Staying Ahead in the Mobile Ad Fraud Battle Zone
Advertising fraud is a game of cat and mouse – on desktop and now on mobile too. As soon as the fraudsters went after the significant ad budgets pouring into mobile, anti-measures were introduced. And then mobile-ad fraud evolved, which did not come without a counterattack of its own, and so on.
Faking Impressions & Clicks
At first it was easy with CPM and CPC pricing models. All that fraudsters had to do to put their hands on the money was to fake an impression or a click. In the app economy, the CPI model is most common. The significant source of money to be made quickly drew the attention of the bad guys who masterminded multiple tactics to mimic an install.
In the past year, the CPA model started to gain momentum because the app economy became hyper competitive, leading to high churn rates. As such, a loyal user rather than an install (new user) became the most important KPI to focus on. A CPA pricing model was a natural follow-up. Although more difficult to perpetrate (impersonating a click, install and in-app activity), in-app fraud is a real problem because its hefty payouts offer extra motivation to fraudsters to find ways to meet the technical challenge.
How To Combat App Fraud
The important thing to realize is that minimizing your exposure is definitely possible. An effective way to combat fraud involves a two-pronged approach: prevention and detection after the fact. Whether before or after, reducing exposure to fraud is centered on data and data transparency, the highest form of which is gaining access to raw data. With all the data, machine learning algorithms and humans can detect data anomalies in different forms and shapes.
App Fraud Prevention
For example, IP-related fraud (e.g. large number of clicks / installs / unique identifiers from the same IP, or different IP locations between the ad click and the install / first launch), pattern-related (e.g. click / install every 20 seconds, or players / users from a specific source always drop off at the exact same point in a game / app), device ID-related (e.g. different identifiers for the same device, or performance-related (e.g. sharp increase in install volume, a stark decline in day 1 retention).
Prevention is more of a challenge because it has to take place in real time and prevent an install from being attributed to a fraudulent source. One way to overcome the real time challenge involves the use of of what is known as deep learning (neural networks). These systems offer a newer class of highly accurate machine learning algorithms that train on hierarchical representations of input data.
In laymen’s terms, deep learning is about building a large set of simple functions that interact together to approximate a solution to a problem. A lot of the advancement in this field is around image processing (e.g. Microsoft Fetch which can identify dog breeds, Google deep dream etc.), but they can be applied in a wide variety of areas – fraud included. Models are trained on offline data, and then scoring based on a pre-calculated model is applied in real-time to enable fraud prevention. Obviously, the scale of data the engines train on to develop a model is a major factor in its effectiveness.
Fraud prevention is the responsibility of everyone in the ecosystem. Advertisers should run with reputable ad networks, demand transparency from these networks, enhance their BI with anti-fraud specialists, and work with measurement partners that can offer robust anti-fraud capabilities.
Fighting Mobile Ad Fraud in the Future
Networks also have an interest to be transparent and at the very least fraud-aware. It will help them land more budgets by being able to pinpoint that single fraudulent source, and shifting budgets to all the other legitimate ones. Ultimately, advertisers should only pay for ads that delivered real and measurable value. That’s the goal. Everyone stands to gain when this goal is reached.
Fraud is part of digital life. But with the proper measures in place and an industry-wide collaboration, we can ensure that in this game of cat and mouse, Jerry will always be one step ahead of Tom.
I’d like to thank everyone who took part in our “Talking Fraud” blog series: Pepe Agell from Chartboost, who represented the network’s perspective; Patrick Witham from Product Madness, offering the advertiser’s POV; and Johnny Thwaites from Forensiq, offering the fraud specialist angle.