Why do you have no idea about data analysis?

The article explores common misconceptions and problems in data analysis and provides practical solutions. The article concludes by emphasizing the importance of combining business scenarios, clarifying problems and goals, logical reasoning, testing, and accumulating experience through multiple rounds of testing. Whether you are a novice in data analysis or a professional looking to further improve your skills, this article will provide you with valuable insights and practical guidance.

Many students complain: "I have no ideas when doing data analysis!" In fact, there are many reasons that can lead to this result. Today, let's take a systematic inventory.

In some cases, it is the data analyst’s own problem. The following three are common:

1. Question 1: Finding nails with a hammer

There are many methods for mathematics, statistics, and operations research, and reading books can make people feel fulfilled. Therefore, some students who are obsessed with reading books start to use hammers to find nails. For example, today I saw the statistical normal distribution, which was so cool, so I saw everyone as normal. Tomorrow I will see the chapter on regression analysis, and I will want to see everyone doing regression...

Doing this will cause trouble. For example, some students calculated the benefits of activities and had to do a regression between the activity funds and the total performance, and then looked at the R-squared value and said that the activity had no effect. As a result, they were naturally criticized by the business.

Moreover, doing this does not mean that you have truly understood the book. If you really understand it, you can at least distinguish between:

Is it a sample statistic or an overall statistic?

Is it a prediction problem or a classification problem?

It is marked but not labeled

Is there any internal logic in the data at hand?

Only by going deep into the business scenarios can you know which method in the book is suitable. And various book methods have fixed application scenarios.

2. Question 2: Putting shoes on your feet

This is a brother of the previous question, both are nerdy behaviors, but the book has been changed from "Statistics" to "Management":

*Because there are 4P in the book, draw the boxes of the four P first

*Because there is PEST in the book, draw four boxes first

*Because there is RFM in the book, let’s calculate RFM first

Then what? ... Then I felt dizzy and didn’t know what to do, and then I was criticized: “What the hell are you doing…”

The solution is the same as the previous one. First understand the business scenario, find the real problem, and then organize the method. Instead of taking a set and applying it to everything you see. Data analysis serves the business. How much the business knows about the problem is the starting point of the analysis (as shown in the figure below).

Question 3: Dismantle everything

This is also very common, no matter what the problem is, pull a bunch of cross tables first.

* For example, when analyzing DAU, we can cross-reference DAU with dimensions such as gender and age.

*For example, to analyze GMV, we can cross-reference GMV with gender, age, and other dimensions.

It is also called: the soul of data analysis is comparison, and the core is disassembly

The result is: without logic and assumptions, the more comparisons you make, the more confused your thinking becomes. You often end up comparing Apple and Rhino. Moreover, this aimless crossover often leads to a distorted business thinking. The business department will hold you back and ask you to explain one sentence after another: why is there a 5% difference here and a 3% difference there? In the end, your thinking becomes more and more confused...

So comparisons can be made, but only by listing the assumptions first, labeling them, and comparing apples to apples can discoveries be made.

In some cases, the problem may not be with the data, but the data is just taking the blame. The following are four common examples:

Problem 4: No business goals

for example:

*Do indicator monitoring, what are the indicator assessment requirements? I don't know

*When doing activity analysis, what indicators should be improved? I don’t know

* Do product analysis, what is the purpose of product revision? I don’t know

Then I don’t know how to analyze it…

In this case, I really don't know how to analyze it. It's like archery. You need a target first to know if you are shooting accurately. If you don't even have a target, just close your eyes and shoot randomly, and then analyze whether this random shooting has the effect of changing the world, bah! What the hell is the analysis?

Of course, most of this problem is caused by the business. But I would like to remind the students who are doing analysis to first clarify the goals. And actively remind the business department: if the goals are not clear, the analysis will naturally be unclear. Otherwise, it is easy to be blamed. Many business departments do not set goals for themselves, and then force data analysts to write: "This random shooting greatly improves the company's performance!" Then when they are questioned by the boss, they say "This is all written by the data analyst, I am innocent"...

There are many ways to set goals, don’t say you don’t know how to do it (as shown below).

Problem 5: Mixed goals and self-deception

This problem is the opposite of "no goal", that is, the business department did a little work, but the result was exaggerated. For example, they put out a 10 yuan coupon, and then started to brag: this 10 yuan coupon can increase GMV, wake up old users, and attract new users. Anyway, it has great effects, and then they ask data analysis to analyze the actual effect of each effect, and give feasible suggestions...

Many students are confused. What the hell is this? How should I analyze it? If you don’t know what to do, it’s just nonsense. Each type of business practice has a fixed form. There are not so many “one trick to rule them all”. So if you want to form ideas, you have to understand the common routines and fixed forms of the business, so that you can distinguish the situation when facing this nonsense.

6. Problem 6: Poor business methods

The most common projects are: user profiling, churn prediction, product recommendation, etc. There is a lot of data, but no business tricks.

You worked hard to predict the churn probability of different users, but what happened? The business side just sent a text message to recall all the users… Yes, the response rate of all the users was less than 1%. What difference does it make whether you build a model or not? In the end, the business side complained: "Your analysis is useless."

The same is true for product recommendations. Many companies can’t come up with a few strong products. They recommend products just for the sake of recommendation, and then come to blame us and ask, “Why is the analysis useless?”

This kind of questioning will make students who work with data fall into deep self-doubt: "Am I thinking in the wrong way?" However, there is no need to doubt that it is not a problem with the thinking, but a problem of inability to implement due to the lack of business methods.

To break this situation: you need to do a good job of basic analysis, have a basic understanding of products, users, and business methods, and know the current capabilities of the company. This way you can easily identify: Is it that my thinking is wrong, or are these guys just good at this?

7. Problem 7: Lack of iteration and accumulation

Good data analysis models are developed through iteration, not just dropped from the sky. The right way is to define a goal, conduct multiple rounds of testing, and find out the upper and lower limits of each business method. Only in this way can we see which method is useful, discover the internal logic, and accumulate analysis experience.

But some companies just like to take crooked paths, such as:

1. When making a business plan, there are a lot of goals, such as "we need XX, XX, and XX, coordinate XX, and work together to XX". It is not clear which direction to measure.

2. Analyze, analyze, and analyze again every day, but don’t do any testing

3. Analyze, analyze, and analyze again every day. After the analysis, the business uses a completely different set of ideas to test

4. If you cannot achieve the goal, change it and try to cover up the problem.

Doing this is like flying without a head. There is no need to accumulate effective experience, and in the end, nothing will be gained. However, students who are deeply involved in this just feel their heads buzzing, and like to doubt: Is it because my thinking is not clear... This is really not the case, this is a standard case of random self-made.

summary

Data analysis should be closely integrated with business, and so should analytical thinking.

Combined with specific business scenarios

Have clear problems and goals

Argue logically

Verify results through testing

Gain experience through multiple rounds of testing

This is the right way to make your analytical thinking clearer and clearer.

Of course, some companies have bad environments, which leads to students being PUA at work, "Your thinking is not clear." At this time, as long as everyone does their job well and accumulates more practical experience on specific issues, they will have the opportunity to leave such a stupid company and find a more suitable job. Therefore, it is very important to discuss details. If you don't consider the details, just stay in theory or superficial, you will make a fool of yourself like at the beginning of this article.