Mobile Antifraud Overview
According to the AppLift study “Fighting Mobile Fraud in the Programmatic era” , the share of fake mobile traffic is about 34% of the total traffic, if expressed in money, this is more than $ 4.5 billion in losses. Advertisers are looking for ways to evaluate the quality of KPI traffic, try to filter IP and compile black and white lists, reluctantly give offers to the network - however, clever bot-guides bypass these obstacles, continuing to pour garbage traffic, taking profit from the advertiser, deceiving grids and buying even more power for their bots and farms.
Despite all the ways to fight fraud on its own, this is a losing strategy. You just spend a lot of resources trying to reinvent the wheel and maybe save a small amount of money by discovering the most lazy and stupid froder. However, those who are smarter will continue to quietly divert profits from under your nose. Therefore, the solution to the problem is an outsourcing, namely, the connection of the fraud protection system.
')
In Mobio, we work hard and hard at clearing traffic, testing different systems, which we want to talk about under the cat.
Imagine that you need to analyze a huge number of conversions by parameters such as: ip, user-agent, device-info, click time, conversion time, proxy, etc. A good anti-fraud system automates this process and “rejects” suspicious conversions (for example, finding that the installation was done using a proxy or VPN). A good system will analyze not only each specific conversion, but also compare them all in a general slice, in order to establish any dependencies characteristic of froder.
After studying the recommendations of colleagues and Google search results for the query: “mobile antifraud system”, I selected several candidates for testing:
I will immediately say about those systems that had to be excluded:
Kraken is a Russian development based on the fact that the source of the conversion is checked for any of the well-known bot and insist farms. The exact algorithm is not disclosed. The cost is low, however, an adequate test did not work out - the system only checks selectively and on certain days of the week. Support comment: “There is no other option for work, the system was not created for this purpose from the very beginning.”
FraudLogix is one of the solutions that are suitable for an ad network, they only have an IP blacklist. In fact, the analysis is performed only by one criterion, which is not enough for a high-quality anti-fraud. In addition, the test scan marked with frod's absolutely normal addresses included in the list for checking false positives. The result is disappointing - low accuracy and just one criterion for analysis is clearly not the option that should be used to protect against fraud.
Kount - I can only say one thing about this system: a prepayment for a year in the amount of several hundred thousand dollars will not suit anyone. We, for example, did not fit.
So, there are only three systems that we could use. I will tell about each of them in more detail:
Forensiq is the loudest name on the list. It is often referred to in articles on anti-fraud. Perhaps my assessment will not be entirely correct, since I used the version integrated into the tracking system Affise , but what I saw weakly pulls on "the best fraud protection system." The report comes in the form of a mark in front of each conversion indicating the level of risk: low, medium, high, and to the listing of suspicious criteria. The inability to somehow track the scoring algorithm and communicate with the representative of the system reduces the final score. I didn’t have any complaints about the quality of the assessment - there were no false positives, Forensiq correctly determined froders. By cons can be attributed to the relatively high cost estimates.
FraudShield is a very interesting option. Beautiful dashboards, ease of integration with HasOffers, the possibility of a free test for a month are all pretty pleasant factors. Customizing each of the criteria (for example, you can turn off checking for motivated traffic from incent-offers, or remove “white” partners from scanning) allows you to fine-tune the system to fit your needs and traffic. The support is very active in Skype, telling about the result of each test - what the system thought was suspicious, how it can be interpreted and seen in the reports, what should be changed in the settings for more accurate detection in the future. The price also does not bite. However, there is a fly in the ointment: customization of evaluation parameters harms the integrity of the report. If you make the level of "strictness" of the system too low, you can skip a lot of fraud, if it is too high, almost all traffic will fall under the filters. The system has enormous potential, but it needs some work.
FraudScore - formerly known as Clearflow, is a favorite of the tests performed. During the test and use the system has established itself as the best of options. Unlike the aforementioned FraudShield, the settings cannot be customized here (perhaps this is the main drawback - you cannot exclude anyone from scanning), but this allows the analysis algorithm to be complete and accurate. It is also worth noting the responsiveness of the support, the presence of a self-learning system based on the neural network, as well as ease for perception. Also there is a constant testing and updating of algorithms, the author of the article was convinced of the effectiveness of himself. In general, our company chose this particular system and during the time of cooperation it remains satisfied with it.
Of course, this comparison is not the ultimate truth. For example, we could not verify systems using analysis through integration into the SDK of an application - simply because we are not application manufacturers. From the point of view of the advertiser and developer, these options may be more optimal. Nevertheless, the author was faced with the task of finding the optimal system for estimating the traffic of an advertising agency. And I can say that now it has become much easier to detect the intruder, as well as to show him the evidence of fraud.
I wish you success in the fight against fraud.
Despite all the ways to fight fraud on its own, this is a losing strategy. You just spend a lot of resources trying to reinvent the wheel and maybe save a small amount of money by discovering the most lazy and stupid froder. However, those who are smarter will continue to quietly divert profits from under your nose. Therefore, the solution to the problem is an outsourcing, namely, the connection of the fraud protection system.
')
In Mobio, we work hard and hard at clearing traffic, testing different systems, which we want to talk about under the cat.
What is the anti-fraud system for?
Imagine that you need to analyze a huge number of conversions by parameters such as: ip, user-agent, device-info, click time, conversion time, proxy, etc. A good anti-fraud system automates this process and “rejects” suspicious conversions (for example, finding that the installation was done using a proxy or VPN). A good system will analyze not only each specific conversion, but also compare them all in a general slice, in order to establish any dependencies characteristic of froder.
Which system to choose?
After studying the recommendations of colleagues and Google search results for the query: “mobile antifraud system”, I selected several candidates for testing:
I will immediately say about those systems that had to be excluded:
Kraken is a Russian development based on the fact that the source of the conversion is checked for any of the well-known bot and insist farms. The exact algorithm is not disclosed. The cost is low, however, an adequate test did not work out - the system only checks selectively and on certain days of the week. Support comment: “There is no other option for work, the system was not created for this purpose from the very beginning.”
FraudLogix is one of the solutions that are suitable for an ad network, they only have an IP blacklist. In fact, the analysis is performed only by one criterion, which is not enough for a high-quality anti-fraud. In addition, the test scan marked with frod's absolutely normal addresses included in the list for checking false positives. The result is disappointing - low accuracy and just one criterion for analysis is clearly not the option that should be used to protect against fraud.
Kount - I can only say one thing about this system: a prepayment for a year in the amount of several hundred thousand dollars will not suit anyone. We, for example, did not fit.
So, there are only three systems that we could use. I will tell about each of them in more detail:
Forensiq is the loudest name on the list. It is often referred to in articles on anti-fraud. Perhaps my assessment will not be entirely correct, since I used the version integrated into the tracking system Affise , but what I saw weakly pulls on "the best fraud protection system." The report comes in the form of a mark in front of each conversion indicating the level of risk: low, medium, high, and to the listing of suspicious criteria. The inability to somehow track the scoring algorithm and communicate with the representative of the system reduces the final score. I didn’t have any complaints about the quality of the assessment - there were no false positives, Forensiq correctly determined froders. By cons can be attributed to the relatively high cost estimates.
FraudShield is a very interesting option. Beautiful dashboards, ease of integration with HasOffers, the possibility of a free test for a month are all pretty pleasant factors. Customizing each of the criteria (for example, you can turn off checking for motivated traffic from incent-offers, or remove “white” partners from scanning) allows you to fine-tune the system to fit your needs and traffic. The support is very active in Skype, telling about the result of each test - what the system thought was suspicious, how it can be interpreted and seen in the reports, what should be changed in the settings for more accurate detection in the future. The price also does not bite. However, there is a fly in the ointment: customization of evaluation parameters harms the integrity of the report. If you make the level of "strictness" of the system too low, you can skip a lot of fraud, if it is too high, almost all traffic will fall under the filters. The system has enormous potential, but it needs some work.
FraudScore - formerly known as Clearflow, is a favorite of the tests performed. During the test and use the system has established itself as the best of options. Unlike the aforementioned FraudShield, the settings cannot be customized here (perhaps this is the main drawback - you cannot exclude anyone from scanning), but this allows the analysis algorithm to be complete and accurate. It is also worth noting the responsiveness of the support, the presence of a self-learning system based on the neural network, as well as ease for perception. Also there is a constant testing and updating of algorithms, the author of the article was convinced of the effectiveness of himself. In general, our company chose this particular system and during the time of cooperation it remains satisfied with it.
Conclusion
Of course, this comparison is not the ultimate truth. For example, we could not verify systems using analysis through integration into the SDK of an application - simply because we are not application manufacturers. From the point of view of the advertiser and developer, these options may be more optimal. Nevertheless, the author was faced with the task of finding the optimal system for estimating the traffic of an advertising agency. And I can say that now it has become much easier to detect the intruder, as well as to show him the evidence of fraud.
I wish you success in the fight against fraud.
Source: https://habr.com/ru/post/314896/
All Articles