The biggest difference between you and a data analysis expert is...

In the world of data analysis, what separates novices from experts is often not the skill of using tools, but the accumulation of labels for a deep understanding of the business. This article uses a store analysis example to reveal how to deepen data analysis by building business-meaningful labels, thereby providing more valuable business insights.

Many students feel that compared with experts, they have made little progress in their work. They keep looking at the data and comparing year-on-year and month-on-month, but they have not made any progress in two or three years. Behind this, there is a big problem: the lack of accumulation of labels with business meaning leads to only looking at data sporadically, which makes it impossible to derive conclusions with business meaning and accumulate business analysis experience.

Data analysis experts pay great attention to the accumulation of labels. Today we will use an example to show you the difference. Please hold on tight, we are about to start.

1. Problem scenario

A classmate submitted a store analysis report, pointing out that store A's performance ranking was lower than other stores, and suggested improving it. However, he did not expect that such a simple sentence would immediately stir up a hornet's nest. Colleagues in the business department began to argue:

Colleague A: A is a newly opened store, so we shouldn’t compare it with other stores. A is actually very good.

Colleague B: Although A is a newly opened store, it is a standard store and cannot be compared with a mini store. A is actually not good.

Colleague C: Although A is a standard store, it is a bargain store and cannot be compared with ordinary standard stores. A is actually very good.

Colleague Ding: Although A is a bargain hunting store, its marketing strength is no less than that of ordinary stores. A is still not good.

Colleague E: Although A has a strong marketing effort, the marketing investment is not heavy, so A is still very good.

…

Everyone was arguing.

The final conclusion is: "The data analysis is not done in depth, there are only numbers but no interpretation, and it needs to be combined with the business for in-depth analysis." The students who were left to work on data were in a mess:

"What are you talking about???"

“How can I go deeper into the Dharma???”

So, how do we conduct in-depth analysis?

2. Key to Breaking the Game: Business Implications of Tags

The biggest problem here is that the various details of the business discussion cannot be directly expressed as a data indicator, which makes it impossible to quantify, let alone analyze. The key to quantifying business is labeling. Attention! When many students mention labels, they instinctively think of "gender, age, packaging size, packaging color" and other labels that are ready-made in the database and directly imported from basic information. Most of the time, these basic labels have no direct business meaning, and their ability to interpret business is very weak, so they need secondary processing to be useful.

Labels with business meanings are labels that directly point to issues of concern to the business, differentiate problem indicators, and provide guidance for business behavior. For example, "This store manager is not competent" is a label with business meanings. If it is confirmed that the manager is not competent, the next step is to change the store manager or provide training, which provides clear guidance for business behavior. This kind of labeling requires a complex conversion process and data verification, and it cannot be achieved overnight.

So how do you fight? Let's take a look at it step by step.

3. Start by organizing your business assumptions

Since we are labeling business issues, we must first start by sorting out the "assumptions that affect business indicators". This way, the labels we create will directly point to business issues. For example, for the questions at the beginning, we can list them according to the discussions of the business department:

1. Business object to be described

2. Indicators for measuring the quality of business objects

3. Assumptions affecting indicators

4. Hypothesized direction of impact

This way, we have a clear list of tags to be developed (as shown below):

The next step can be to develop them one by one.

4. Make simple labels first

During development, first create labels that can be directly calculated using basic labels + existing data indicators. This type of label is also called a rule label. That is, after the business is given a calculation rule, it can be directly calculated based on the basic label + existing indicators. This type of label is fast to obtain and easy to verify.

For example: whether the store is newly opened. In theory, you can just classify according to the store opening date. For example, if you stipulate that stores opened 6 months or less are all newly opened stores, then stores opened 7 months or more are old stores, and stores opened 6 months or less are all new stores.

Here is a key question: how to come up with the standard of "6 months". There are two ways to do it:

The first one is that if the business department has a consensus, we can directly use the business standard. For example, if everyone agrees that it is 6 months, then it is 6 months.

The second type is that the business does not have a consensus on a specific value, but has a concept, such as:

1. During the new store opening phase, the probability of store closure is very high

2. In the new store opening stage, store turnover/order volume is on the rise

At this point, although there is no clear standard, the business provides a method to find the standard. We can count the life cycle data of all stores to see where the turning point of store closure probability/sales/order volume is, so as to clarify the standard (as shown in the figure below).

In principle, even if the business verbally gives the first standard, I suggest that you guide the business to come up with the second standard. Because the second standard is the standard with business logic. If different business departments quarrel one day, or the business leader changes and no longer recognizes the first standard, the second standard will be the principle of adjustment.

Similarly, the store area label can be typed in this way. First list the business assumptions:

1. The larger the store area, the higher the corresponding cost

2. If the business area is large, the income should also be high.

After that, you can make a list of the existing store areas, look at the parameter range, and make labels (as shown below).

A special reminder here is: many students do labeling without communicating with the business. They make judgments based on their own feelings or data distribution. For example, they spend three months on how to distinguish new stores... This kind of closed-door work can easily be challenged by the business and cannot be combined with business scenarios, ultimately making the labeling work a self-entertainment.

With simple labels as the foundation, we can then deal with complex label situations.

5. Common complex tags

Common complex situation 1: A business problem needs to be described with several tags. For example, the tag "promotion" may need to be described separately in terms of promotion form and intensity. For example:

1. Promotion scope: the number of SKUs participating in the promotion

2. Promotion intensity: users get a discount ratio based on the original price.

3. Promotional forms: buy one get one free, discount for a certain amount, gift, add one more...

(As shown below)

Perhaps a business scenario requires a combination of several tags to be clearly described.

Common complex situation 2: Labels that are merged from two or more basic labels (also called comprehensive calculation labels). For example, "a bargain store" means: this store is large, but the rent is lower than normal, and the customer flow is not much less than normal, so we found it. At this time, the bargain store is spliced ​​from three basic labels (as shown below).

Similarly, for example, "this store manager is not capable", how to prove his incompetence may require demonstration from several dimensions, such as performance, number of jobs, and personal resume. The more dimensions there are, the more weights are involved. There is a set of methodology for weighting, and if you are interested, I will share it separately later.

Common complex scenario 3: Labels are used to predict future situations, not situations that have already occurred. For example, we predict that this store is a "high potential store", so we require it to perform better than normal. Note! Prediction itself is a complex task. It can be based on rule judgment and modeling. There are also several ways to model. Therefore, it is a bit complicated to handle. If you are interested, I will share it separately later.

In short, after a bunch of complex calculations, the labels are now ready and can be used for analysis.

6. Comprehensive Utilization of Labels

The direct application of labels is to quantify complex business problems and then analyze and test them. For example, the complex business reasons at the beginning of the article can be directly compared in a single dimension using labels to test the statements.

If multiple labels are superimposed, complex analysis logic can be constructed and deduced layer by layer. This complex analysis logic is what we often call "in-depth analysis". Generally speaking, considering many situations is called "comprehensive analysis" and the number of deduction layers is called "in-depth analysis" (as shown in the figure below).

Of course, labels can be used in more than one way. For example, labels can be used as feature values ​​for further modeling and input into the model for comprehensive evaluation/prediction. Many students’ evaluation models/prediction models are inaccurate because they lack label accumulation and directly feed a few simple raw data into the model.

For example, tags can also be used to deduce business actions. For example, "store manager is not capable enough" or "marketing efforts are insufficient" can directly lead to conclusions such as "I want to train store managers" or "I want to increase marketing investment."

In summary, labels are an important part of in-depth analysis, modeling, and business suggestions. Students can try to create more labels with business meanings, especially for businesses involving "blind box" status, such as online advertising, offline sales follow-up, product selection, etc., where labels play a greater role.