Classical method of data analysis: structural analysis

Do you feel that the data we see every day is useless to ordinary people like us? We want to see it but cannot understand it. That is because we only have data but no method to interpret it! This article will introduce you to a classic method to help us understand the data. If you need it, you can take a look.

Do you feel that the daily, weekly, monthly, quarterly and annual reports you prepare are useless?

I update the data numbly every day, but no one reads it. Do I have to get the numbers temporarily when I need to?

Because there is only data but no method to interpret the data!

Numbers are valuable only when their meaning is understood. Structural analysis is a simple and quick way to interpret data, and it is also the ancestral craft of data analysts.

1. Let’s start with the average number that everyone hates

People naturally hate averages, and always think that using averages is nonsense, like "I am the average height of Yao Ming, what's the use of that?" But here's the question: Why is it that averages are so hard to use, but people still like to use them?

Because: convenient.

Let’s take the simplest example:

Known 1: The customer needs 50,000 products

Known 2: The production line produces an average of 5,000 pieces per day

Q: How many days does it take to produce?

Answer: 10 days

Although the actual number may be 4,879 pieces per day, it doesn’t matter. This deviation does not affect the overall decision. It is very convenient to use the average to calculate quantity, arrange manpower and material resources, and allocate resources to estimate progress. Especially on the supply chain side, production capacity, material consumption, and delivery time are relatively fixed and do not fluctuate much. Therefore, it is very useful.

But on the marketing side, the situation has changed.

The 80/20 rule is more popular on the marketing side:

• 20% of customers contribute 80% of revenue
• 20% of products create 80% of profits
• 20% of sales win 80% of orders

This situation is very common.

Therefore, when the business data is very balanced and stable, you can directly use the average. When the distribution of business data varies greatly, you have to look at the distribution structure. This is the origin of the structural analysis method: de-averaging and discovering internal problems (as shown in the figure below).

2. How to do structural analysis

Structural analysis approach:

Step 1: Identify the target. The target of the structure to be observed is the user, commodity, channel, product, etc. Select the target well.

Step 2: Find the indicator. For example, if you have selected the user to observe, then make it clear which indicator you want to observe: user payment, activity, registration time, regional analysis, etc. Here, you must avoid trying to cover everything. If it is too complicated, not only will it be troublesome to extract data, but people who want to see the data will also have no interest in reading it. It is best to focus on the core indicators.

Step 3: Observe in layers. When it comes to observation, students who work with data tend to think of box plots. Yes, box plots are a way to observe structure - but the business department can't understand it. If you want to free up manpower, the more intuitive the method, the better. Therefore, it is recommended to use the following two methods, which can be understood at a glance. Summarize the form. Take the user structure as an example. Are our users of the large R type or the large DAU type?

Step 4: Directly determine the direction of subsequent operations: whether to continue panning for a large amount of sand, or to simmer over low heat and serve the majority of people. At the same time, if there are already business benchmarks, the structure of the benchmarks can be used as a basis for judgment.

• We hope to develop high-end users, so the proportion of Class A users must reach 20%.
• In our benchmark store, Class A users account for 20%, so other stores must also achieve 20%!
• Our business model is to eliminate the weak with the strong, so the proportion of Class A users in each batch of new users must be 20%.

This simple and direct judgment is very useful for improving the usage rate of daily reports!

3. How to use structural analysis method

Use scenario 1: Quickly understand unfamiliar users/products/channels.

• For example, physical enterprises can visit stores in a certain place to see the customer structure and the proportion of goods on display.
• For example, Internet companies look at the time distribution of channel traffic and the number of conversions.

Only a few indicators are used to quickly infer the overall situation and identify problems, which makes the work efficiency very high.

Use scenario 2: Monitoring changes in user/product/channel health

• For a stable business, nothing happening is a good thing, and a stable structure means there are not many problems.
• For new businesses, the sooner they get close to the ideal structure, the better the development.

With clear standards, we won’t be bothered by short-term fluctuations and can focus on doing big things.

Use scenario 3: Monitoring the effects of major policy launches

• Internal observation: Is the policy playing a positive role?
• External observation: whether it triggers fluctuations in key groups

In this way, you can use regular reports to quickly identify problem points, and you don't have to drag out a bunch of irrelevant dimensions and cross-cross them again and again. This allows you to respond faster in emergencies.

The above work can be done completely by regular reports, without the need for temporary data collection and without taking up extra time of data analysts. Therefore, it is very easy to use.

IV. Summary

Similar methods include matrix analysis, trend analysis, and funnel analysis. The common point of these methods is: using a set of logical indicators to establish clear benchmarks, monitor business changes over a long period of time, and quickly draw conclusions. These methods combined with reports can greatly improve work efficiency. Business departments can locate problems, and data analysts can free up workloads, so that they can do more in-depth analysis.

Interestingly, the popular practices on the market now are:

Q: How do you analyze a 30% drop in DAU?

Answer: If we break it down from the dimensions of basic characteristics, source channels, DOU/DNU, etc., you will definitely find that there is a big difference between two columns, and that is it.

This approach is very problematic.

• Only look at the time point, not the trend
• Only look at major changes, not subtle developments
• Don't find the key dimension, try one by one with your eyes closed
• No fixed indicators, no detailed indicators

This will only increase a lot of ineffective work and make the business develop the bad habit of "not looking at the reports, but only making phone calls".

• "Hey, why is it going up again?"
• "Hey, why did you fall again?"
• “Hey, why did it go up and then down?”
• “Hey, why does it go up and down again?”
• …

I am busy dealing with such trivial issues every day, so I can forget about any in-depth analysis or modeling.

Therefore, if you want to free up your energy, you have to build a good monitoring system, make more use of this short, quick and small method, and make more use of the fixed indicator system and search mechanism, so that it can operate efficiently.

Author: Down-to-earth Teacher Chen

WeChat public account: Down-to-earth Teacher Chen (ID: gh_abf29df6ada8)