We Sat Down To Talk With Marketers From Matching App 「with」! Using SDK Data for Fraud Protection
Mr. Sato ： To start, I wanted to ask when was it that you started turning your attention towards fraud protection?
Mr. Yamamoto： Around February of 2018. To increase user acquisition from the end of 2017, we began adopting a non-incentive type of advertising menu while expanding the delivery destinations. In terms of results we were observing strong acquisition and increased our budget because of it but after running it for 3 months, when a single month was costing about 3 million yen, I started having my worries about if expanding like this was good or not. It was because I heard that a non-incentive type of advertising menu has a lot of ad fraud that you have to be careful.Mr. Sato ： After that, what kind of actions did you take?
Mr. Yamamoto： We also have a game software department and they have already introduced fraud protection functionality to measure SDK. I was thinking that if we already started to distribute non-incentive advertising menus on “with”, why not consider adapting this functionality too? Based on the advice of the game software staff, I decided to consult with the person in charge of SDK.Mr. Sato ：So you partnered with other departments and learned more about ad fraud?
Mr. Yamamoto： That’s right. We did information exchanges between departments. I guess that a lot of people working in gaming are more prepared to deal with these types of fraud protection. When I was talking with the person in charge of SDK for fraud protection about my concerns, he suggested doing a one month free trial of the function . Since I wouldn’t know anything unless I gave it a shot, I decided to try it. And the result was that 90% of all installations were fraudulent.Mr. Sato ：I can imagine that if you had these worries and saw those numbers then you would have felt uncomfortable. Mr. Yamamoto： I prepared myself for about 20% being fraud but when I saw those results, it was eye opening and there was a sense of crisis in me such that I felt that I had to rethink my approach towards digital marketing from the very beginning. On this advertising menu, it doesn’t matter if it is organic or not – fraud will happen in the advertising results and I thought that if I didn’t have this kind of function in place I wouldn’t be able to protect myself. So that’s when I signed up for the feature. Mr. Sato ：I see, were there problems you could see after you introduced the SDK fraud protection function? Mr. Yamamoto： I felt a little relieved when the trial version came out with accurate results but even after introducing the function, conversion rates remained extremely low and the number of installations were increasing while the number of member registrations remained constant. This type of thing was happening several times in a month. Because of that, I felt that even if you implement a protection function like SDK, I can’t say that there is a perfect countermeasure. Mr. Sato ：You didn’t talk to the agencies when this was happening? Mr. Yamamoto： I talked to them about the abnormal values from the conversion rates like “Don’t you think that this period’s results are a little off?”. It was outside of my claims but finding out where ad fraud was occurring and how it was happening remained a black box. At first I thought that it would be good if the cost was not wasted due to being unclaimed but it was occurring quite frequently and I kept feeling like I was falling into a trap. So I got the feeling that I wanted to understand this problem firsthand. Mr. Sato ：This is probably specific to Japanese people but if you ask the agency for subtraction support for a claim, there are places where it is going to be difficult to get into the details. Mr. Yamamoto： Yes, I was asked to negotiate with the media side firmly through the agency, and I was satisfied with the subtraction process, but I felt a little uneasy about the lack of understanding on the details As it happened, I talked to other agents with which I had a deal in the game division, and I heard that it could be solved with other tools, which ended up being SpiderAF. They told me that SDK’s fraud protection function detects ad-fraud in real time and does not post back, but also that there are some things that are difficult to prevent. For example, if the language setting of the installed device takes an English conversion, it does not detect that it is fraud in real time, as it may happen that the person has it set to English. However, if 80% of the installations are in English when you see it later, and given the fact that “with” is a service offered only in Japanese, weird possibilities happen. After that, I explained how to detect fraud by looking at the data after chasing data and decided to give the free trial of SpiderAF a try. That was exactly one year after we tried the SDK trial. From the SpiderAF trial results I felt that the strength of SpiderAF is that it is great for fraud detection on other device languages, devices sold only overseas, old OS versions and other additional information that can’t be found without looking while SDK’s real time detection is great as a CTIT (Click to Install Time) standard detection. Mr. Sato ：Looking at the details when you were trying the SpiderAF trial, the anomaly on the CTIT popped up and was spotted on the SDK side, but if the language or the device is suspicious it can be detected later – especially on Android we could detect a lot of unwanted ad fraud in real time. On the other hand, with iOS there is a lot of ad fraud that can be detected by the SDK side on this trial. Have you found any good things about fraud protection with SDK or SpiderAF etc.?
Mr. Yamamoto：I think that communication with the agency has become more constructive as the tool introduction has increased my knowledge of fraud. Not only can the cost be properly billed out, but it is also great that the reason is clearly understood. Though I thought that CTIT was a resource for only identifying fraud, by introducing SpiderAF, language and devices etc can be used as a fraud standard, and data can be viewed and judged from different perspectives.
Mr. Yamamoto： Look at the data carefully! I also liked looking at the data, but I did not have the correct knowledge and I did not know what data to look at. What I think now is that if you use the SDK measurement, you have all the data you need, so just learning the right data will tell you what to do. Of course, it’s good to rely on an agent, but I think it’s important that you look at the data you have first-hand.Mr. Sato ：Thank you very much!