Say goodbye to routine analysis and explore infinite possibilities with your mind

In different scenarios, the process of data analysis is often very different. Is there a way to help us sort out the ideas of data analysis in different scenarios and not be limited to fixed scenario analysis routines? This article introduces the influencing factors behind each step of the analysis routine in the process of abnormal analysis, helping you better understand abnormal analysis and respond to different contingency scenarios. It is recommended for friends who are interested in data analysis to read.

There are many scenarios for data analysis, here are some of them:

• Change Analysis
• Product Analysis
• Business Analysis
• Operational Analysis
• User Analysis
• Channel Analysis
• Retention Analysis
• Conversion analysis

The data analysis process is different in different scenarios.

This creates a problem: with so many different scenarios and different processes, it is basically impossible to remember all of them.

So there are many data analysts around me. When faced with specific problems, they usually search for analysis processes online first, and then start subsequent analysis based on the existing processes.

Is there a way to remember all the analysis processes?

In the world of Go, you will find a large number of fixed patterns, which are routine ways of playing.

For those who do not understand why the chess is played this way, they can only master these patterns by rote memorization, which is not very efficient, and they will be helpless once faced with new chess patterns.

However, if you can understand the reasons behind the fixed way of playing, then what you remember is no longer just the routine formula, but you can deduce the fixed way of playing by yourself.

In this way, when faced with new chess situations, you will be able to respond to changes flexibly instead of being helpless.

I think it is the same as the Go formula to remember all the analysis processes. If you understand why each analysis process is the current few steps, you can deal with it easily even when facing business problems that you have never seen before.

As an author in the field of data analysis thinking, I decided to try to analyze the derivation process of each analysis process to help you get rid of routine formulas. You don’t have to remember the routines, but no routines are better than having routines.

Today I will share with you the analysis routines for abnormal movement analysis.

1. Analysis of abnormal movements

When data indicators change, the most nervous people, besides the students in charge of business, are data analysts. Because data analysts are about to face three soul-searching questions:

1. Why is it rising?
2. Why did it fall?
3. Why not rise or fall?

How to answer these questions?

There are many reference ideas for abnormal analysis on the Internet. The general steps are:

Confirm that the data is accurate . Check whether there are any problems such as database anomalies/data reporting delays/data SQL errors/inconsistent data statistics.

Confirm the type and range of abnormal indicators . Check whether the abnormal type is sporadic/cyclical/trend-like; whether the increase/decrease of the abnormal change is reasonable.

Disassemble and locate the cause of the problem. Initially divide the dimensions and locate the approximate scope of the cause. If it is an e-commerce company, the dimensions of people, goods and places are generally used for disassembly:

• People: new and old customers/by channel/by customer acquisition method/by user value label/by user intention/user portrait (gender, age, region, occupation, etc.)
• Goods: Different categories/different specifications/different prices
• Field: Traffic channels/recent promotions/industry performance

There is nothing wrong with the above analysis process itself, but it certainly cannot be memorized by rote.

How can we derive the above steps through analytical thinking?

2. Clarify the problem

There are at least three basic steps in analyzing any problem:

1. Identify the problem
2. Analyze the problem
3. Solving the problem

Then the first step in abnormal analysis must be to clarify the problem.

The question is, what is the problem?

You may think this question is a bit strange, but a question is just a question, it’s that... question.

Well, okay, you may be right, the problem is indeed... a problem.

A more precise definition of the concept of "problem" is:

The problem is the gap between reality and expectations.

The problem is the gap between reality and expectations.

The problem is the gap between reality and expectations.

Important things should be said three times.

From this concept, we can deduce that the problem consists of three elements: reality, expectation, and gap.

So there are only three reasons why a problem occurs:

The first is that there is something wrong with the current situation . The current situation you see is different from the real situation. For example, due to differences in statistical calibers, data transmission errors, etc., the numbers you see do not match the real numbers.

The second is that there is something wrong with the expectations . The expectations do not match the actual situation. For example, the transaction volume will drop significantly at the end of each year, so the expected value must be adjusted downward to match the actual environment.

The third type is that there is a real gap between the display and the expectation, but the reason is unknown . For this type of problem, we need to find the reason for the gap.

Compared with the routine analysis process, the three types of errors mentioned above, the first type is actually a problem of data accuracy, the second type is a problem of business cyclical fluctuations, and the third type is the problem that really requires in-depth exploration of the causes.

For the third category of issues that require further investigation, we also need to further clarify the issues . When did this gap arise? How long has it been there? How big is the gap? And so on.

The above problems are actually described more clearly through the 5W2H method . Through the description of 5W2H, we can know whether the gap is sudden or continuous .

So far, the first step of the analysis, clarifying the problem, has been basically completed.

Even if you don’t know the routine of abnormal movement analysis, you can still deduce the routine yourself.

3. How to choose the disassembly dimension

There are two types of problems that require further investigation: one is sudden and the other is continuous.

1. Sudden abnormal movement

When data fluctuations occur suddenly, they are often caused by major business changes, policy changes, market opinion, etc.

The best way to deal with this kind of sudden abnormal situation is to consult business colleagues directly.

Sometimes, spending hours on analysis may not be as effective as spending a minute asking a salesperson. Because the salesperson may have just added a new delivery portal or increased the delivery budget yesterday, and he can answer your question with just one sentence.

In addition, business colleagues are usually more sensitive to industry information and can provide you with some valuable information to help you shorten the time of troubleshooting.

2. Persistent issues

For persistent problems, a disassembly approach is usually required to find the ultimate cause.

There are usually two ways to disassemble: one is the horizontal disassembly dimension, and the other is the vertical disassembly process .

Taking the example of a certain e-commerce company whose GMV revenue continues to decline, we can use the above two disassembly methods. For the horizontal disassembly dimension, we can disassemble it according to dimensions such as product category, drainage channel, product price level, user region, etc. The vertical disassembly process can be carried out through traffic funnel analysis.

Here a question arises: which disassembly method should be used first, horizontal or vertical?

I think there is no specific order, because both ways of disassembly will eventually lead to the same result.

If we first use the horizontal decomposition dimension and find that the GMV of a certain category has decreased, then we will next need to decompose the traffic conversion process under this category and find out which step the problem lies.

If we first use the vertical decomposition process and find the traffic problem, then we also need to use the horizontal decomposition dimension again to find out which category has a decrease in traffic.

Therefore, you can use any disassembly method first and achieve the same effect in the end.

Although there is nothing wrong with the order of disassembly, the choice of dimensions in horizontal disassembly will drive people with obsessive-compulsive disorder crazy.

Taking e-commerce business as an example, the three dimensions of "people, goods, and places" can usually be used among the common splitting dimensions.

• For the dimension of people, they can be classified according to new and old users, high- and low-value users, users from different regions, etc.
• The dimension of goods can be divided into categories, goods in different price ranges, etc.
• The venue dimension can be divided into event venues, traffic channels, etc.

However, if you break it down for every dimension, it would take too much time and effort and would not be practical.

So we need to choose which dimensions to decompose first.

Without going into the specific routine, let’s try to deduce it ourselves.

The purpose of disassembling and analyzing the abnormal data is to find out the cause of the problem and then have someone responsible for optimization and improvement.

Therefore, the first priority of disassembly is to disassemble according to the current division of labor of the team.

For example, if an e-commerce company has divided its business into multiple independently operated teams based on categories, then the first priority when facing data changes must be to break them down by category.

Because after breaking down the categories, you can directly find the corresponding person in charge.

Otherwise, you first separate the new and old users, and then tell the business side that the new user has a problem. Then who is responsible for this new user? It is impossible to break up the entire organization and re-divide the work because of a small problem.

According to the division of labor, after the problem is located in a certain category, the person in charge of the category knows that it is his own problem. If the traffic is declining, he will increase the investment, and if the conversion rate is low, he will do promotion. If the problem is not particularly serious, this abnormal analysis is generally over.

Sometimes the analysis will go a step further. If the person in charge of a category wants to have a deeper understanding of what went wrong with his or her category, then after breaking down the category, a deeper analysis will be done.

This step of decomposition can still be done based on the division of labor . There will also be different divisions of labor within the category team, such as user operations, category operations, and traffic operations. Because different companies have different job settings, the decomposition dimension should be selected based on the actual division of labor. If there is no user operation within the team, there is no need to do a detailed decomposition of the user.

If there is user operation in the team, you should also communicate with the business to see what the user stratification model of the current user operation is and whether there are ready-made user stratification, user grouping and other operation plans . According to the original user operation method, the most accurate cause of the problem may not be found, but the implementation is very good. Relying on the original user operation platform, the operation students can directly select user groups for operation.

At this step, you still don’t need to remember the routine, and you don’t need to remember which dimensions to break down into for people, goods, and venues. In actual analysis, you need to know information such as team division of labor and existing strategies. Based on this information, you will naturally know what to break down.

If you answer like this during an interview, it can also give you a lot of extra points. Because others know what dimensions to break down, and you not only know what to break down, but you also know that the principles you guess must eventually be implemented, so you are more practical than others.

IV. Summary

Analysis of abnormal changes is basically a must-ask question in interviews.

There are many systematic routines on the Internet now. Although routines improve work efficiency, the routines themselves simplify the thinking process, cannot cope with all scenarios, and will limit the growth of personal abilities.

If you have an in-depth analysis process like the one above with the interviewer during the interview, I believe it will add a lot of points to you.

In the eyes of the interviewer, you are not a person who only knows how to use routines, you are really good at analysis . And your analysis can really be implemented, you are a practical person.

Your skills are transferable and your results are realizable. I believe you will be the most attractive guy during the interview.

Author: Jason

Source: WeChat public account "Sanyuan Variance (ID: sanyuanfangcha)"