What is multidimensional data analysis? How to do it?

This article deeply analyzes the concept and practice of multidimensional data analysis, points out the common mistakes made by data analysts when conducting business analysis, and provides a new analytical approach. The article not only explains the difference between business and data analysis, but also proposes a systematic six-step analysis method to help readers deeply understand and apply multidimensional analysis from a strategic level.

Students who do data analysis have all encountered this problem: Analyze the problem from multiple dimensions and make suggestions that are meaningful to the business. This topic seems simple, but many students have worked hard to run a bunch of reports, but in the end, they only get a lot of complaints from the business:

"What does all this mean for you?"

“Where is your focus?”

"Your dimension is too one-dimensional!"

Good! Unjust! Wrong!

Why are people still saying "there are not enough analysis dimensions" when there are so many sets of data? Today we will systematically answer this question. The essence of the problem is: the "multi-dimensionality" mentioned by the business is not what you think it is.

01 Multi-dimensionality in the eyes of data analysis

For data analysts, multi-dimensionality often refers to the split dimensions of data indicators. Let's take a simple example: sales in March were 300 million. This is an indicator without split dimensions. If the classification dimension is added, the effect is as follows:

Note: Compared with just looking at the total amount, using multiple dimensions to break down the data can locate the data more accurately. There are two common methods: 1. Add process indicators; 2. Add classification dimensions according to business management methods. For example, if we only look at the total sales amount, we find that it is 30 million short of the target, but we don’t know why it is not up to standard. At this time, if we break it down a little more, for example:

1. Add classification dimension: see which business line is not doing well (as shown below)

2. Add process indicators: See which link has problems from user intention to payment (as shown below)

By adding process indicators + classification dimensions, we can locate the problem more accurately. Even some simple conclusions are ready to come out. Because of this, many data analysts directly understand the "multi-dimensionality" mentioned by the business as "many dimensions". When they hear that they need to do analysis, they shout "Tear down! Tear down! Tear down!" and make a lot of cross tables layer by layer, and produce the data of each classification dimension (as shown below).

However, is “more” alone enough?

02 Multi-dimensionality from the perspective of business

The "multi-dimensional" mentioned by the business is not what this means. What the business thinks about is not the table structure in the database, but specific problems. When the business sees "Sales in March did not meet the target", the multi-dimensionality in their mind is as follows:

Are you stunned?

You will find that simply breaking down the data cannot answer the above questions at all. Yes, it cannot answer any of them. Even just looking at the data cannot answer these questions.

Even if we define the problem as: "The reason why the performance in March did not meet the target is that the three branches in Region A had too few customers intending to sign contracts", even if we define it to such a detailed level, it still cannot answer the above question.

Because there are too few intentions, is it because the competitors have made efforts, the product is not well made, the activities have not kept up, the user needs have changed... or it is not explained clearly. No specific business questions were answered. Naturally, the business looked confused.

03 Real multi-dimensional analysis, do this

In essence, true multi-dimensional analysis does not test data calculation ability, but strategic ability.

Specifically, there are three aspects:

1. Convert business reasons into data-based arguments.
2. Block excuses: Don’t always try to pass the buck to the business owner; focus on what you can do.
3. Find the crux of the problem: Among a bunch of influencing factors, find the most critical one and focus your efforts on it.

Note that these three things are in order. First, clearly list the data argumentation methods to avoid empty talk (if the data cannot be used to prove the reason, just shut up, this is a very good rule of procedure). Then, block the excuses first. Excuses cannot solve the problem, so block all escape routes first. Finally, focus on finding solutions. When thinking of solutions, start from big to small, from rough to fine, and solve the big problems first. In summary, this can be done in six steps.

The first step is to classify the statements made explicitly and implicitly by the business.

For each type of problem, build analytical hypotheses, convert business reasons into data logic, and let the data speak for itself (as shown in the figure below).

The second step is to prioritize excuses.

Excuses are often generated from macro factors, external factors, and teammate factors. So here, the key is to disprove them. As long as their excuses for running away can be overturned, it will be fine. The best way to disprove them is to give examples. Why can others withstand the rain? Why can other business lines continue to grow when traffic is also difficult to handle? (As shown in the figure below).

Another benefit of the example method is that while refuting excuses, it also points out a way to solve the problem. Business people hate people who only talk about problems but not methods, because everyone can criticize, but it is difficult to solve the problem. Providing specific learning objects can greatly stimulate business thinking of countermeasures, thus achieving a win-win effect.

The third step is to address the white rhino and eliminate the obvious major impacts.

For example, regulatory policies, corporate strategies, major external environments, and other factors do play a major role in business operations, and these factors are something that ordinary employees can only accept and cannot change. However, there are strict requirements for such major factors to be reflected in the data (as shown in the figure below).

So if someone wants to blame it on these factors, it depends on:

1. Did this actually happen?

2. Does the data match the trend?

This is a warning: Don’t blame everything on the bad environment. Wherever you go, the bad environment is everywhere. You are the one who affects the environment! First eliminate the influence (or interference) of this big factor and then focus on what we can do.

The fourth step is to address black swans and eliminate obvious emergencies.

If a real emergency occurs, it is easy to find the source of the problem

Positive: promotional activities, commotion among a certain group of users, new product launches...

Negative: bad weather, emergencies, system downtime...

Therefore, it is easier to explain if we eliminate single sudden problems first and find out the causes, then trace back to the previous situations.

Step 5: Focus on the problem points according to the division of labor and then discuss the details.

After solving the big problems, if you want to discuss more detailed issues, you have to lock in the department, decide on the people first and then discuss the plan. I have shared this before, so I won’t repeat it here.

Step 6: Focus on the details.

Please note that even if we focus on an action of a department, it is still difficult to figure out what business reasons caused the problem.

Because business matters are intertwined with various factors and it is difficult to sort them out, such as:

Content operation: communication channels, themes, tone, style, pictures, and delivery time. Any difference may lead to failure.

Activity operation: target group, activity threshold, reward content, participation rules, any one of them may lead to failure

User operation: If you work hard on new users, the retention rate will be poor. If you work hard on retention rate, the cost of new users will be low. If you work hard on both, the investment on both will not be enough.

Product operation: When selecting products, you consider 100 dimensions, but if the product is not online for a day, you don’t know its real performance. If you put it online, it will fail.

…

The instinctive reaction of data scientists may be: We can do AB testing. In fact, most businesses do not have the time and space to do AB testing, and some things (such as product selection and copywriting) have too many influencing dimensions, and countless sets of AB testing are required to test them clearly. And it is impossible to do AB testing for things that have already happened. Therefore, if you want to distinguish intertwined factors, you need more auxiliary methods to cooperate.