When the product’s daily active users (DAU) decrease, how should product operations analyze it?

In this article, I chose a specific problem to write about. Product core data anomalies are problems that are often encountered in work and are also common Internet interview questions. Here I combine some sharing on the Internet and my own experience to summarize some thinking and analysis frameworks so that everyone will have a clear focus when encountering such problems.

Case Introduction

The daily active users of an information flow APP usually remain stable between 790,000 and 800,000, but it suddenly dropped to 788,000 on June 13, and dropped to 785,000 on June 15. The product manager is anxious and asks you to find out the reason for the data drop as soon as possible. This kind of problem is still a headache for most people, because for a product of 800,000 yuan, ten or twenty thousand is not a very large fluctuation, but the cause still needs to be investigated.

When you get this question, do you feel like you don’t know where to start analyzing? It doesn’t matter. Let’s sort out the common routines and then look back at this case.

Key point: First make a hypothesis about the cause of data anomaly, then use the data to verify the hypothesis.

It is not recommended that you break down the data yourself as the first step. There are many factors that affect daily activity data, and it is impossible to break down and compare all dimensions one by one. It is easy to waste time without any valuable discoveries. The core of analyzing the causes of data anomalies is to combine past experience and various information to find the most likely cause hypothesis, verify the hypothesis through multi-dimensional analysis of data splitting, and locate the problem. During the process, a new hypothesis may be established or the original hypothesis may be adjusted until the cause is located.

Step 1: Confirm the authenticity of the data

Before starting the analysis, it is recommended to confirm the authenticity of the data. We often encounter bugs in data services, data reporting, and data statistics, which will result in abnormal values ​​appearing in data reports. Therefore, find products and R&D related to data flow to confirm the authenticity of the data.

Step 2: Preliminary data splitting based on several common dimensions

Calculate the impact coefficient: Each data must be compared with the previous normal value to calculate the impact coefficient.

Impact coefficient = (today's volume - yesterday's volume) / (today's total volume - yesterday's total volume)

The larger the impact coefficient, the more it indicates that this is the main drop point.

The above are several common preliminary splitting dimensions. Through preliminary splitting, the approximate scope of the causes can be located.

Step 3: After locating the abnormal range, make further assumptions

Further investigation will be carried out based on the impact scope initially located. Based on the three dimensions of hypothesis making, it is recommended to set up a special group for data anomaly issues, bring together relevant product, technical, and operational personnel, and understand what product, operational, and technical adjustments were made around the time of the data anomaly.

Comprehensively consider the most likely causes of previous data anomalies, adjustments to product operation technology, and the impact of preliminary positioning. Then, combine your own business experience to determine several most likely hypotheses, prioritize these hypotheses for data verification, and check them one by one.

Finally: Break down your hypothesis and establish your reasons

In addition to the above, there are too many dimensions that can be subdivided for analysis. Logically speaking, the key point is that after a hypothesis is verified, the data can be split into finer dimensions based on the hypothesis being true. We need to remember this analysis method. When we guess that the data is abnormal due to a certain reason, as long as we find the opposite of the segment represented by the reason for comparison, we can prove or disprove our guess until we find the real reason.

Case Study

The above is the analysis routine for core data anomalies. When you first received the problem, you didn’t know where to start the analysis, but now you feel that there are actually many points to start with? Let’s go back to the case just now. According to the above routine, we first split the active volume of new and old users, as shown in the following figure (old users on the left axis, new users on the right axis):

It was found that the daily activity of old users was relatively stable, but the number of new users dropped sharply since June 13, so the impact coefficient of new and old users was calculated:

Old user influence coefficient = (77.89-78)/(78.8-79.5) = 0.16

New user impact coefficient = (0.98-1.5)/(78.8-79.5)=0.84

The new user impact coefficient is 0.84, which means that the decline in DAU is caused by new users. After clarifying the scope, we can further segment the new users. What do they consist of?

New users = channel 1 + channel 2 + channel 3 + other channels , so we split the daily activity of new users by channel:

Through channel segmentation, we found that the number of new users in Channel 3 has dropped sharply since June 13, so we located the problem in Channel 3, and it should be the channel effect of Channel 3 that has problems. Contact the person in charge of channel 3 to identify the specific reasons. Is the channel lead volume decreasing? Channel conversion rate decreased? Problems with the channel platform? After finding out the cause, solve the problem accordingly and formulate a channel optimization strategy.

Final words

This concludes this article. I have described in detail the analysis routine for core data anomalies and provided a small case that is easy for everyone to understand. I believe that next time you encounter such a problem, you will at least have a clear starting point.

There is something else I want to say to everyone: In order to make it easier for everyone to understand, the data of this small case is fictional, and the problem locating process is also relatively simple. However, in actual business, the reasons for data anomalies may be multifaceted (this article only discusses some internal factors, and the external environment and competitors will actually affect core data). Sometimes it is also necessary to establish a statistical analysis model to do some quantitative analysis.

It may take several days to troubleshoot the problem. This process is tedious and boring. If the verification fails, you may feel frustrated, and you may work for a long time but still fail to find the cause.

In fact, this is a very normal thing. Data anomaly analysis is a headache even for a senior data analyst. Therefore, we need to pay more attention to data changes in our daily work. As we become more familiar with the business and our data sensitivity increases, we will become more and more proficient in analyzing data anomalies and find problems more quickly.

I hope this article is of practical help to everyone. If you want to know more about Internet data analysis in the future, please follow, like and forward it. You are welcome to discuss more topics together.

Related reading:

1. Product operation and promotion: How to compete for traffic?

2. How can product operations increase the number of new users and retain them?

3. Product operation: 2 major ways to get started to accurately capture private domain traffic!

4. Product Operation | How do stranger social products guide users?

5. How can product operations conduct good competitor research and analysis?

6. Product operation and promotion | 5 underlying ideas for traffic growth!

7. Product operation: application of data system under the growth model!

8. How can product operations conduct good competitor research and analysis?

Author: Zhao Xiaoluo

Source: Zhao Xiaoluoluo, WeChat: luoluo963