A few days ago, a friend who was engaged in operations complained to me that the user portrait I made after staying up for several nights was called garbage by the boss. Whether it is marketing personnel, operation personnel or product managers, "user portraits" can't be avoided, but I often hear partners complain that this word is too big, and I don't know where to start.
Lao Li summed up a set of user portrait learning methods for everyone, from theory to practice, and taught everyone how to make user portraits.
What is a user persona?
Simply put, user portrait = tagging users. For example, if you pay attention to Lao Li's headlines and read data analysis content every day, then you will be labeled "data analysis", "workplace", etc. The next time you open the headlines, the recommended ones are "How to change career data analysis", "Data analysis essential tools" and other articles. User portraits cover a wide range. The following figure summarizes 7 aspects that you can refer to.
What is the use of doing these user portraits?
The process of making user portraits is to convert unimaginable user behavior data into user tags that can be intuitively recognized.
Although the role of user portraits is far more than these, it is generally inseparable from the following aspects
1. Precision Marketing: Divide user groups into finer granularities through user portraits, and use SMS and email for precision marketing
2. Data analysis: After classifying users according to their attributes and behavioral characteristics, analyze the distribution characteristics of different user portrait groups through data
3. Product application: User tags are the basis of many data products, such as personalized recommendation systems, CRM infrastructure, etc.
4. Data mining: Build recommendation systems, search engines, and advertising systems based on user portraits to improve service accuracy
How can we make a set of user portraits that can be implemented?
Many companies have made user portraits, and may only implement some static tags, mainly based on the basic attributes of users. The entry level is to do some user surveys and telephone interviews. The advanced level is to use some background data to get the ratio of male to female of 7:3, and the ratio of South China to North China is 4:6, which will be scolded by the business side. It is not really used in the actual business, it has value to the business, and it eventually becomes a formalism.
Therefore, in order for user portraits to really play a role, a set of user portraits that can actually be implemented is a necessary condition.
1. Clarify the purpose of the business side to make user portraits
Here, I will first point out the wrong thinking order of most people: it is not because there are user portraits that the portraits are used to improve the business, but the business needs to build user portraits. For example, a content-based community hopes to commercialize the model by launching a knowledge payment module. Based on this, we need to sort out the business goals and problems to be solved, and create user portraits according to the problems to be solved. Only by understanding the purpose of the business side to make user portraits, can the data labels that more meet the needs be selected later.
90% of people do wrong user portraits, what should I do?
2. Collect user data
Common user data can be divided into static data and dynamic data. A 2D user portrait is constructed from static data tags, such as Xiao Ming, male, 25 years old, Beijing; while dynamic data tags + static data tags, a 3D user portrait is constructed. For example, Xiao Ming logged in for 30 minutes on 3.11 and visited a certain page 4 After the second, add the product to the shopping cart, these are dynamic information, which are constantly changing over time.
static data:
The most basic information elements of users, such as name, city, education, registration time, registration method, etc.
dynamic data:
Dynamic data: the changing behavior information of users, opening web pages, adding to shopping carts, purchasing items, etc. are all dynamic behavior data of users.
3. Build a user portrait model
After collecting basic user data, we start with the basic profile of the user, analyze the user's age, region, industry and other dimensions, and model the user portrait.
①Who (user) - which users
Representation of users, easy to distinguish users and locate user information
②when (time) - what time will it happen
The time span and time point of the user's behavior, such as browsing the page for 20s, clicks the button in 5s, and returns in 17s, that is, the time span is 20s, and the time points of the behavior are 5s and 17s respectively.
③where (location) - user behavior touchpoints
The touch points of the user contacting the product, such as which pages the website visits, which buttons are clicked on the APP, how many times are refreshed, or other interactive behaviors.
④what (event) - trigger information point
That is, the content information accessed by the user, such as mainly browsed categories, brands, descriptions, production dates, etc., these contents also generate corresponding tags.
⑤action (action) - user specific behavior
For example, e-commerce users add shopping cart, search, comment, purchase, click like, favorite, etc. The data model of user portrait can be summarized as the following formula: user+time+behavior+touch point, a user is tagged because of what time, place, and what he did.
Different products require different label combinations, and different label combinations form the model of user portraits.
4. Data visualization
After building the model, use BI tools to perform data visualization analysis on the previously generated user portraits, generally analyzing specific groups. For example, core users can be subdivided according to user value, and the potential value space of a certain group can be evaluated to make targeted operation, etc.
Overall customer analysis
and the commonly used data analysis model here is the RFM customer analysis model, using BI tools to calculate the RFM indicators of each customer, here I use FineBI (available at the end of the article), through customer name, consumption time, consumption The amount is used to process the three original fields of the last transaction interval R, transaction frequency F, and transaction amount M, as shown in the following figure.
However, it should be noted that these three indicators are not rigid and unchanged, and should be flexible according to their own industry characteristics. For example, in the financial industry, the time of the last purchase may not be applicable. In this case, you can consider using the holding time of financial products instead. R, which can better reflect the length of time for users to establish contact with financial enterprises.
Dividing users according to three dimensional indicators is equivalent to placing users in the following cube:
Therefore, we need to segment the indicators, set thresholds, that is, set positive and negative values for the indicators, and ensure that the three indicators divide users into eight quadrants. The most commonly used method is to divide into equal frequency and equal width, such as calculating the average value of the user's purchase cost.
However, the average value is only suitable for homogeneous data. For some irregular data, the average value will cause a large error.
At this time, we need to use the function of aggregation. In short, the aggregation function is to divide a bunch of data according to different internal characteristics. The difference between different types of data is generally very large, so that we can find the large amount of data. the "center point" rather than the average point.
In FineBI, we can directly use the aggregation function. The aggregation index selects "Order Amount", "Time", "Number of Times", the aggregation number is "3", and the aggregation method is "Euclidean Distance", so that the final aggregation result can be obtained. Now, we can finally calculate the aggregated R value, aggregated F value and aggregated M value of each customer, which is the reference value we will use.
User classification
After we divide the three indicators separately, and combine them according to the following figure, we can get eight quadrants, representing 8 types of customers:
Finally, FineBI is used to make it into a visual data analysis template, so that we can conduct customer analysis according to the needs.
例如图中的面积图，可以显示出公司各类客户的占比，显而易见一般挽留客户与一般发展客户占据多数，说明该公司的用户结构不是很合理，需要尽快采取措施进行优化；
而右侧的环形图则代表着各种类型客户的购买数量，可以看出复购率越高、愿意花大价钱的客户买的数量越多；
Comment and reply "RFM" to get tools and templates~