With Fraudsters Now Faking Viewability and Video Completions, Advertisers Risk Being Gamed
Before joining programmatic advertising and audience targeting firm Dstillery as chief scientist, Claudia Perlich focused on machine learning at IBM Watson. That highly analytical and data-driven experience has proven valuable in helping Dstillery combat fraud. Perlich spoke with eMarketer’s Lauren Fisher about the types of fraud pervading the advertising space, and why ultimately attributing all the way down to a sale will be key to avoiding it.
eMarketer: What types of fraudulent activity are you seeing?
Claudia Perlich: There are still a lot of bots posing as people in the programmatic environment that marketers should be keenly aware of. We have seen a lot of improvements on measurement with companies like Integral Ad Science and White Ops measuring what they call “blatant traffic fraud” that’s being committed by bots, not humans.
But what I find really interesting is how good these dark entities are at faking almost any metric marketers use to allocate money. And that’s also true for things like viewability. As long as we have some weakness in our measurements, these things are easily gamed. So I caution against jumping on viewability as a solution to bot traffic, because that is and will be gamed just as much as any other metric.
“I caution against jumping on viewability as a solution to bot traffic, because that is and will be gamed just as much as any other metric.”
eMarketer: What can advertisers do?
Perlich: What advertisers should have is a healthy skepticism against too-good performance and a desire to run a lot more experimentation where you measure true impact instead of metrics like viewability or clicks that don’t typically get to the question of whether you truly influenced people’s behavior.
eMarketer: Are you already seeing evidence of fraud gaming viewability and other metrics? What does that look like?
Perlich: Yes, and I want to make it clear: This is a hard problem to solve. The reason we’re having conversations about fraud year after year is that it’s really hard to do ad measurement right. There’s no silver bullet, so I can relate to this on a scientific level.
There are inventory providers who are clearly making their inventory appealing by artificially inflating their viewability numbers. And I can’t shake the feeling that something is wrong with that. You cannot predict whether a person will really scroll down. You will always have a certain kind of post-impression assessment. But we see publishers with 100% viewability. We also see publishers with a 100% clickthrough rate.
eMarketer: What else are you seeing?
Perlich: Something that’s prevalent and is a big problem in video is video completion. Measures of bot traffic that are the highest we’ve seen have been on video complete pixels, because there’s a lot of money in video advertising. And the moment video complete became a metric, you saw bots being the only ones who really had the patience to sit through all the ads.
“The reason we’re having conversations about fraud year after year is that it’s really hard to do ad measurement right. There’s no silver bullet.”
eMarketer: The Media Rating Council just came out with an addendum in which they broke fraud or invalid traffic into two distinct categories: general invalid traffic and sophisticated invalid traffic. Most vendors filter for general invalid traffic, but do you feel like there’s a lot of sophisticated invalid traffic out there?
Perlich: My sense is what we are looking at is almost the same as email spam. This will continue to be a weapons race. As long as the economics of fraud still work, meaning you can make money with it, we will continue to see the development of more sophisticated tactics.
What makes it even harder is that the industry constantly switches metrics. So if we see fraud attacking clicks, we move onto viewability. But if you look at reports from different viewability vendors, you can get five different measures, which tells me that we aren’t very good at measuring it. This is now an opening for the next generation of bot traffic to figure out how to game one of those vendor’s specific implementations of measurement.
So as we constantly evolve to new things that we don’t really know how to measure, the bots are good at exploiting those new weaknesses that don’t understand yet.
eMarketer: For that reason, you hear a lot about people talking about not using these types of metrics and moving, as you suggested earlier, toward some more meaningful measure of actual impact. Offline sales would be an example.
Perlich: We have seen bots fill out forms and schedule test drives, so closing the loop is really the only true way to go. But the reality is that you’re still looking at a set of companies, like luxury cars, where nothing is going to be solved because so few are being sold every day that you simply never have enough information to make a judgement call or optimization.
But if there is a high rate of purchase, like for CPGs [consumer packaged goods], you can do some insightful work. But it is limited to scenarios where you can link it and you have enough purchase activity to have statistical solidarity.