For the sake of this article, we have two ways for attempting to answer the question about the total size of the ad fraud market. One is by looking at the numbers already available through secondary research sources, and the the other is to use reasoning. First I will focus on looking at numbers from secondary sources.
The Numbers
- Telemetry says “more than 50% of ad views are by bots” [1]
- Solve Media says “more than 50% of ad traffic fraud” [2]
- Oxford University says “Between 88% and 98% of all ad clicks were by bots”[3]
- Science Channel Veritasium says “Over 75% of Facebook likes fraudulent”[4]
- Incapsula says “61.5% of all traffic is bots”[5]
Fraud reduction company Fraudlogix promises on its website that “demand side partners are guaranteed a 30% improvement in their core success metrics, or their money back!”[6]. This would suggest that Fraudlogix is confident with its ability to remove at least a 30% level of online ad fraud.
In their research, both Oxford and Veritasium find fraud levels at way over the 50% mark, stated by vendors Telemetry and Solve Media. The question is why the recent ANA report finds fraud at roughly 20%? This is explained by the methodology they’ve chosen to use, limiting the focus to one kind of fraud only.
Given that it’s hard to find evidence to suggest that online advertising fraud levels would be less than 20% of total ad spend, so it seems fair to expect that the level is 20% or more.
Marketing is rapidly becoming a trillion dollar industry. Depending on the attribution model in use, it roughly contributes to 20%-30% of all consumer spending, which in turn represents roughly 1/8 of the global GPD. In other words, the media efficiency alone is likely to be connected with roughly 10% of the global GDP.
This way we can see that the health of the marketing industry is intimately connected with the health of the global economy. Due to this, media technology should be seen as critical infrastructure and there should be far greater focus on information security by the vendors.
The aspect that gets talked about less in various papers and commentaries, is the (actual) economic impact of ad fraud. In low involvement categories (e.g. candy) advertising could contribute 50% towards total sales, and in high involvement categories (insurance) it could still contribute 5% towards total sales. So is it fair if we say that roughly 20% of all sales is because of advertising?
Using US as an example, where roughly 70% of the GDP is consumer spending, this comes to roughly US$12 trillion in commerce per annum. If 20% of that is coming from advertising, and advertising suffers from 20% overall fraud level, then advertising fraud becomes a US$480 billion dollar problem.
I realise the many caveats within these number, but still want to use them to illustrate the point.
We can summarise this section in two points:
- The direct cost of online advertising fraud could be higher than anticipated
- The indirect cost of online advertising fraud could be far greater than the direct cost
The Reasoning
The fact that we have 5 research sources saying that the number is very high, does not confirm to us that the number indeed is high.
The question is, how much have we done about online advertising fraud as an industry? The answer is not that much. Because of this it seems likely that the problem is more sophisticated than our understanding of the problem. In which case it is likely that it would be larger than we think it is.
The nature of “cops and robbers” type of games is that there are always robbers looking for ways to do things in ways the cops don’t know about yet. When the robbers have roamed free for years, it’s possible that there is a lot the cops don’t know. The smarter and better resourced the cops are, the harder it gets for the robbers to do this.
When this happens in a neighbourhood, the robbers move to the next one. This means that it’s relatively simple to move fraud around. When this happens in the whole city, the robbers either move away or find something entirely different to do. This means that unity is needed to remove fraud.
For this reason the methodologies we see being used for detecting and understanding the online ad fraud problem are likely to be better at detecting one kind of fraud over the other. This is a key point as there are hundreds of known play-books for executing online ad fraud models, many of which are widely available for anyone interested in the “black hat” forums. This means that because there are many kinds of online ad fraud, many different methods are needed for eradicating it.
The first thing is to share openly between the various industry stakeholders. Without shared resources and shared practices, it is not possible to reduce fraud substantially. While it’s easy to move a problem like fraud around, it’s very hard to make it go away.
Due to the way fraud reduction solutions typically work, solving one problem better than another, the solutions will have a tendency of leading to many false positives.
The next question is on the way the fraud reduction solutions are built. If they are more likely to have false-negatives more than false-positives, this would lead to over-reporting. If they are more likely to have false-positives, and less likely to have false-negatives, then this would lead to under-reporting. Having built these solutions myself, I’ve learn it’s harder to sell solutions that produce false-negatives (less available inventory) than it is to sell solutions that produce false-positives.
If the case is indeed that fraud reduction solutions tend to have a false-positive problem, then 5 methods saying “it’s 20%” means actually that there is much more. When each of the technologies find a result at 20% level, it means they’ve all missed some.
The other question about the vendors is if they are all doing exactly the same thing, or a different thing. If they are doing 100% the same thing, then they are finding the same 20% and making the same false-positive mistakes, then the result is the same. Vendors and researchers actually use propriety solutions with only limited transparency to the underlying methodology.
This means that they are all missing something, and they’re all also seeing things that the others are not seeing. This way 5 saying 20% becomes much higher than 20% just by us knowing the factors discussed above.
Conclusion
Over the past 10 years or so, having researched this topic extensively first hand, I’ve always thought that it would be a big problem and most likely more than 50% of the overall media spend.
In the light of the above reasoning and numbers, it seems safe to assume that the actual level of fraud is 50% or more of the total investment.
As we already discussed, why ANA reports it at 20% in its recent paper with White Ops is explained by the fact that the method focuses on one type of fraud activity only.
References used in this article:
[1] Ad Fraud Has a Chicken Little Problem
http://digiday.com/publishers/ad-frauds-chicken-little-problem/
[2] More than half of U.S. traffic, per Solve Media
http://www.adweek.com/news/advertising-branding/bot-problem-keeps-getting-worse-154585
[3] Quantifying Online Advertising Fraud
http://oxford-biochron.com/downloads/OxfordBioChron_Quantifying-Online-Advertising-Fraud_Report.pdf
[4] Veritasium on Facebook Ad Fraud
https://www.youtube.com/watch?v=oVfHeWTKjag
[5] Incapsula. Report: Bot traffic is up to 61.5% of all website traffic
http://www.incapsula.com/blog/bot-traffic-report-2013.html
[6] Online Advertising Fraud Solutions
http://lp.fraudlogix.com/