surreptitiously replaced by spam or other ad that wasn't what was originally agreed to.
More than that, it can catch the new tricks that fraudsters use to try to get malicious ads to viewers.
For example, some criminals will target only a very specific group of users, said Tobias Silber, at New York-based GeoEdge Ltd.
"We mimic user characters to catch dynamic threats, to catch all the different threats that target very specific users," he said. "We mimic user characteristics -- we will check form different IPs, devices, and even carriers."
The system also checks if ads load slowly or create other problems for the websites where they run.
"A lot of platforms are taking this very seriously. making sure that only clean ads are going through," he said. "They do recognize the problem."
So far, the investment in the screening technology has paid off for Xertive.
"You never know what happens next week," said Ben-Asher. "Nothing is perfect. But we know that it works right now because we don't have complaints."
And the work has paid off.
"It's great to to be in the top 30, because we are a small company, and we're competing with giants," she said. "It's nice, it feels good."
This shows that advertisers don't have to assume that fraud is just a normal part of online advertising.