The purpose of the enterprise to make label portraits;
Application scenarios and application processes of labels and portraits;
Practical method of constructing label and portrait system;
Securities industry case.
I have contacted customers from all walks of life. In the process of communicating with them and communicating their needs, it is obvious that in the infrastructure and application of data, in addition to data analysis, more and more attention is paid to the application of data assets in more business scenarios. The construction and application of label portrait is one of the very common needs and expectations.
In fact, I think labels and portraits are similar to middle-level system modules in commercial value. Specifically, data assets are essentially data sources obtained by purchasing, but enterprises hope to realize asset realization and continuously expand asset value on the basis of data sources.
In this process, enterprises need to transform data into a product that is really valuable for business output, and then realize the application of upper-level business on these products, such as marketing and personalized recommendation to customers similar to CRM products, and truly turn data into a weapon to realize business value. Many enterprises realize that this middle layer is a label portrait.
So, what is the more specific application purpose of tag portrait construction?
First, the purpose of the enterprise to do label portraits
Although many enterprises will consider different emphases when making labels and portraits, they are all abstractly analyzed and can be divided into the following categories (as shown below):
Figure 1 the use of label portraits
Note: Click to enlarge to view HD pictures.
Most customers who are in the exploration stage of labels and portraits will focus on similar customer life cycle management, in-depth development of high-value customers, cross-marketing and other angles in the early stage (such as the left figure 1). The essential reason is that enterprises hope to do better in mining existing customer assets and customer operations.
With the decrease of demographic dividend, the cost of users' acquisition is getting higher and higher, especially for companies with relatively mature businesses such as banks and securities. Although hundreds of millions of users have been deposited, the number of truly active users is not much, and the value of users released is relatively small.
In the past, some personal business service assets of banks needed to reach more than 6 million to enter the category of private banks, so the value of long-tail customer groups was ignored in the business scope. Now enterprises hope to tap the value of this group of people, but due to the limited cost, they can't serve this group of people with financial managers and professional financial services. Banks began to learn from internet financial management and internet operation methods to tap user value.
At the same time, enterprises begin to attach great importance to data, hoping to better serve these customers by minimizing the cost of data and data assets, which is the relatively mainstream demand of the financial industry at this stage.
Another kind of demand is mainly related to personalization (as shown in figure 1 on the right). The reason why the two types of requirements are separated is that the left side of Figure 1 considers several aspects with the idea of grouping, and divides customers into several categories, but it is not refined to provide customized services for a certain customer. On the contrary, the right side of figure 1 will be more personalized, and the overall input cost will be significantly higher than that of figure 1.
For example, personalized recommendation itself consumes more data resources, basic hardware and labor costs than the picture on the left. Therefore, each enterprise has different stages of development, business needs, input-output ratio, input cost and so on. This determines whether the enterprise is mainly on the left side of the above diagram or on the right side of the diagram 1.
In fact, from the early application, we will also recommend customers to pay attention to the left part of the diagram 1 first, because relatively speaking, this part can produce greater marginal value with less investment. When this part reaches the ceiling of business promotion, it can be further added by means of the right side of Figure 1.
In other words, after the way on the left reaches a certain upper limit, enterprises need to use more extreme means to achieve breakthroughs, such as personalized push, personalized recommendation, personalized real-time marketing and so on.
For example, head e-commerce companies have basically realized personalized real-time marketing. When the user is ready to buy a product, it is lost on the payment page, indicating that the customer has the will to place an order, but he has some doubts or forgets to come back after being interrupted. After about ten minutes, the system will basically push the customer's marketing in real time and push the user to place an order.
Of course, if the business develops rapidly, there are clear scenes and sufficient resources, and you want to have both, it is certainly possible.
Second, the application scenario summary of label portrait
Tags and portraits are actually data reprocessing, which can be divided into four application scenarios according to different processing outputs (as shown in Figure 2 below):
Fig. 2 Application Overview of Label Portrait
(1) fine operation
Enterprises are gradually changing from extensive to refined, hoping to cut the user group into finer granularity, supplemented by SMS, push, email, activities and other means, driven by strategies such as care, recovery and encouragement.
(2) User analysis
User portraits are also a necessary supplement to understanding users. After the number of product users is expanded, it needs to be supplemented by user portraits to study, such as what characteristics the new users have and whether the attributes of core users have changed. Tags are essentially descriptions of users, so the processing of tags is equivalent to analyzing the user information close to the business at a deeper level, which will reduce the scenarios of re-running some business analysis and user analysis based on the original data.
Here I emphasize that when using data, I find that many enterprises ignore the thinking about the meaning behind the data, but when using data to explain the business characteristics, enterprises need to analyze the user characteristics represented by the data in a deeper level, because with this thinking, it means that the business of enterprises really serves users from beginning to end, rather than using these businesses as a means to serve users.
Therefore, when enterprises begin to explore the sustainable development of business, it is very important for enterprises to understand and recognize users. For example, when I consult some securities customers, I will first guide enterprises to sort out the existing data and take stock of customer assets from the customer's point of view.
(3) data analysis
Tags can be understood as one of the rules of hierarchical classification of users. After these data are accessed, the data query platform can support richer and deeper analysis and comparison. In addition, the application of data analysis can be more like a concept, that is, the application of products.
(4) product application
User tags are the foundation of many data products, such as advertising system, personalized recommendation system, CRM infrastructure and so on. In fact, the technical requirement of automatic operation is the labeling system.
Third, the typical application scenario of label portrait
Many enterprises will use push to recruit new, active and recalled users. In fact, if an enterprise introduces an external third-party enterprise to support the push business, the third-party enterprise will charge according to the company's DAU or MAU, which means that the enterprise pushes customers 0 times and 5 times a month, and the cost is the same.
On the other hand, when an enterprise pushes a user 10 times, the latter pushes the user better than the accurate push three times (all three times have effects). Therefore, the application of push emphasizes skills very much, which needs to be considered from two aspects: cost and the effect produced by push users.
For example, the purpose of an enterprise is to increase wealth management transactions through push. We can first find the pushed users through two-layer label screening. The first layer is life cycle label screening, such as users in loss period, and the second layer is behavior label screening, such as users who have checked bank wealth management products in the last seven days. The reason why the secondary screening is set up to extract these people is because these people are users who can migrate and potential marketing users of wealth management products.
If the enterprise does not take corresponding measures for these users, these potential customers will be lost, but if the enterprise accurately markets these users, it will be of great value not only to the wealth management business, but also to the overall business.
Figure 3 Typical application scenario
This is an example of a real scene (as shown in Figure 3 above). Students who operate the wealth management business screened out 50,000 people through the above rules, and accurately pushed these people. After the push, the number of people who opened the App reached 5680, and the opening rate reached 1 1.4% (the data was processed to some extent).
In fact, it is very good that the financial push opening rate can reach 5%. However, by judging the characteristics and needs of these target users and adopting directional push, the push opening rate can reach nearly three times, which means that if the enterprise adopts inaccurate push mode, the user group to be covered needs three times that of accurate push.
At the same time, adopting inaccurate push mode will have two main effects:
First, the push qualifications of those users who are pushed ineffectively are occupied;
Second, it is likely that other businesses can accurately push this group of people and produce better results.
Therefore, the use of precision push is to improve the overall efficiency, but many enterprises rarely account for the effectiveness of input resources in specific business, which is also one of the necessities for enterprises to do refined operations.
Fourth, the typical scene flow of tagged portraits
Figure 4 Typical Scenario Flow
So, how is the example we just talked about generally operated and realized in the actual business process?
Specifically, we can be divided into four steps. First of all, according to the established goal, determine the attribute description of a crowd, which actually corresponds to the business strategy of the enterprise.
For example, the goal is to increase the transaction volume of wealth management. I take the growth of wealth management transaction volume from lost users as one of the strategies, and define it as potential customer marketing for lost users. At this time, the crowd strategy will be vividly portrayed. Crowd strategy is described as dealing with high-value customers. Non-financial users with financial intentions can choose corresponding labels from all dimensions, quickly select the list through labels, then reach users accurately, and finally evaluate the marketing effect. This whole process is also a common idea in personalized marketing or group marketing.
Through the previous introduction, we know that labels and portraits play an important role in making enterprises bigger and stronger, but now many enterprises say that they have made a user portrait system, which may only realize some static labels, focus on the basic attributes of users, or make a tall user portrait report, but it is not connected with the business system, and it is not really used in actual business, which is of no value to business. Therefore, many enterprises make labels and portraits with good intentions, but they have become formalism.
Therefore, without label portraits, enterprises can also drive business and achieve growth. There is a big gap between "doing" and "doing well" and "having" and "using" label portraits. In order to drive business and achieve growth, enterprises need label portraits, not label portraits for the sake of having label portraits, and cannot put the cart before the horse.
Therefore, the correct steps from the establishment to the application of the case tag portrait can be summarized in the following figure:
Figure 5 determines business goals and sets them.
Figure 6 defines the characteristics of the target population.
Fig. 7 Extraction of Label and Attribute Value Definition
Figure 8 Effect evaluation
Five, the four keys to establish a complete labeling system
To establish a complete label system, we need to pay attention to four points: understanding the acquisition form of labels; Clear business forms, list business purposes and collect labels; Classify and define tag pools; Label maintenance.
Below I will expand one by one:
Fig. 9 How to build a complete labeling system?
1. Learn how to get a label.
Figure 10 Understanding the acquisition form of labels
First of all, we all know that there are many kinds of labels, but judging from its implementation rules, they can be roughly divided into the following categories:
(1) Tags based on statistical classes
As the name implies, this kind of tag can be used for user registration, user access and consumption data statistics, and it is the most basic tag type. For example, fields such as gender, city, App usage duration, weekly average startup times, and monthly average consumption amount form the basis of user portraits.
(2) Tags based on rule classes
This label is generated based on user behavior and certain rules. In the actual tag development process, the rules of such tags are determined by the operators and data personnel through consultation. For example, the number of transactions within 90 days from now is greater than 3, which is the definition and caliber of the "active transaction" label; The segment > 20 in continuous 12 months is the definition and caliber of the "frequent flyer" label.
(3) Tags based on mining classes
This kind of tags is a probability model, and the probability is a numerical value between 0 and 1, which needs algorithm mining to generate. For example, judge whether a user is a man or a woman according to his behavior habits, and judge his preference for a certain commodity according to his consumption habits.
It should be noted that the combing of data sources and the application of basic rules are the prerequisites for the application of mining class tags. If the data source of the enterprise has not formulated rules and laid a good foundation, you can temporarily ignore the mining class label. Because the establishment of rules seems to be building a small ladder, it is more reasonable for you to put forward the first step if the ladder is built and proves that your data quality and application space exist.
Because it will involve the input of cost and the application ability of personnel, it is a gradual process, including the gradual process when we serve customers.
In addition, the user's natural attributes, user transaction data, user asset data, user behavior characteristics, and third-party source data in Figure 10 are all tag classification based on data sources or specific business scenarios. In fact, the labels finally presented are generally from a business perspective, and labels are related to application scenarios and sources of statistical attributes.
This is done because such a tag is defined from the user's point of view, so the user can know what the tag stands for, not how it is extracted. So when we really organize labels for customers, most of them are from the user's point of view.
2. Clear the business form, set out for the purpose of business, and collect labels.
(1) resume business process
Figure 1 1 Resume business process
User portraits are first based on business models. The business department has not even thought about the business model, and the data department can only cook without rice. However, the data department also needs to pay attention not to build a car behind closed doors. This is actually the same as making a product. Even the needs of users are not fully understood. In a hurry, an APP goes online, and the result is often ignored. Therefore, the first step for enterprises to create labels is to be familiar with the business.
In fact, when we consult customers, some customers will think that we have not done his business at first, and we are not clear about its business form, so we can't accurately sort out the labeling system. However, they will soon give up the idea. Because our consultants will have a systematic, methodological and migratory ability.
For example, I was engaged in the game industry before I entered the mutual gold industry, but within two months, my experience of mutual gold products and my grasp of this industry can reach a level that I can't reach in this industry for three years.
Why?
I will try all the products in the industry, make real investment, sort out their product experience and marketing strategy, and use the system and methodology I learned before to adjust this application scenario in this process. In fact, you will find that real consultation and users' own understanding of the business are mutually matching processes. A business person is probably familiar with this business, but he is not familiar with how to systematize this set of things and deliver them to others for application. So this is also the value that enterprises need to refer to.
(2) Clear business purpose
Figure 12 illustrates the business purpose
This step needs to be clear about the purpose of the label. Every company, even every operation, has a very different labeling system. For example, if an enterprise wants to make personalized recommendations, it will be more valuable to label things or people's hobbies, but if an enterprise wants to operate users, it will be more valuable to label the retention and activity of users. Therefore, the establishment of the label system is ultimately closely related to the business purpose of the enterprise.
(3) collecting labels
Figure 13 Pushing labels from policies
Regarding the aggregation of tags, it is necessary to combine the operation strategy and application scenarios of enterprises, so that the definition of user groups can be attributed to atomic tags. Considering which data sources are involved and the discrimination of label assignment, these will eventually become the sorting principles of enterprise label architecture and label data sources.
(4) Classification and definition of labels
Figure 14 Classification of Labels
The above picture is a general label architecture. Although not necessarily suitable for every enterprise, it is a good reference frame when sorting out labels and portraits. Through this framework, enterprises can sort out business scenarios and goals, and then reverse design labels. However, the design and application of tags must be based on the understanding of business and architecture.
This framework actually applies the concept of classification, which will involve tag generation rules, tag hierarchy and attribute parameters, which will be marked in the system. Ce Shen data is consulting customers, and ultimately it is given to customers in combination with application scenarios, such as which level and which business system the label belongs to, which is basically the most direct label system for people with specific applications.
Figure 15 Definition of Label
(5) Maintenance of labels
The maintenance of labels is often the focus that is easily overlooked. In fact, labels also have a life cycle, from requirements to generation, to approval, to implementation.
For many enterprises, it is not difficult to generate a label. Many of our customers have sorted out three or four hundred or thousands of labels themselves, but after these labels are generated, there is no clear update rule. Update rules include: label update cycle, such as real-time update and monthly update; Label update dimension, under what circumstances the update of a specific user is triggered, such as under what circumstances the risk rating of a certain type of user is updated; Tag update permissions, such as who can update this tag library; Eliminate useless tags, for example, only 60 tags will be used in the tag library, but there are 90 tags in the tag library, many of which occupy resources. Therefore, label maintenance is a very important system engineering.
But many enterprises didn't realize this, or realized it, but in the end it went away.
The above are the key points to build a labeling system. In fact, a good label design should have the following characteristics:
Figure 17 Label Design Features
I also want to emphasize that it is very important for enterprises to reverse design labels with business requirements instead of generating labels with any data they own. The essential difference between these two ideas is whether the enterprise is goal-oriented or system-oriented. If they are only system-oriented, it is difficult to make valuable products. Therefore, we attach great importance to the interaction between business departments and technical departments. Label generation is not only related to IT departments, but also has a strong correlation with business usage scenarios.
I made a summary of the construction of the label system, as shown below:
Figure 18 Complete labeling system establishment process
Six, the case of the securities industry
Figure 19 Implementation Process of Project Management
The above is a schematic diagram of the realization process of Ce Shen Label Portrait Products. As shown in the above figure, we can find that our overall project management is very detailed, including seven steps: project preparation, system deployment, labeling requirements sequencing, continuous product delivery, labeling system implementation, delivery and training, and post-delivery support.
1. Demand survey: sort out business scenarios, operation strategies and requirements.
Figure 20 Demand Survey
Specific to the demand research stage, the above picture is a demand combing framework that we initially established according to the customer's business scenario, operation strategy and demand. Because the purpose of designing the label system is to do user operation, it will sort out the business system of the enterprise from the dimensions of new customer cultivation, active retention, transaction promotion, capital retention, sinking awakening, loss prevention, lost recall, user experience and major customer operation, but if the purpose of designing the label system of the enterprise is to make personalized recommendation, the design idea is completely different.
Therefore, before sorting out the labeling system for each enterprise, it is necessary to clarify the objectives, sort out a basic business demand framework according to the objectives, and then supplement the information on the framework through preliminary research, such as interviews and data search.
2. Extract label requirements from business requirements.
After defining the business requirements of the enterprise, we sort out the label requirements according to the business requirements of the enterprise, and sort out the corresponding label strategies, and finally formulate the corresponding labels, which are the labels that the enterprise will use in the final scenario application.
Figure 2 1 demand sequencing
As shown above, we finally classified the labels of securities customers into several categories.
The first category is the basic information of users.
Basic information includes user identification, activation information (important information during drainage or innovation, such as when users will come, what channels, etc.) and other information. ), risk characteristics (used more in the financial industry) and so on.
The second category is the user's account characteristics.
Because the users of securities companies have multiple sets of fund accounts, no matter which fund account the users operate, the ultimate marketing of the enterprise is the users themselves, and the most important information is about the users' own characteristics and preferences, so we will make a layer design for the users' accounts.
The third category is business characteristics.
The reason why we divide transactions, wealth management, information and services into four categories is related to the operating system of enterprises. They divide specific operations according to sectors.
Therefore, in order to establish a label that truly represents users in a specific business scenario, we will split the business layer, and the label corresponding to the business layer represents the characteristics of the business itself. For example, the label of wealth management is completely different from the transaction label, and the stock label will involve individual stock preference, but wealth management products actually have a weak preference for specific products and pay more attention to product types. There are also active functions and value labels that are easy to understand, so I won't go into details.