Finally, we can also look at the changes in user portraits. Dependent attributes, such as some common demographic characteristics, behavioral aspects, is the user group of certain behaviors changed. Finally, there are some preference type dimensions, such as the preferences of different users for different types of articles and so on.
02 DAU anomaly analysis summary
By the way, it is best to follow the above-mentioned way of analysis: Seasonal Factors-External Factors-Structure-Portrait
Generally, abnormal changes in DAU are basically caused by products/operations/technology. And the most lethal is technology. A technical bug may cause great harm, and it may be irreversible. Operations, the impact on DAU is probably the most common.
When we analyze the structure and portrait behind. The structure may be more able to change, and the changes in general portraits will not let you find out at once. Maybe you have noticed that our portraits have changed after a long time. But like some very unusual changes, you can find problems more from a structural point of view.
1) The overall DAU dropped on a certain day
We look at the changes of core indicators every day. One day, we found that DAU has dropped a lot. At that time, we thought it was due to the recent strict user experience and the governance of PUSH, so we didn’t pay much attention to this problem.
Later, I carefully studied this problem and found that it was caused by the decline of a certain model. At this time, it attracted everyone's attention. Just at that time, the version was launched, and the platform of the touch system was updated, so I pushed forward the research and development. Go to check the version problem, check the problem of reaching the system. The entire research is jointly developed, mobile development, Zhongtai touch, data warehouse and analysts. Finally, the problem was located, which was caused by the touch system update. After the country email list repair is complete, the entire DAU is restored.
During this period, I also found that the DAU of a certain group was rising. It was found that the DAU of this part was rising because the operating classmates used new touch materials.
Since then, I have established a complete detection DAU reporting system. After the problem is discovered later, some factors can be ruled out for the first time.
2) There is not much change in DAU but a certain group finds the change
After building the DAU detection system, one day I found that the entire DAU has not changed much, but a certain port has changed very abnormally. We later continued to disassemble the source of the DAU of the entire port and found that it was because the new version was released from different terminals. , resulting in the failure of a sharing function, and then the DAU brought by this sharing drops greatly. Finally, after fixing the issue, the DAU for that port was restored.
03 Analysis of abnormal decline of indicators
Based on the analysis of DAU changes and the long-term research on abnormality at work, I have summarized the abnormality analysis framework. The abnormality analysis has four steps: trend-external factors-internal structure-portrait
Whether the positioning is abnormal.
External factors: Changes related to product/operations/technical communication, pay attention to attribution issues.
Internal structure: The composition of this indicator, for example, DAU can see different devices, and some other indicators can also have their own dimensions.
Lastly is the image.
have to be aware of is
When dismantling indicators, you will have the first dimension to reflect the dismantling based on business experience.
When dismantling, sometimes the proportion of different categories under the same dimension is different, we must pay attention to the contribution rate of each category. For example, when DAU is divided into different models, the number of users of IOS and Xiaomi is generally quite different. When you are positioning, you see that the total DAU has dropped by 100,000, but ios has dropped by 30,000, and Xiaomi has lowered the price by 20,000. At this time, the contribution rate of IOS is 3/10, and the contribution rate of Xiaomi is 2/10. . Because, sometimes it is possible that Xiaomi's contribution rate is greater, then it may be that Xiaomi has caused the decline in DAU, not IOS.
Personal experience feels that the best order is the order of 1234 in four steps.