Data Operations: Distribution Analysis & User Retention Analysis

1. Distribution Analysis Method

1. What are the common group divisions?

(1) By frequency of events

For example, on the page of a public course on a professional skills improvement platform (see the picture below), we can not only view the PV, but also view the PV by the number of user views, and see how many people have viewed it 1 to 3 times and how many people have viewed it 3 to 5 times.

The difference in the distribution analysis method is that we usually only look at how many times the page has been viewed, and then average it. What we see here is just an average. Many users may view it dozens of times a day, while some users may only view it once. If we look at the number of views of the entire page alone, we cannot get the real usage information of users. If we use distribution analysis to look at the number of times users browse a page, we will have a clear understanding of the number of times this page has been browsed.

(2) Distribution by time of day

Let’s take the above professional skill improvement as an example: the number of views is mentioned above, but this is the course page. We not only need to look at the number of views, but also the length of time the course is viewed. For example, for the same viewing, some users watch for 1 hour, and some users watch for 3 hours. This is obviously different.

(3) By consumption amount range

Finally, we can also divide it according to the range of consumption amount. We are a professional skills improvement platform website, and there will definitely be different users who purchase different courses.

Here we can not only view the number of orders, but also divide the intervals according to the consumption amount, so that we can know how many users on our platform spend 0-500 yuan, how many users spend 500-1000 yuan, and how many users spend 1000-2500 yuan.

The above are the three main analysis methods of distribution analysis. Its operating principle is: to observe the distribution of events in different dimensions in order to understand the information of other dimensions of the event in addition to the cumulative number and frequency.

2. What are the applicable scenarios?

Scene 1

It is already known that a group of users have completed a specified event, but the user group needs to be segmented and divided into different groups according to different dimensions (such as degree of dependence) and value (payment amount) for subsequent maintenance or analysis.

For example: select the users who are particularly dependent, set up a special project for user operation, and operate the users. You can also carry out some operational activities for users who pay large amounts.

Scene 2

Now that we know the number of times a single event is completed, we want to know the distribution of these times after splitting them into different dimensions, so as to have a clearer understanding of the completion status of the event.

For example, if you compare the number of views of different contents, there may be no difference. However, if you distribute the number of views of several contents by time, you can see more detailed information, such as the viewing time period of users of a certain content. More users view content A in the morning, while more users view content B in the afternoon.

From this, we can find that different contents have some characteristics in their time distribution. At this time, different types of content can be recommended to users according to different time periods.

2. User Retention Analysis

1. Why should we look at retention?

In the short term: To understand the quality of a channel, we usually look at daily retention.

Measure the current and future performance of users coming from this channel on a daily basis. It should be noted here that when using [X-day retention] as the comparison standard, interference from other daily data should be avoided. In the long run: When observing the entire market, we usually look at weekly retention/monthly retention in units of weeks/months to measure the health of the product and observe the user stickiness on the platform. Remember to remove the duplicates.

The following product shows the performance of new users added from January to December within a one-year period and their retention in the following months.

As can be seen from the figure below, with the continuous optimization and iteration of the product, the monthly retention rate has increased, which proves that the iteration and operation direction of this product is relatively accurate.

Think about it: Why choose monthly retention rather than daily retention when verifying the long-term value of a product?

This is because the retention data itself does not fluctuate too much. If we look at daily retention, there will be a lot of data to process. The large amount of data often prevents us from focusing on the right place. It just so happens that we usually iterate a version for half a month or a month, which often fundamentally affects retention.

Therefore, observing the changes in product retention over a month can help us better understand changes in the long-term value of the product. This is the most common way to calculate retention.

2. How to calculate retention

(1) Common calculation methods

The calculation method of overall retention is to cross-de-duplicate the user ID at a certain time with the user ID at another time.

However, the retention of the market will be affected by many factors. Let’s take a simple example: if your product launches an activity and introduces low-quality channels (this happens all the time), then the retention rate will plummet on the next day/week/month.

Another possibility is that when you organize this event, the number of users coming from low-quality channels surges, causing server crashes, etc. Therefore, every aspect, whether it is product, operation, technology or market, will have an impact on retention.

(2) Accurate retention

There are two methods to calculate precise retention: First, filter the user IDs that have performed specified behaviors and calculate them separately.

For example, for an online reading product, the following is the overall retention. Here we need to distinguish whether users who read a certain type of book are more likely to stay than other users.

At this time, you need to filter out the users who have read a certain type of books (such as inspirational books) and only look at the retention status of this group of users. After checking, we found that the retention rate of users of this type of books is higher than the market as a whole. Does that mean that there are other types of books that are lower than the market as a whole?

Therefore, through such observations, we can know that different types of books have different abilities to attract users, and then judge the operational quality of different types of books and the value of user stickiness.

Second, divide users into different groups based on their different attributes and observe the differences in retention between them.

For example: We are the product managers of Wangzhe Pesticide. If we look at the data through the overall market, we cannot see the reason. At this time, we can divide users into different user groups. Here we divide users into zones. Through zone division, we can find that some zones have better weekly retention than other zones. Then we can study why this zone performs well, and then copy the good points of this zone to other zones.

The last architecture diagram summarizes:

Well, the above is my sharing about distribution analysis method and user retention analysis method. I hope it can provide you with some ideas and inspiration. Welcome to communicate.

Author: Cai Cai

Official account: Caicai Lao Products (caicailaochanpin)