Ten pictures, detailing how to do user segmentation?

A classmate asked: Our leader asked us to do user segmentation, but no matter how we segmented it, it seemed that there was no scientific logic behind it, and we were often criticized: "What's the point of your segmentation?" Are there any standard layering specifications?

Before answering this question, let’s first look at a simple question. Suppose a business revenue situation is as follows, how would you interpret the data? (Total revenue = number of paying users * average payment per paying user)

1. General role of user segmentation

In essence, user stratification is a special form of user segmentation: segmentation based on user value. Those at the upper level are high-value users, and those at the lower level are low-value users. The biggest benefit of user stratification is to eliminate averages.

Looking at the averages, we can come to this conclusion. But in reality, if the average value is reduced by 25 yuan, does it mean that the business has to find a way to increase it by 25 yuan?

——Of course not. Because we don’t understand the composition of these more than 10,000 users. If we tell you that these 10,000 people are composed of the following two forms, would you still think that raising the per capita price by 25 yuan would be enough?

This is a visual demonstration of the power of user segmentation. We will find that the trend observed through the average is correct, but the derived execution plan is often wrong. When it comes to execution, user segmentation is needed to more easily identify the real problems and develop feasible plans.

2. The special role of user stratification

User stratification also has a special function, which is: the products/services/experiences that an enterprise provides to high, medium and low-end users are limited. There are often fixed high, medium and low-end packages, high-end/standard/low-end products, and advanced/intermediate/elementary VIP services.

When we observe users separately in different levels, it is easy to see intuitively whether there is something wrong with the products/services/experience we provide and which level of customers we are losing. This kind of analysis is very directional. It can quickly locate problems and help operations find breakthroughs.

For example, the AB form in the above example has been simplified for the sake of convenience, but it represents two very classic business forms, the large R type business that relies on wealthy users and the large DAU type business that relies on a large number of ordinary users. (Big R and Big DAU are terms in the gaming industry. These two forms are the most clearly distinguished in the gaming industry, so they will be directly applied here). Their common user stratification forms are as follows:

Knowing this, and looking back at the AB patterns, we can more accurately locate the problem:

After this level of interpretation, isn’t it a more thorough analysis than just looking at an average and then saying, “The average order value is too low, we need to raise it!” This is the further role of user segmentation: by combining the segmentation of business behaviors, business problems can be quickly located.

3. Common Mistakes in User Segmentation

After looking at the examples, some students may say: "It seems that the stratification is very simple. User value, paid ≥ active ≥ registered, can't I just stack a pyramid (as shown below)? I see that this is how it is stacked online."

A: This is a common mistake in user segmentation: lack of focus. Remember, the purpose of user segmentation is to quickly identify problems. If you simply stack registration, activity, and payment into a pyramid like the picture above, the graphic looks impressive, but in essence, isn’t it just the three indicators of user volume, activity rate, and conversion rate repeated in the form of a pyramid chart?

The meaning of the graph and the report are the same, so making an extra weird graph is like taking off your pants to fart. This is also the origin of the complaint at the beginning: "What's the point of your score!"

Some students may say: "I see examples, that is, one dimension is cut into several segments, so I will look at the dimensions of payment and activity, and make a picture similar to the following: The imagined layered effect, isn't it good?"

A: This is another common mistake in user segmentation: dimension crossing. It is common to see overlaps between user payment and activity indicators. When the classification dimensions intersect with each other and one layer of users contains another layer, it becomes difficult to interpret. At this time, it is better to directly make a matrix classification to see it more clearly.

In summary, the reason why user segmentation is often superficial is largely because: the students doing the analysis lack the awareness of detailed thinking, and excessively pursue drawing a layered diagram to appear to be thinking comprehensively, ignoring the role of this diagram in the business and ignoring the fact that the business itself has different needs at different stages.

4. Basic Ideas of User Segmentation

It is actually very simple to do user segmentation. As shown in the figure below, it only requires two things: classification dimension + classification standard.

We have said before: The biggest user of user segmentation is to quickly locate problems and provide business breakthroughs. To achieve these two goals, you need to do this (as shown in the figure below):

● Classification dimension is the key issue of current business

● Classification standards are directly related to business actions

The key issues and business actions for business development are not completely whimsical or "following the instructions of the leadership", but are highly related to the life cycle of product/business development.

Every time when I talk about "We need to understand the current key business issues", some students will say: I will just go and ask. Well, direct communication is a good thing, but if you don’t understand anything, not only will you be too lazy to share the business, but you may also not understand a few words you say occasionally. It is even possible that the operators themselves are confused and do whatever the leader tells them to do without using their brains. Therefore, students who do analysis still need to have some understanding of it.

5. Select classification dimensions based on development stage

Usually, the launch of a product/business will go through five stages (as shown in the figure below). The core indicators and key issues to focus on in each stage are different.

Traditional companies will choose to sell off products at the end of their product life cycle and wait for the next generation of products to be launched. Internet companies are more likely to make multiple iterations. At different stages, the business concerns will be different, as shown in the following figure:

With these foundations, we can make preliminary judgments about the current situation and communicate with business more smoothly. By determining the key issues of current concern, you can lock in the classification dimensions and then look at the classification standards below.

6. Set classification standards based on business actions

Under these three restrictions, operators often choose a blockbuster strategy, using a blockbuster product/competitive service/high-quality customer experience to attract users and achieve their goals. Often when users are in the novice stage, there is an entry-level product; during the growth stage, special preferential rewards will be set up at a certain point. These nodes become natural classification standards.

In this regard, traditional companies are doing better than Internet companies. Traditional companies rely on gross profits from selling products to survive, so they have clear definitions of how much feedback to give to customers.

Generally, a fixed proportion of the gross profit is allocated as feedback, and then the proportion of competitors is referred to to choose the main gear to form a competitive advantage. The corresponding classification standards can also be directly applied to business standards (as shown in the figure below).

Under the guidance of this stratification standard, it is easy to find the corresponding problems according to the changes in stratified data. Just like the effect in the example at the beginning, when we see that there are fewer users of a certain level, we immediately realize that we are looking for the wrong people and we need to review the competitiveness of our product. This will provide clues for further in-depth analysis.

On the contrary, most Internet companies, except for a few leading ones, are still at the stage of working hard and burning money for subsidies. There is a lack of clear product line planning and competitive strategy. As long as there is money in operation, they will issue coupons to the utmost, expand the scale and go public to raise money. Therefore, Internet companies often cultivate the kind of mindless operation: if there are not enough registrations, issue coupons; if there is not enough activity, issue coupons; if there is not enough retention, issue coupons.

If you are a data analyst, you will find that your company's operations are really brainless. You just look at which AARRR indicator drops and then organize some short-term activities. There is no overall planning and no strategy at all.

Then you can try to refer to the situation of competing products and do a competitive product analysis to clearly distinguish the differences between your product and competing products among users with different consumption/activity levels, and help them see: we actually have an advantage in XX level and a disadvantage in XX level.

Therefore, we can develop user segmentation strategies to further optimize the system. After all, we are engaged in operations, not sports.

VII. Summary

User segmentation may seem simple, but if we explore it in depth, we will find that there are many business details involved. Many students are troubled by this. They are all troubled by the following problem: My leader asked me to divide the users into high-end ones. Is 8,000 considered high-end, 10,000 considered high-end, or 12,000 considered high-end? Yes, he was struggling to death over this line.

He knows nothing! He is still hoping that there will be a machine learning algorithm that will calculate and tell him: "Artificial intelligence AlphaGo tells you that 10,000 is the perfect standard. Anyone who disagrees will be bitten to death by AlphaGo."

This is going in the opposite direction. User stratification is the result of business operations and is applied to business operations. Therefore, seemingly simple stratification is easy to explain and easy to locate problems as long as it is combined with the business. Let’s share this with everyone.

Author: Down-to-earth Academy

Source: Down-to-earth Academy