What Are Major Challenges In Detecting Ad Fraud?
In the past few years, global mobile spending has been on a constant rise. With worldwide app store consumer spend reaching $120 billion in 2019, mobile ad spending saw a sharp increase, too. According to a report, global mobile ad spending reached a whopping $190 billion in 2019 – and is estimated to reach $280 billion by 2022.
This comes at no surprise whatsoever. The numbers mentioned are enough to lure in fraudsters and scammers to take advantage of this multi-billion dollar market. Advertisers and publishers will be negatively affected by this, even worse than before. A study predicted the total loss advertisers would face in 2019 as a result of fraudulent activities would be a staggering $42 billion – this is 21% higher than the $35 billion that was lost to global advertising fraud in 2018. According to another 2019 report by Interceptd, mobile ad fraud costs an average of $50 million per day to marketers and organizations.
Taking all these alarming figures into consideration – it is important to note that, on the other hand, there has also been substantial progress made in terms of ad fraud detection tools and methodologies. To understand this better, let us get caught up on the terminology and what it really means.
In this article/
- What is ad fraud?
- Mobile ad fraud and how it is done
- Ad fraud detection techniques
- Challenges faced in ad fraud detection
- Limitations of machine learning and AI
What is Ad Fraud?
In the simplest of terms, ad fraud is the act to generate revenue using fraudulent techniques in digital advertising. Usually, this is categorized into two broad terms – desktop and mobile ad fraud.
You can read our comprehensive ad fraud guide here.
Mobile Ad Fraud and How it is Done
Mobile ad fraud is when fraudsters and scammers drain advertising budgets of publishers, advertisers, and online businesses via organized and systematic methodologies. Fraudsters, in 2020 and beyond are converging their unique skills to attack the full marketing funnel. Why? Because the full marketing funnel represents their total addressable market. Mobile ad fraud can be carried out in a number of different ways, such as:
- Fake Impressions
- Click Injection
- Ad Stacking
- Cookie Stuffing
- Fake Installs
- Click Spam
- SDK Spoofing
- Bots & Emulators
- Device Farms
Ad Fraud Detection Techniques
In order to reduce the impact of digital ad fraud, marketers and companies have made noticeable advances in ad fraud detection over the past few years. Some smart ways to prevent ad fraud are:
- Have adequate ad verification tools
- Target audience on social media platforms
- Monitor and ban IPs
- Partner publisher trustworthiness
- Anti-fraud software and tools
- Use an ad fraud prevention tool that has machine learning capabilities
- Monitor relevant metrics
However, even after having one or more of the above practices implemented, companies are still losing tons of money to ad fraud. Why is that?
Challenges Faced in Ad Fraud Detection
Yes, there have been some positives as far as ad fraud detection technology is concerned. However, the industry as a whole still has a lot of development to do in terms of understanding the acceptance and replication of ad fraud principles across numerous channels and businesses.
Let us now have a look at some of the major challenges in detecting and preventing digital ad fraud:
Advanced Mechanisms Accessibility to both – Marketers and Fraudsters
One of the major reasons why mobile app fraud instances can never be truly controlled is because even though advancements in technology are good for marketers and ad fraud detection companies, fraudsters benefit from this as well.
With all the technology readily accessible to anyone these days, fraudsters and scammers have all the latest mechanisms and systems in place to bypass ad fraud detection tools and software. And make no mistake; they manage to do it pretty well. For example, programmatic is traditionally thought of as a ‘clean’ space for marketers and advertisers. However, our reports show how this channel is also subject to fraudulent activity.
Limitations in ads.txt Files – Programmatic
Back in 2017, ads.txt was launched by Interactive Advertising Bureau (IAB) – a text file on a publisher’s website that has all the authorized vendors listed that can sell their inventory. This enabled ad buyers to verify whether the inventory they were buying was legitimate or not.
The above figures demonstrate the fact that over three quarters in 2018, websites that had ads.txt implemented incurred a lower display ad fraud rate as compared to the ones that didn’t.
However, even though it does assist advertisers and publishers in reducing the impact of domain spoofing, one of the types of mobile ad frauds, it still does not necessarily reduce bot traffic.
In addition, there is a probability that ads.txt files might have errors on the publishers’ end. This will further reduce its benefit since the demand-side platforms (DSPs) will not be able to differentiate between authorized and unauthorized inventory accurately. According to a study, out of Alexa’s 1000 top-ranked publishers, 27% have inaccurate ads.txt files.
Can’t Really Differentiate Between a Real and a Fake Click
Despite all the upgraded software and methodologies used to detect fake clicks, there can never be a fool-proof mechanism to differentiate between a genuine and a fake click.
These fake clicks are done via click bots that are designed to imitate human behavior in real-time. In some cases, actual human beings are hired to make these clicks that do not follow a set pattern. Thus, whilst it might seem like a simple exercise, it actually requires sophisticated deterministic, and in some cases, probabilistic approaches to detect and block fake clicks.
Limitations of Machine Learning & AI
Over the past couple of years, the use of Machine Learning (ML) and Artificial Intelligence (AI) has seen some major improvements and upgrades in ad fraud protection. Research has shown that by the year 2022, digital platforms using AI & ML to penetrate specific markets will account for 74% of total online and mobile advertising expenditures.
However, there are limitations. For instance:
- Blacklists – this is one of the old-school techniques to block traffic using IP addresses. Over time, this has become relatively easier to bypass for fraudsters. They achieve this by changing their IP addresses.
- Rule-based Algorithms – whilst rule-based systems (we call them deterministic rules), are adequate to combat simple ad fraud techniques such as click injection, it is not enough to take on more sophisticated methods such as SDK spoofing. Thus, given that there is a wide range of fraud and fraudsters, it is best to opt for ad fraud prevention methods that have a diverse set of ad fraud detection and prevention tools.
- Requiring a data-history – machine learning requires a set of data, from which the system or software developed can learn from. Whilst, in most cases, the pay-off far exceeds the cost, this is not always a feasible approach for ad fraud, as ad fraud is constantly evolving. Thus, machine learning is often best used in conjunction with other forms of ad fraud detection.
Detection and prevention of digital ad fraud of any type will always remain a significant challenge for the people who are tirelessly working to find ways to achieve success. This is an ongoing cat-and-mouse game where marketers and ad fraud prevention companies will have to work side-by-side to constantly develop new tools to reduce the negative impact of these fraudulent activities. Marketers need to stay well informed and updated with the latest technologies, methodologies, and techniques that evolve in the digital marketing and advertising arena. On the other hand, publishers and advertisers need to opt for the best practices that can minimize, if not fully eliminate, the impact of ad fraud on their digital advertising.