CAAF Spotlight: AdColony takes a different approach to fraud prevention
As CAAF goes from strength to strength, we intend to shine a spotlight on the members of the coalition - and in this new edition of our series, we spoke to Eric Dickinger, VP of Growth at AdColony, to find out more about their views on ad fraud, and their membership of Adjust’s Coalition Against Ad Fraud. Below you'll find the full interview.
What’s the single biggest achievement AdColony has made in fighting fraud?
We started seeing fraud ramping near the end of 2016 and into 2017. At the time there were simply no “out of the box” or “off the shelf” solutions that matched our expectations. We initiated a project with a few of our data scientists to build new algorithms that would detect anomalous behavior across the network.
It took quite a bit of iterating throughout Q1 & Q2 2017 coupled with feedback from our sales teams and clients but by the end of Q2, fraud on AdColony had peaked and begun to retreat. By Q4, our fraud levels (as determined both by our internal models and by client feedback) had fallen to very low levels, similar to early 2016 and before.
Does AdColony conduct its own fraud research? Have there been any big findings?
We’ve taken a bit of a different approach than the market.
Whereas most of the literature (including the great work from Adjust) has focused on specific types of fraud or the intent of fraud, we have instead focused on the faster identification, automated blocking, and prevention regardless of type.
Though some fraud may be considered sophisticated, much of it is quite basic - we’ve found that fraudsters do minimal work to drive the intended result, expecting that often it will go undetected. It’s a logical approach from their perspective - why spend significant time building complexity way beyond what network systems will ever work to detect?
How does AdColony’s tech side take on mobile ad fraud?
As described above, we use machine learning models to identify odd behavior using a few dozen vectors. We’ve built data tools on top of these systems so that all of AdColony’s commercial teams can easily see and act on the data for their campaigns or apps.
How does AdColony teach advertisers about the dangers of fraud?
In our experience advertisers have generally been well-schooled on the dangers. The fact that so many advertisers use sophisticated approaches to running their campaigns means they are very sensitive to increases in fraud (expressed typically as drops in ROAS).
The one area we have focused on however is that fraud and low-quality overlap but are not synonymous. Some inventory may provide fewer paying users, but that does not (necessarily) mean that the app itself is fraudulent.
Is there a single big takeaway that you think advertisers should be aware of when it comes to ad fraud?
It’s pretty easy to concentrate too heavily on sources of poor traffic, rather than viewing media buying holistically.
Finding fraud on a network does not mean that said network is fraudulent, and it shouldn’t fundamentally alter one’s approach to buying with that network. An advertiser raising concerns about fraud that results in no change from the partner is in many ways more problematic than the fraud itself.
Why did AdColony want to join the Coalition Against Ad Fraud?
AdColony has historically played an active role in establishing new industry standards, and fraud is no different. Adjust is a great partner of ours, and the CAAF was an obvious vehicle to collectively improve industry approaches to fraud.
What's been the best thing about joining CAAF?
There are great minds across our industry, and many of them are part of CAAF. The group has been a great source of sometimes opposing perspectives that is necessary for establishing strong, actionable standards.