How to build a user portrait system? Take the insurance industry as an example
This article was published in the community by the author PM He Xiaoze
The essential purpose of big data technology is to mine valuable information from massive data, and it is also the most widely used technology in the field of insurance technology. This article mainly introduces the user profile system, one of the application of big data technology, as well as the construction ideas and specific applications in the insurance industry.
01 Pain Points in the Insurance Industry
Before talking about the specific implementation of big data in the insurance industry, let's take a look at the pain points of this industry in order to understand how the technology is applied.
As for insurance companies, they do not understand users and operate extensively. Many of their products are standardized gift packages, resulting in expensive products, inaccurate sales during sales, and not necessarily meeting user needs. Insurance companies also have low sales efficiency and high overall costs.
On the user side, the purchase process experience is poor. After being misled by sales, they feel cheated when they buy products that are not suitable for them. It is difficult to settle claims and have low trust in insurance.
Although insurance demand is low-frequency for users, it is just needed for most people. At present, the average insurance depth (premium/GDP) of countries in the world is 7%, Japan 9%, the United States 11%, Hong Kong, China 19%, Taiwan, China 19%, and China as the world’s second largest economy, the mainland’s insurance depth It is only 4%, which is only more than half of the world average, which is far behind the developed countries, and there is still much room for improvement.
The main reason is that the overall national financial management and risk awareness are still in the process of popularization, and the individual needs of different groups of users for insurance have not yet been met.
On the one hand, industry problems need to be solved urgently, on the other hand, the continuous progress of information technology. Together, insurance technology has become one of the important driving forces for solving industry problems. Therefore, the insurance industry has entered an era of high-speed information transformation.
02 The main technology and implementation links of insurance technology
As a product, insurance, from production to sales to use (claims), is mainly divided into the following four major links from the business perspective:
Product design, marketing, underwriting, and claims services
The technologies used by insurance technology are mainly divided into the following five categories:
Cloud computing, big data, artificial intelligence, internet of things, blockchain
The combination of technology and business links will derive different specific landing projects, as follows:
Source: 2020 China Insurance Technology Industry Research Report-iResearch
Source: 2020 Insurance Innovation and Development Research Report-Yiou
This article focuses on the specific implementation of big data technology. From the above two industry reports, it can be seen that among the technologies currently implemented in insurance technology, big data is the most mature and the most widely used, and it can be applied in all aspects of insurance business. .
In the past, the insurance industry had limited ways to obtain user data. It was only possible to understand user needs through partial questionnaire surveys. The pricing of products was mostly based on relatively general data reports such as the national average life expectancy.
However, Internet insurance has been greatly improved due to the way of contacting users, and the understanding of users has also become more in-depth.
Insurtech’s big data application is to organize these scattered user-related data, correspondingly carry out more refined insurance product pricing design, and more accurate matching with the actual needs of users, thus realizing insurance companies’ Cost reduction and efficiency enhancement also provide users with products that truly meet their needs.
03
Ideas for building a user portrait system in the insurance industry
The user portrait system is the underlying basic system for the application of big data. To fully understand the users, differentiated design of products for different user groups, refined operation of marketing links, and personalized recommendations, it must be based on the label data of the portrait system . In addition, the unique business characteristics of the insurance industry also need to be considered.
Regarding the idea of building a user portrait system in the insurance industry, I personally sorted out the following directions:
1. Analysis of the status quo of enterprise data
Whether it is necessary to start building a big data platform and to what extent depends on the current degree of dataization of the enterprise. For how to evaluate the degree of dataization of an enterprise, you can refer to the data management capability maturity model DMM, which is divided into 5 levels. If you reach the second level, you need to start building a standardized data platform.
However, this model is a more general model in the data industry. For the Internet insurance company I work for, I personally evaluated it from the following specific directions.
1) Does the company have enough products and services?
2) Is the user's attribute hierarchy diverse and complex?
3) Has the volume of enterprise users reached a certain scale?
4) Does the company have obvious data island effects, and it is difficult to unify data from multiple business lines?
5) Is there much room for improvement in corporate sales efficiency?
The answers to the above questions are all yes, so building a big data basic system such as user portraits is more in line with the needs of the current company's development stage.
2. Business characteristics of the insurance industry
The user portrait system is actually a relatively common underlying system for big data practice. It is already very mature in the Internet industry. It has already been implemented for many e-commerce, video, and social platforms. Many recommendation systems that achieve thousands of people are also based on user portrait tags. data.
But this does not mean that in the Internet insurance industry, you only need to copy the user portrait system. The most important thing is how to integrate with the insurance industry, reduce costs and increase efficiency for enterprises, and provide users with better services and products. Therefore, it is necessary to analyze the business characteristics of the insurance industry first.
The business characteristics of the insurance industry will directly affect the label definition of the user profile system and subsequent applications. I sum it up as the following features:
1) Low frequency, large amount
Compared with e-commerce such as daily necessities, grocery shopping, clothing, and household appliances, insurance is similar to RVs. It is a low-frequency product type with an average single amount of several thousand or tens of thousands of yuan, which is a large expenditure for many users. This determines that insurance sales must first address the first order transaction rate, not repurchase, or user activity.
2) Long decision process
Low frequency and large amounts also lead to a longer decision-making process for users. For example, when buying a critical illness insurance, users generally shop around, hoping to choose the most cost-effective and most suitable product, and there is less impulse consumption. This also determines that many e-commerce companies capture the user's psychology of greed, comparison, and show off, and gamified event marketing is not suitable for the insurance industry.
3) High degree of artificial dependence
Due to the complexity of insurance product terms and responsibilities, many users do not have the ability to self-understand and complete orders. They need the role of an agent to intervene to help users. This determines that the application of personalized recommendation in the insurance industry has great limitations, because no matter how scientific or complex your recommendation model is, users cannot directly trust the results of machine recommendations.
4) Single profit model
Since the pricing of insurance products is standardized (reported by the China Banking and Insurance Regulatory Commission), general e-commerce price promotions cannot be used, and most of the earnings of Internet insurance intermediary platforms are derived from the commission settlement of insurance companies.
3. The focus of user portraits in the Internet insurance industry
Based on the above business characteristics of the insurance industry, when designing the user profile system, it is necessary to consider how to help the business. You can think from the following key directions:
1) Quantification of user behavior data
The premise of insurance sales is to analyze what users really need, and then sell the corresponding products to users smoothly. To understand the insurance needs of users, first quantify the user's behavior, such as product browsing, activity participation, order placement, underwriting, claim settlement and other behavioral data, and then transform it into the user's insurance needs through a complete data indicator system. This is also the advantage of Internet insurance over offline insurance. It is not necessary to directly talk to users for inquiries, and summarize the needs of users through various data records analysis, and use the same set of logic to realize the needs of all users, without manual one-to-one communication. .
2) Improve sales efficiency
The sales efficiency here includes not only the process of users self-understanding the product and completing the insurance, but also the manual intervention to assist the user in understanding the product and completing the process of insuring.
The self-service insurance part, such as the user's existing data, can help the user to quickly complete the underwriting and improve the ordering process experience. Even when the data is abundant enough, automatic underwriting can be realized, making the user's insurance order process as simple and fast as ordering takeout.
The manual sales assistance part uses user feature tags to assist sales staff in understanding customer demand characteristics and insurance intentions, and improve the efficiency of communication and conversion.
3) Improve LTV
The LTV of a user in the insurance industry can be understood as the insurance needs of all family members of this user.
There are four main types of insurance needs according to insurance types: critical illness, medical care, life insurance, and accidents.
Through the data tags of the user portrait, the different insurance needs of each member of the user are quantified, the protection gap of the user is analyzed, and the corresponding gap is filled with the appropriate product through relevant precise activities and marketing, so as to improve the user’s LTV and at the same time. The overall sales premium.
And some focus directions that are not suitable for the Internet insurance industry:
In order to accurately recommend products, establish a complex algorithm model, increase user activity, increase repurchase rate, and product price play.
4. Ideas for dismantling label requirements
The core of the user portrait is the label system. Only when the label is defined can subsequent applications produce good results. When designing user portrait labels based on the above analysis of the business characteristics of the insurance industry, the following two points need to be paid attention to:
-
First of all, we must define the correct business goals, and then gradually disassemble them into labels that can be implemented, so as to avoid the creation of many labels that do not have much effect on the business, and to avoid wasting development resources.
-
When defining labels, try to follow the MECE principle as much as possible to make labels independent of each other and completely exhaustive. Don't let labels with overlapping concepts and relatively obscure borders be generated.
For example, there is a goal called to increase the conversion rate of the first order. The goal can be disassembled according to the following ideas, and then the data label corresponding to the smallest goal dimension can be designed. In this way, these data tags can assist in the refinement of the operation plan when doing related operation activities, and can also evaluate the operation effect by tracking the data changes of the tags.
04
User portrait system composition
The following briefly introduces which parts of a complete user portrait system are mainly composed and what are their functions.
The main structure is as follows:
1. The underlying labeling system
The core of the user portrait is the user tag. After the tag data is available, user analysis of various conditions can be performed.
The labeling system can be understood as a data warehouse, which collects data from various business systems, embedded data systems, log data, external systems, etc. After preprocessing the ETL data, it performs modeling calculations according to the definition of the label to obtain the standard Unified and applicable data labels. Then maintain a certain data synchronization frequency and provide it to users in various forms of data services, such as a visual management background or an API interface.
2. Data service layer
The label management background is the main presentation form of the label system data service, which is mainly divided into label management and user group management.
The role of tag management is to visually manage and operate tags, including adding, deleting, checking, and removing tags. The application can easily view existing tags, tag definitions, and data distribution of tag values.
User group management is a data unit that establishes refined user classification. Filter out a batch of specific users in a certain way, such as free combination through tags or directly import users, and then perform related management, analysis and applications based on this batch of users. For example, setting up exclusive marketing reach, analysis of user tag data, providing data to external calls and so on.
3. Data application layer
The data application layer is to achieve specific business goals through tags or clusters. For example, common single-user portrait analysis, user group portrait analysis, personalized recommendation, precision marketing, etc.
The single-user profile data is actually the tag value of all tags of an account. When a business party wants to know the profile of a user during the application, it only needs to return the tag value he needs.
The user grouping portrait data is a created user grouping label data collection. The label source data can be exported for free analysis, or the grouping data can be made into intuitive icons according to a certain template, which is convenient for business side analysis.
The above three parts are just a brief description. Because there are too many content, and many related articles in the industry have introduced the system in detail from the functional level, I will not repeat them here. The main structure of the entire user portrait system has been organized into the following brain map.
05
Application of User Portrait System in Insurance Industry
After the user portrait system is completed, it needs to be used by the business side and the operation to find problems and iterate continuously to continue to help the business and operation achieve their goals, that is, the data application layer mentioned in the previous section. The application of user portrait mainly includes several directions in the figure below.
Combined with the business characteristics of the Internet insurance industry, the main application methods can be divided into the following categories:
1. Precision marketing
With refined tags, users can be grouped based on different tag combinations, and more accurate content marketing can be performed on user groups with different feature combinations. The content of marketing is customized by the operator, and the marketing method can be SMS, WeChat , Emails, in-app messages, resource popups, etc., so as to improve the effectiveness of marketing.
This application is similar to the user portrait application of most Internet products, and the core is the free combination of user tags.
2. Sales assistance
Due to the complexity of insurance products, Internet insurance still has a large number of orders generated by manual conversion of users. In the process of communication between agents or sales consultants and users, the main characteristics and needs of users are provided to sales consultants in advance through single-user portraits. , Can better help sales to convert.
Such an application method is also suitable for manual intervention processes such as customer service, and the core is to improve the user experience and facilitating efficiency of manual links.
3. Analysis of user portraits in groups
For the characteristics of this batch of users in the user group, the corresponding conclusion can be obtained by analyzing the tag value. In many reports that we commonly see, the common characteristics, age distribution, price distribution, etc. of a group of users are all portraits of the grouped users. Data analysis application.
The easiest way is to support exporting the label value data of user groups, and then do secondary processing analysis to make a chart. The complicated method is to group users into groups directly in the background of user portraits and visually present the data situation by means of icons.
4. Risk control, anti-fraud and automatic underwriting
As a financial product, insurance has an important business link that is anti-fraud, anti-money laundering, and anti-fraud insurance. In traditional offline insurance, insurance companies know little about user information and data, and there are many users who are insured with illnesses, which leads to an increase in the insurance company's loss rate. Therefore, insurance companies will restrict the interception of the risk control and underwriting interface for online insurance, but the insurance company still lacks user data, and the interception rate is still not accurate enough, or the scale is too strict, and many normal users are intercepted.
For insurance intermediary platforms, especially large-scale platforms with a lot of data, they have unparalleled data advantages. Through the underlying portrait system, the user’s label can be calculated very well, which can help insurance companies to control and underwrite the risk, even Automatically match different underwriting methods according to the situation of different users (the price may need to be increased in the case of non-standard underwriting), which can not only improve the user's quick insurance experience, but also enable insurance products to sell for thousands of people.
5. Customization of auxiliary products
The traditional insurance production process is that the insurance company has already set the product form and price, and then finds suitable users for sales. However, if users with similar label data can be gathered through the user portrait system, and the insurance company can be used for customization, then more refined product pricing and design can be achieved. This idea is the same as C2M in e-commerce. .
For example, ZhongAn Insurance has a customized insurance for users of thyroid nodules, "Zhongan Million Medical Excellent Edition", which probably does this.
6. Personalized recommendation
Personalized recommendation is very common in many social entertainment and e-commerce companies. It is realized by thousands of people, constantly optimizing algorithm models based on usage data, and then continuously recommending similar content. The personalized recommendation in the Internet insurance industry is to analyze the user's insurance needs through the label data of the user's portrait, and then match the corresponding insurance type and product to achieve more accurate product recommendation.
However, as mentioned above, the current users' trust in the results of personalized insurance recommendations is very low, and many platforms can recommend a small number of products, and the labor cost to implement this function is also high, so there are few applications in the entire industry.
06
The pits stepped on during the product landing process
For business or operation personnel, it is enough to know the role and application method of the user portrait, but for the product manager who is specifically responsible for the system, he must also have experience in various details. In the specific implementation process, there are often many detailed problems. As a result, the progress of the product project is slow, or the application effect is poor.
Combined with my personal experience from 0 to 1, I mainly summarize the following points worthy of special attention for reference by product peers.
1. The scale of label definition
1) Which dimensions of data are not suitable for labeling
Not a specific data, but a collection of data. For example, product rankings, the 5 most suitable products.
2) What data is not suitable for label storage
Data that changes frequently and has a large amount of data, which can be converted twice when the business is in use. For example, the product name.
2. Re-split the data type to facilitate specific use
In general, the only data types are string, numeric, boolean, and set. But for the convenience of specific use, it can be further divided. For example, the numeric data type can also be split into age, general numeric value, date, etc., so that the combined operation can be more convenient when performing label combination applications.
For example, the date of birth label is generally stored in the form of a birthday timestamp in the table. If the general numeric type is converted to the "age" data type for storage, users with "age=10" can be filtered directly when the label is used; if Converted into a specific year, month, and day of the "date" data type. When the tag is used, users with "birthday date> specific year, month, and day" can be filtered.
3. Definition of label type
Tag types are generally divided into fact tags, model tags, and prediction tags.
The early stage of the system is mainly to define fact labels and model labels, so that the system can run and be used by the business side. It is not necessary to do prediction labels in order to do prediction labels.
Because predicting tags need to combine multiple user characteristics to establish a more complex algorithm model, and even label products, according to user-based collaborative filtering algorithms or commodity-based collaborative filtering algorithms, user preferences can be predicted. This is already the category of recommendation systems, and as mentioned above, the current Internet insurance industry's reliance on personalized recommendation functions is not obvious.
4. Verification of data accuracy
After the data is online, how to verify the accuracy of the data is very important, otherwise the business side cannot use it directly.
There are probably several ways to verify data accuracy:
1) Compare the label data with the data source
Mainly check the completeness and accuracy of the data.
Integrity refers to whether the data is missing. For example, some label data has a value in the data source, but after combined calculation, the label value is empty.
The accuracy is mainly to verify whether the value is consistent with the value of the data source. If it involves a label with a decimal point, pay attention to the rounding rule, whether to round up, round up, or directly round off.
2) User coverage calculation
User coverage refers to the coverage of these tags on the user as a whole, and some individual tags with abnormal coverage should be checked separately.
For example , the label of "gender" has three label values: male, female, and unknown . According to normal logic, the ratio of male to female will not be very different. If the ratio of male to female label value is found to be 1:5, then the data may be calculated error.
3) Business utilization
As the business develops, there will be more and more labels. The business side's use of tags can also reflect the quality of tag data from the side.
If you find that some tags have never been used by the business after the system has been used for a long time, it can be understood that the usage is 0, then this tag either has a definition problem, does not meet the business requirements, and needs to be modified or deleted, or the data is abnormal , The business side is unwilling to use it, and it needs to be checked.
Conclusion
The user portrait system is a basic system for the implementation of big data. With this foundation, the upper-level refined operation, digitization, and intelligent operation can be carried out. The insurance industry is still in the transition process from online to intelligent. Whether it is an insurance company or an insurance intermediary, making good use of the system can better empower business, improve efficiency and user experience, and make the insurance industry progress to a more intelligent stage.