Understanding the User Operation System: User Segmentation and Segmentation

User stratification is a division based on the general direction, which is what core goal you want users to work towards, while user grouping is to divide them into finer granularity to improve the effect. The two complement each other.

What is user operation ?

It aims to maximize user value by improving activity, retention rate or payment indicators through various operational means. In the user operation system, there is a classic framework called A AR RR, which stands for new additions, retention , activity, dissemination, and profitability (which has been covered in previous articles).

User stratification

However, going from user activity to profitability are not two simple steps. If users are considered active when they open the product, does that guarantee that the business model will be profitable? An excellent user operation system should be a dynamic evolution.

Evolution is a pyramid -level division of user groups, with the upper and lower levels being dependent on each other.

First, the status of the user group will continue to change. Taking e-commerce as an example, they will register, download, use products, recommend, review, purchase and pay, but they will also log out, uninstall, and churn. From an operational perspective, we will guide users to do what we want them to do (here is to pay), which is called the core goal.

The core goal is certainly not achieved overnight; users have to go through a series of processes.

Not all users will complete the steps as we expect, and each step will show a funnel-shaped conversion . We see the entire process as the evolution of the user group.

The above picture is a typical bottom-up evolution, summarizing the ideal behavior of the user group.

Since the user group is no longer a simple whole, operators can no longer operate in a rough manner with a one-size-fits-all approach, but need to conduct targeted operations based on different groups of people. This is called refined strategy or user stratification.

Its greatest value to operators is the use of different strategies through layering.

• New users : I want them to download the product, and a common strategy is new user benefits;
• Downloading users : I hope they can use the product. At this time, we should use a novice guide to familiarize them with it.
• Active users : I hope to increase their frequency of using the product, so the operations staff must continue to operate, solidify the user's usage habits, and be interested in the product content;
• Interested users : I hope they will make a payment decision and purchase the product. Operations can use different promotional and marketing methods.
• Paying users : These are my target users , and I hope they can maintain this status.

Different user levels adopt different measures. Operations are also limited by resources. When we can only invest limited resources, we often choose the core group, that is, the paying users mentioned above. Because according to the 80/20 rule, only the core group can contribute the greatest value.

A typical example is that in gaming companies, there will be dedicated customer service staff or even dedicated telephone lines to serve RMB players with sweet voices. Ordinary players may use the same automatic reply that never changes.

I believe everyone has understood the stratification, so how should it be divided?

In fact, there is no fixed way to stratify, we can only set up a system that suits local conditions based on the product form. However, it has a central idea: division according to indicators, because indicators are a clearly measurable standard that is far superior to the empirical intuition of operators.

The above picture is a simplified game user stratification, and the indicators of each layer are quantifiable. In order to make the upper and lower levels of users clear, the groups should be as independent as possible. That is, when calculating RMB players, the rich players should be excluded, and when calculating ordinary players, the upper two levels included in the results should be excluded. In this way, the operation can be more targeted.

Operations personnel can then build hierarchical reports based on this, and develop various methods to improve the data based on data trends.

Next, let’s think about what form Zhihu’s user stratification takes? Its core is content production by big Vs ? Or will more users participate in Live to generate revenue? It’s quite difficult to decide. In fact, in many operating systems, user stratification is a two-tier structure.

It targets two complementary cores to form a double pyramid structure.

Under this structure, its core users include both content production influencers and consumer loyalists, who represent two types of operation strategies:

1. Content production direction : In the early stage, we used the invitation system to acquire outstanding talents from various industries, maintained relationships through operations personnel, and encouraged content production. The product mechanism will also encourage big Vs to create and produce better.
2. Content consumption direction : find out the content interests of ordinary users, guide them, and cultivate their payment habits. Increase the exposure of Live, Zhihu , and e-books, and design various coupons to promote user usage.

This type of double pyramid structure brings together content producers and content consumers to form a virtuous cycle for the entire platform: big Vs create content, attracting ordinary people, ordinary people pay for the content , and big Vs earn profits.

Double pyramid structure of user stratification is not uncommon. Take the e-commerce we are familiar with as an example, there are buyers and sellers. The way buyers operate is already well known, but what about sellers? Store opening tutorials, seller university, store decoration, exposure display, store backend, various auxiliary products... Operations also need to help sellers grow, so sellers can also be divided into levels such as ordinary sellers, senior sellers, big customers, and super sponsors.

Is O2O a two-tier structure? Of course. Online refers to users, while offline refers to various offline or service entities. These sellers are more engaged in sales promotion and marketing maintenance, but we can still use the layered thinking to operate. Others include internet celebrities and the public who do live video broadcasts , big Vs and grassroots on Weibo, companies and employees using recruitment apps, and so on.

Different products will have different forms, and different user segments can also be used for different stages of the same product. In the early stage of a product, the goal of user segmentation is to attract more users and KOLs . In the later stage, it will be closer to the business direction, which requires operations to set up flexible segmentation.

For user stratification, generally a four- or five-layer structure is sufficient. Too many layers will become complicated and not suitable for the execution of operational strategies.

User Segmentation

Does the user operation system only have user stratification? Not exactly.

User stratification is a top-down structure, but the user group cannot be fully summarized by structure. Let's think about it simply. We have divided the paying user group based on whether they pay or not. However, there are differences in this group. Some users spend a lot of money, some users buy frequently, and some users have bought but don't buy now. How should we segment them?

If we continue to add more layers, the conditions will become complicated and the business needs will not be met.

Therefore, we use horizontal structured user segmentation. The groups within the same layer can be further divided to meet higher level of refinement needs.

To understand how to understand user segmentation, let’s take the following case as an example.

Men and women show significant differences in consumption-centric products, and they are two different groups. The core goal of grouping is to improve operational effectiveness and maximize the value of operational strategies. In e-commerce products, it is normal to distinguish between men and women, but in tool- type apps, it may not be necessary.

This is what I have always emphasized. Layering and grouping can only be established based on product and operational goals.

Next is the practical application of clustering.

The RFM model is a classic method in customer management. It is used to measure the value and profitability of consumer users and is a typical segmentation.

It relies on three core charging indicators: consumption amount, consumption frequency and the time of the most recent consumption to build a consumption model.

1. Monetary : The amount of consumption is the golden indicator of marketing. The 80/20 rule states that 80% of a company’s revenue comes from 20% of its users. This indicator directly reflects the user’s contribution to the company’s profits.
2. Consumption frequency : Consumption frequency refers to the number of times a user purchases within a limited period of time. The more frequently a user purchases, the higher his loyalty.
3. Recency : measures user churn. The closer the consumption time is to the current user, the easier it is to maintain the relationship with him. The value of a user who made a purchase one year ago is definitely not as good as a user who made a purchase one month ago.

Through these three indicators, we can easily construct a coordinate system to describe the user's consumption level, forming a data cube with three indicators:

In the coordinate system, the two ends of the three coordinate axes represent consumption levels from low to high, and users will fall into the coordinate system according to their consumption levels. When there is enough user data, we can divide it into about eight user groups.

For example, if a user performs well in terms of consumption amount, consumption frequency, and time of most recent consumption, then he is an important value user.

If the last consumption time of an important valuable user is relatively long and he has not consumed again, he will become an important retention user. Because they were once very valuable, we don't want to lose users, so operations and marketing personnel can specifically target this group of people to recall them.

Different quadrants in the figure correspond to different consumer groups. Are you willing to simply treat them as one, or treat them differently based on the group of people?

This is the RFM model, which was once frequently used in traditional industries and can be transplanted for our use in consumer-oriented operation systems. It is not only the core of the CRM system, but also the core of consumer user segmentation.

There are two mainstream grouping methods for the RFM model.

One is to establish indicators and use the indicators as the basis for division, which is similar to user stratification.

The judgment and establishment of indicators require the experience of business experts: what is considered a high consumption frequency, what is considered a low consumption frequency, and how much consumption is considered valuable, all of these are matters of knowledge. And it needs constant revision and improvement.

The above figure is a simplified division. The actual application will be more complicated because the indicators may not be representative. Most of the fee-related data will show a long-tail distribution. 80% of users are concentrated in the low-frequency and low-amount range, while 20% of users generate most of the revenue. This is the difficulty in division.

Indicators are generally divided into descriptive statistical quantiles, such as the median, first quartile, and third quartile.

Another method is to use algorithms to establish user groups through data mining without the need for manual division. The most common algorithm is called KMeans clustering algorithm, and its core idea is "birds of a feather flock together".

We use the data of a company on the Internet to build a Python model. First, we perform dimensionless processing (z-score) and clean up abnormal extreme values.

The three columns of data in the above figure are standardized user consumption data. The closer the value is to 0, the closer it is to the average level. Since the r value is the time of the most recent consumption, the smaller the value, the closer the time is, and the larger the value, the longer the consumption is.

Through the three RFM indicators (called features in machine learning), we first create a visual scatter plot. The figure below is a scatter plot of the most recent charge R and the charge amount M. Each dot represents the charge-related data of a user.

The scatter plot does not yet reveal the pattern of user grouping, and we can only make a preliminary judgment that most of the data show a concentrated trend.

Since the core idea of ​​the KMeans algorithm is "birds of a feather flock together", it uses distance as the objective function. In short, the closer two users are in distance, the more likely they are to be similar, so KMeans finds similar groups, which are called clusters. The greater the distance between clusters, the more independent the user groups are, which is called clustering; the closer the distance within a cluster, the more similar the users are, which is called clustering.

Talk through the charts:

The users marked with red circles are more likely to be similar and belong to the same user group. Because their data on the two indicators R and M are similar, they are all people with low consumption amounts and have made recent consumption.

As for whether it is true or not, let the algorithm solve it. The specific algorithm principles and processes will not be demonstrated. We assume that we can divide user groups into five categories and then see what these groups are like.

The different colors in the picture above are the user groups calculated by the algorithm.

• Red user group : represents high consumption amounts. Because their numbers are small, there is no obvious distinction in the time of their most recent consumption, but it was not long ago. These are the product's fathers and financiers.
• Green user group : represents users who have a tendency to churn. These users do not spend too much, so operations can adopt appropriate recovery strategies.
• Purple user group : represents users who have recently consumed and spent less money. Operations need to tap into their value and develop and cultivate them.

Cyan and blue do not seem to be clearly distinguishable. What if we change the dimension of the scatter plot?

Switching to indicators R and F provides a different perspective. The cyan user group has made more purchases than the blue user group. The blue user group has a lower consumption frequency and needs more incentives. The purple user group has a fairly high consumption frequency.

At this point, the user groups have been clearly differentiated. Can you accurately summarize the characteristics of these users? Although the long-tail form will affect readability to a certain extent in terms of data distribution, operators can still make corresponding operational measures for different groups.

Observe the final results through the scatter plot matrix (the picture may not be clear):

The above is the content of the RFM model. It can dynamically provide users' consumption profiles, and offer a basis for refined operations to marketing, sales, product and operations personnel. This is also one of the applications of data mining in user operations, which everyone should understand.

How to divide groups is a science. If there are too few groups, the distinction will not be obvious; if there are too many groups, there will be no business value. How can you operate more than twenty groups? The number of groups is to strike a balance between data and business.

In summary, there are two methods of grouping: one is to artificially divide user groups through indicators and attributes. The other is to give business significance to the results through data mining. Anyway, the ultimate goal is to improve operational effectiveness and value.

We can use the RFM model and try to broaden our thinking. Can we come up with new ideas? You can definitely try it.

• Finance : investment amount, investment frequency, and time of last investment;
• Live broadcast : duration of live broadcast, last viewing time, and reward amount;
• Content : number of comments, number of words in comments, number of likes on comments;
• Website : number of logins, login duration, and last login time;
• Games : level, game time, game recharge amount.

These are some references I simply listed. They may not be accurate, but they are just for your reference. The segmentation strategies for different products are also different. For example, for hotel products, accommodation is not a solid demand. Is it necessary to add the dimension of time? Maybe the accommodation would be better in separate groups.

It should be noted that the number of groups is not fixed. It can be two or four, depending on business needs, and mainly covers most users. Just don't use too many of them, because firstly it will be complicated, and secondly KMeans clustering doesn't perform well with multiple features.

Through user stratification and user grouping, I believe everyone has understood the cornerstone of the user operation system. User stratification is a division based on the general direction, which is what core goal you want users to work towards, while user grouping is to divide them into finer granularity to improve the effect. The two complement each other.

If the user base reaches a certain size, stratification and grouping may not be a good method, because as the granularity of the user base's attributes expands as the product continues to expand, no matter how it is segmented, it will be difficult to meet the complexity of the user base. This is common in various platform products. At this time, it is necessary to introduce a user profile system. At this time, user stratification and grouping are only part of the profile.

Mobile application product promotion service: APP promotion service Qinggua Media advertising

This article was compiled and published by the author @秦路 (Qinggua Media). Please indicate the author information and source when reprinting! Site Map