Retention rate Analysis case--The mystery of the next day's retention rate plunging 50%

The author of this article, Yu (@ Small white Study data analysis) recently in the retention analysis, encountered a lot of situations, but also often asked the author, why his game suddenly the next day the retention rate dropped by half.

If the retention rate is only a simple indicator, it is very limited to your value, today and you talk about a case, this is not long ago solved the problem, I believe it will help a lot of people. This will also be the first article in the analysis of retention rates, which is shared with you later.

The statistics found a 50% decline in the next day's retention rate for 3rd, but no changes were shown in the overall trend of dau.

However, by looking at the installation, the number of user registration, found that there is no significant increase in installation volume fluctuations, but the user's registration volume suddenly increased. The following figure is a screenshot of the system statistics

Let's take a look at the number of user registrations

From the above data, the initial determination is two cases:

new server old player brush number

For the first case, I did the following trends for registration and installation.

The game's official website got the timetable for the game to open the service

In addition to the January 6 wave crest is due to the game made soft text, stimulating the growth of game users, the other red circle (except January 16) are in the weekend to open new clothing to stimulate the growth of new users, working days opened new clothes did not appear crest.

For example, January 3, January 7, January 9 and so on. The game on January 18 to open a new suit, according to the experience just now, January 18 will not appear a large crest, but from January 18 ~ 20th appears a larger crest. This excludes the impact of the new server on the weekday.

Then there is the second situation, the old players have the possibility of brush number. Then we need to do two things:

continue to look at the breakdown of data, such as registered active accounted for ratio, registration and installation conversion rate, player-day game times, retention trend performance data to continue to find data problems during the operation of the situation, easy to locate problems.

Here we first say 2nd, I found an activity in the game forum: After the new dress opened, new gangs in the first 3rd after the service, summoned 10 players to join their gangs, that is to send a large number of gold coins.

As a result, the basic definition of the problem is here. But we have to look at the problem at that time from another level, that is, from a data perspective.

number of games in a single day

Obviously found that the number of 18-20-day games increased significantly, which is a sign of the trumpet increase, brush number, because just now we saw that the installation volume of this period did not increase, but registered a substantial increase.

Single Game Long

The length of a single day game has always been relatively smooth and stable, but in the 18-20 days of 3rd, there has been a significant fluctuation, that is, the user of the game when the length is not high, that is, a large number of low-level accounts.

Retention Trend Performance

The retention rate allows us to quickly locate the problem, whether it is a problem with the quality of a new user, a day or several days of external events that result in a retained change.

if it is a user quality problem, then this batch of users of the new next day retention rate, 2nd, 3rd, such as the retention rate will be low; if it is an external event, then the retention rate of new subscribers in different batches will be very low in a certain statistic day;

Let's take a look first:

Changes of the next-day retention rate of

It is clear that the next day's retention rate was significantly lower than the 18-20-day three-day decline, and the next day's retention rate returned to normal after three days.

Next, we look at the 18-20-day retention trend and the retention trend after 21st.

Here we can clearly find that the trend of the 18-20-day retention curve is generally lower than the 21-23-day retention curve after the trend performance, that is, 18-20-day http://www.aliyun.com/zixun/aggregation/18538.html "> New user quality is not high."

Because of the large number of old users to refresh the number of logins caused by data growth, such users in fact, the active degree is limited, that is, in order to gain benefits, the use of the trumpet cheat to obtain incentives, and at the data level performance is difficult to see.

In other words, this is the operation of the design of the problem, indirectly affected the performance of various data. As for the second case, here is not to say, in the following article, will talk about this problem.

Here is very simple, the analysis of the retention rate is not isolated, nor just look at it, driving the retention rate analysis, can help us solve a lot of operational problems.

For example, today's discussion is due to the fact that the comparison transaction set up by the operation activity causes the data to drop, or because the interference from external events causes the data to fall. The single retention rate index is not very meaningful, but the comprehensive utilization of other indicators, combination positioning, analysis of problems, shows its role. In later articles on retention rates, it will continue to be the case for the analysis of retention rates.