Some operators who do put channels, each channel are put in, traffic is particularly high, but the activation volume in the single digits. It is also possible to activate the number of clicks are high, but low retention rates. Costs are spent, but the results did not come out. To analyze their own background data, but not the conclusion.
I want to say is that the premise of the analysis is to get data fly. If the data is not accurate, the conclusion can be imagined -
Of course, and then fly platform, there may occur in some cases do not fly. why? As the saying goes, there is the list where there is brush list, there have data statistics platform cheating workshop.
There are many unknown amount of channel brush studio in the mobile Internet ecosystem, these studios at a very low price the same low quality contribution of user data.
Statistical Analysis Platform SDK earlier forged these packets based on the packet plaintext jason, studios can easily use the program to simulate new active, retention, duration, and other user data. With the development of statistical analysis platform, many analysts platform to launch the SDK, developers based on binary protocols can also call their own encryption switch. These technologies enhance the statistical platform of security and data accuracy is improved. If the security protocol App upgraded to version SDK, the amount of brush chamber has been difficult to take the form of analog data packet directly to the amount of the brush.
The so-called foot step ahead, platform-platform, an amount of brush chamber brush amount tricks. In addition to the use of distributed human flesh brush amount of ways to brush the amount (the form can refer to based on the integral wall task); technical strength are able to write scripts, modify the real machine parameters, drive real machine run (Interested students can learn iOS look igrimace the amount of brush tool). These actions have almost no much difference with the real user behavior, it is difficult to distinguish these data technically.
In fact, there are experienced operators who can still tell the difference between true and false to the user through a number of data indicators.
Channels impact assessment
(1) retention
Sometimes the amount of channel brush will choose to import user data on these important points in time the next day, on the 7th, 30th. We will find APP data on these key time points the following day, the 7th, 30th, significantly higher than other point in time. In fact, the real user retention curve is a smooth exponential decay curve, if you found your retention curve abnormal fluctuations renewed steep drop, basically intervened channel data. One can imagine that such user quality is very poor and do not have commercial value.
Retention curve can not only help us to determine the quality of the channel, reference may also be given a lot of advice on the operation promotion and product optimization. Retention is so important, then retention is how to calculate it?
Defined retention: Added one day users, the proportion of n days after the visit, is this day n date retention. For example, if we get on February 1 of 1000 new users, these users have 400 users visit on February 2, February 8 there are 200 users visit, the February 1 Nisshin user the next day retention rate was 40%, 7 retention rate is 20%.
Retention is common indicators in the industry to determine the quality of the user. In the mobile Internet industry, the next day if an App retention rate of 40%, 7 day retention rate of 20%, on the 30th retention rate of 10%, this App retention rate is higher than the industry standard of. In general, the retention rate of retention tools applied to gaming apps, retained high retention rates high frequency applications in low frequency applications. In addition to factors of application types, retention even went App user experience, and other related promotional methods.
(2) a user terminal
Each channel has its own user base covering their user terminal will be different. For example, millet application store user may TOP10 models are millet phone, mobile users might MM vast majority of mobile operators users.
Excluding these app stores have special channels, the channels most users of the mobile Internet terminal with terminal distribution is similar. We can understand these data by looking at the industry data, these data as a benchmark, to compare the analysis of data App.
For example, can focus on property terminal equipment, operating systems, networking, operator, location of these mobile devices. Below I've listed some tips, please share with brick.
Method One: Focus on low-cost equipment rankings
You can focus on the analysis of new subscribers channels or start ranking the user's device. If you find a low-cost device under abnormal ranking front that merits our focus. These data can be found in the terminal property distribution statistics platform.
In particular, no simulator iOS platform, all user data needs to be triggered by a real machine. Many will choose to brush the amount of studio iPhone5c buying used to do brush the amount of real machine. Had to make a channel to promote such stepped pit friends, find a channel that 75% of the equipment is iPhone5c, iOS devices accounted for more than the top5 of. Then they found that the retention channel and other indicators are far from satisfactory, finally found this channel uses a lot of iPhone5c to brush the amount.
Method Two: Focus on accounting for the new version of the operating system
After my many years of work experience I found that many channels will brush the amount of delay in the studio version of the operating system adaptation. It is recommended that personnel channels when viewing channels the user's operating system, the operating system can be distributed to all mobile phone users comparison. If you find a channel below, the new version of the operating system (such as iOS8.x) does not exist, there is a possibility that the studio's technical cooperation in this channel has not yet adapted to the latest operating system.
Method three: focus on the use of wifi network
Some friends asked me, in the proportion of users use wifi reached below 90%, this proportion in the end of normal is not normal. To answer this question, we first need to understand the current situation within some.
It is a high-speed network environment, whether new users or active users, wifi use relatively large proportion. From the user's behavior, if you look close friends, you will find we tend to use when downloading App wifi (flow Guia), compared to when you start App, the current sensitivity of the network will be worse. In other words, the ratio of new subscribers to use wifi wifi will be greater than the starting proportion of users.
In addition, the proportion of wifi associated also with the type of application. If you are a type of online video applications, the proportion of wifi may be more than 90%.
If you are a small flow of App, while being able to see the comparison of the data for clues on the wifi new users and active users, may really be a channel in the mischief.
Method four: targeted delivery is also very important
To do for a long time within the industry have a friend taught me an experience that cheating Fujian more, we can focus on considering blocking, cheating and more areas in the development of strategies put in the time. The blacklist can also be customized according to APP actual geographical division put in effect.
In addition, when we need to focus on delivery is also possible to select some areas served based. For example, north of Guangzhou in these areas of high consumption, such as three or four lines of these areas is relatively blue ocean. We need to verify that the user and our consistent strategy put in the time to view the data.
(3) user behavior
Method One: compare user behavior data
App to do if a relatively long time, visited pages, long time use, access interval, frequency of use of these behavioral data will tend to stable. App of different behavioral data is different. Brush possible amount of studios can simulate the behavior of the user plausible, but it is difficult with the daily data of your App to do exactly the same.
Long a user's channel, frequency too high too low is questionable. When we usually do channel data analysis, these data can be compared with the entire App or the Android Market, the application treasure these large applications store data as reference data for comparison.
Method two: what's new users, active user data-hour time point curve
Many studios brush amount to forge bulk import data by way of data or a timing device initiated. In this case, the new and the start of the curve will be steep and steep drop. New real users and are starting a smooth curve.
In general, users of the new start and reach a peak after 6 pm. And compared to the start of the new trend will be more obvious.
We can time-sharing data from different sources to compare, find anomalies. It should be noted that the comparison of this behavior data needs to follow the principle of a single variable. In other words, in addition to different channels, other factors in the experiment must be identical. If we select channel A on Wednesday the number of active channels and the number of active B do comparison in Saturday's two data certainly it is different, not comparable.
Method three: View user visits a page name details
Some studios will hit the other high frequency appkey App. In this way, we may find that channels the user's data is very beautiful, but careful observation can be found, the page name in a large number of pages not define yourself. By comparing the page name, you can navigate to this form of cheating channels.
If AndroidApp, this name is the activity or fragment; if it is iOSApp, this is the name of the custom view. This not remember it does not matter. I remember looking for developers to look at the list of names specific page, compare page detailed statistics backend user access, you can see the difference.
(4) Conversion Analysis
Analysis of conversion data may not only help us to cope with cheating channels, it can also help users determine the quality of our different channels, improve delivery efficiency.
Each App has its own target behavior. For example, the target behavior electricity business class applications that the user purchase goods. Applications need to look at games like pay within the application. Social networking applications will focus on user-generated content. Operators need to target behavior to define and design applications.
If a user is a real flow, he will experience a click, download, activation, registration, until the process is triggered target behavior. We can use these steps to make the funnel model, observe the conversion of each step. Step closer and funnel, the more difficult to cheat, the higher the value of the user of the system obtained, we pay higher user costs. Operators need to monitor the behavior of the target, when the channel promotion, visit the conversion rate target behavior, raise the marginal cost channels cheating.
Anti-cheat module
In addition to ready-made addition to using statistical analysis tools, you can also apply to have R & D staff to develop their own anti-cheat module. Anti-Cheat antivirus software module is similar in principle, we can define some behavior patterns, added to anti-cheating blacklist database module. If a new device meets the definition of patterns of behavior, it will be judged as a cheating device. Each operating personnel can be defined according to their own App. I have listed some common patterns of behavior:
(1) No abnormal apparatus: frequent reset idfa
(2) ip abnormalities: frequent replacement location
(3) abnormal behavior: purchase a large number of specials, etc.
(4) packet is incomplete: only startup information, do not have the other user behavior pages, events and other information
Written in the last
Operations staff to do long-term and channels together (dou) as (zheng) psychological preparation. Good use of the data is the first step ××× Long March.
It is recommended that a reliable statistical tool APP channels of -shareinstall
shareinstall One such feature is the installation of App channel statistics. Channel-link channels substituted package, each mounting channel accurate positioning of APP. Higher accuracy channel statistics, allowing policy makers to promote a more accurate understanding of the effects of different channels, different channels of efforts to promote the decision to provide the basis for decision-makers can more accurately grasp the promotion of data extension workers.
Each operator will be able to hope that through the use of data, pick out the appropriate channels, improve revenue streams delivered.