This is peng Wenhua 161 original, by the way to you happy New Year! \
Today is New Year’s Day, I give you a happy New Year! To my good friend:
The magic test,
Diligent study profuse;
As much money as a cow,
Up the bull!
Today’s question comes from Mr. Li in Shanghai, who is doing customer segmentation in the industry. Want some customer classification and clustering related strategy and information. I have a lot of information, but I don’t know if I can give him some good advice.
To be honest, although I have done a lot of customer classification before, but always feel that it is not very good. There are a lot of similar articles on the Internet. Let me share with you my understanding.
In addition, I challenge the Spring Festival not closed, every day to share original articles, welcome to add my personal wechat: Shirenpengwh, join to urge more group, small whip urges me to write quickly
Simple customer segmentation
The core purpose of customer segmentation is to refine the operation. In fact, it is to develop operational strategies for different users to maximize profits.
So the simplest idea of customer segmentation is segmentation. You’ve probably heard the term “high net worth people,” which is one of the most popular customer segments of the traditional marketing era.
In general, user segmentation should follow the MECE principle, and all of the above methods are already MECE. But it’s not absolute. There are exceptions.
Limited by the data and technology available at the time, customer segmentation was mostly done in CRM because it was the only way to get all kinds of information about users. The logic of subdivision is very simple, mostly from a single dimension for segmentation.
For example, “by customer net value, by customer data source, by consumption frequency, by age, by monthly accumulative consumption amount segment” and so on. This kind of customer segmentation method is primitive and rough, far from being called “customer segmentation”. But this is the easiest way for anyone to think about, and the easiest way to understand.
So in the early days of data trading, sellers would label the data source in an attempt to reflect the value of customer data on the name. Of course, now the sale of personal privacy data has entered the criminal law, we must not touch ha.
Business analytical customer segmentation
Further, some people summarize and refine the logic of customer segmentation from various angles, such as the segmentation from the user life cycle. We use different strategies for customers in different life cycles, hoping to prolong the time of users in the mature stage and create more value.
Such as:
According to the user life cycle, such as “potential users, new users, paying users, repurchase users, lost users”, the life cycle of different industries is not exactly the same;
According to the user operation process, such as AARRR, RARRA, “new users, users, interested users, intended users, paying users”, etc.
According to the user points grading (loyalty), such as traditional membership card level, Taobao “crown, diamond”, etc.;
According to the user’s various tag segmentation, this degree of freedom is very large, very rich information. \
These methods are very intuitive, and business units love them the most. And the corresponding strategy is also very clear, basically by name can think of meaning.
These still seem to slice customers from a fixed dimension, but generally these dimensions are understood and processed by the business.
A simple example: The different levels of credit cards are a customer segmentation model with very complex rules. A card can only be upgraded if certain conditions are met. Of course, the rights and interests enjoyed are not the same. There are even people and companies that specialize in the rules of keeping credit cards.
Combined customer segmentation routines
All of the above are methods of customer group segmentation from a single dimension, which we can get along with a lot.
So if we go one step further, how can we break it down? Bingo! In that case, each single attribute can be combined and subdivided, and RFM is a typical one.
This model is very easy to use, spread very widely, high degree of recognition, can explain strong, the corresponding strategy is also very clear.
RFM model is the classic model of user segmentation. It divides users into eight groups based on Recency, Frequency, and Monetary spending.
In fact, RFM is essentially a quadrant model, but instead of 4 quadrants, there are 3 indicators, each of which is separated into 0 and 1, and divided into 8 quadrants in total. And when we use RFM, we can also carry out various variations, such as changing a index, separating “high and low” into “high, medium and low” and so on. But no matter what happens, the essence remains the same.
One of the variants of RFM is to replace the M in RFM, such as RF+ age. Of course, as long as you can think of it, can be any combination, such as monthly consumption level + user age + region. I have done user segmentation in the automobile industry before, and I have done the combination dimension of customer preference + complaint frequency + activity participation, which can distinguish loyal customers from difficult customers.
Again, this approach can be combined almost indefinitely, if you can come up with a combination of two or more key business metrics, instantly you can slice customers into any group.
The segmentation of customer groups in the field of mobile communication has also achieved a mechanism. Various chaotic packages are not dreamt up by operators, but calculated by various algorithms using different groups of data. \
Speaking of algorithms, we can also use a variety of clustering algorithms to achieve customer segmentation. But the result of these algorithms is much less interpretable than all the previous segmentation methods. And some algorithms have random seeds and other factors, the results of each execution is not the same. For example, the k-means clustering, the K value is either artificially set or random. That’s what happens at random. The same data, it is likely to appear the following radically different situation.
This makes it very difficult for us to explain to the business. We have to make it all up. One of us wanted “big data” + “algorithm”, and we had to ensure that the execution was idempotent, that is, the result of each execution had to be the same, or we couldn’t explain it to others. In the case of clustering with random numbers, that’s not good. Finally, we changed the algorithm, so as long as it’s a result set, no matter how it’s executed, it’s a result. Alas, we are too difficult.
Of course, in addition to K-means, we can also use KNN, hierarchical clustering and other clustering algorithms. With the exception of hierarchical clustering, which is slightly easier to explain, the results of other algorithms have to be compiled
Is there a better way to group customers? Of course there is.
Business insight customer segmentation
Maybe a buddy said, I know what you said, it’s not much use to be honest.
This score how to say, if only customer segmentation, we move these. However, if we want to guide the operation, we need to analyze the problems of users at different levels in detail, followed by appropriate operation suggestions. I have to work with operation students to set strategies, make implementation plans, follow up, adjust and optimize.
However, we say back, in addition to these moves, we still have a more classic user segmentation method?
The answer is yes!
But one step further is not a generic approach, at least to the industry and scenario-specific. For example, the FMCG industry has a very classic “Ali eight groups” :
All FMCG industries can refer to these eight categories for customer segmentation. Here you may notice that this is clearly against MECE’s principles. Yes, but these eight categories already cover the majority of the population, and what’s left doesn’t matter.
So how do these eight categories come from? Business insight! There’s no other way.
Of course, some insights are also very interesting, and will be analyzed from the user’s “constellation” perspective, also do not know whether it is confusing. For example, Tencent’s user insight into Xi Tea has such a conclusion:
There are also many business insights that are combined with social realities, namely the legendary PEST model (political politics, economy, social society, technology). Such as:
If you look at the subdivision above, there is no mathematical logic at all. It is also completely inconsistent with the MECE principle, and there is a gap between various parts, but it does not affect its business interpretation. Because it’s mixed with insight into the various phenomena of today’s society.
Jd.com and Nielsen have also done very in-depth research on the user life cycle, and their insights are interesting. Their slogan is “achieve wave growth of brand user equity” :
conclusion
User subdivision is an important means of fine operation.
The simplest idea is to slice users from a fixed attribute, such as customer net worth;
Further, from the perspective of business analysis, such as by user life cycle;
Further, multiple dimensions are combined for segmentation, such as the popular RFM model, or KNN, K-means, hierarchical clustering and other clustering algorithms are used.
One step forward is the special insight of the industry, even the special insight of the vertical field, which requires a deep insight of the industry users, such as Ali’s “eight groups” in the field of FMCG.
I have dozens of user segmentation insights for different industries that you can download for your reference.
As for the technical implementation, in fact, if the amount of data is not large, it can be done with Excel; Some more data volume with relational database, write SQL to fix; Anything more is big data, distributed computing.
In the traditional marketing era, more people use SPSS, SAS and other data mining software; There was a time when R was popular; Python is now in the ascendancy. It’s very simple to use, basically organize the data structure, and then switch packages.
And that’s it for today’s share. Welcome to add my wechat account: ShirenpengWH to discuss big data and data analysis. Share an original content to everyone every day, we learn together and make progress together. \
Extended reading: dozens of industry user segmentation insights report, the public account “big data Architect” back to “user segmentation” can be downloaded. \
Enjoy better with the following articles
Hot article | how to build a data analysis system
Dry goods | using peak experience building the strongest user activated! \
Dry goods | what is called understand the business? Five levels of analysis
Thinking why | said your operations team must have a girl?
Dry goods | monthly salary of 3000 and 30000 data analyst is poor?
Dry goods | how to leverage the power of the organization to build a super win the guest system? \
