How to create the correct user portrait?

Many companies actually do not have a clear user portrait , that is, they themselves do not know who their users are.

What’s worse, they don’t even know what user portraits are.

In fact, students who work in operations and product managers should be familiar with the term "user portrait". They have heard it at least eight hundred times, but it seems that few people really understand what user portraits are and how user portraits are generated.

Next, I will explain in detail the concept and function of user portraits and how to establish correct user portraits. I will use several small examples to help you understand. Please keep them in mind.

1. What is user portrait?

User portraits, also known as user tags, are based on the various behavioral data left by users on the Internet. These data are actively or passively collected and then processed and analyzed to generate tags. For example, male, born in the 1990s, white-collar worker, likes to buy electronic products, has a monthly salary of 15,000, etc.

The content of a user portrait can be very broad; as long as it is an understanding of the user, it can be called a user portrait. But the people you want to identify must be your typical users, who will use your products, services or consume your brand in a similar way.

2. The role of user portraits

Regarding the role of user portraits, I think there are three main points:

1. Precision marketing: I believe this is the reason why many operations students are reading this article. From simple and crude operations to refined operations, user groups are stratified, and then SMS, push, tweets and other means are used to stimulate users to use the product, place orders, or recall users.
2. Advertising: When a brand is in a period of rapid business development, it needs a lot of external exposure. Since the promotion budget is limited, the choice of channels requires more typical users to promote to, and this requires the cooperation of user portraits.
3. Data analysis: This can be understood as building a data warehouse and using various tags to connect various data systems. For example, the previous article "Advanced Operation Strategies for WeChat Official Accounts, I suggest you take a look at it with your product manager" talks about using official accounts for targeted push. Then, establish behavioral tags for users' behaviors on official accounts and e-commerce apps and connect them, then build a data warehouse, and then create user portraits. Only then will the refined push strategy be perfect.

3. How to create a correct user portrait?

Next comes the part that many people are most concerned about: how to build a correct user portrait.

First of all, it needs to be made clear that all user portraits are based on business models. Many students start to create user portraits without even figuring out their own business models, business scenarios and forms, which is basically a waste of effort.

Below I will use a short story to help you understand how to better establish a correct user portrait.

Xiao Ming started his own business and developed an APP to sell various snacks. After operating for half a year, the benefits were very good. However, we now find that performance growth is weak and there is still no improvement after increasing investment in promotion. So they came to me and hoped that I would develop a set of refined operational strategies for them to improve their performance.

After an in-depth communication with Xiao Ming, I found that Xiao Ming’s team had always been operating in a rough manner and had not used data to drive business growth at all. So I plan to help Xiao Ming sort out his user portrait first, and then take the following operational actions.

First, I drew out his most basic business process.

According to this chart, I first divided Xiao Ming's users into 5 categories based on whether they had purchased salad:

• Potential users (registered for the APP but did not place an order)
• New user (purchased once)
• Old users (users who have purchased 2 times or more)
• VIP users (consumption over 200 yuan within 90 days)
• Lost users (future consumption within 90 days)

Here is a magical data to share. As long as a user makes a repeat purchase, that is, purchases twice within a certain period of time, his retention rate will increase by 30%.

For potential users, give new user gift packages to encourage their first purchase;
For new users, you need to encourage them to make a repeat purchase as soon as possible, such as sending them targeted push notifications of other products in the same category as their last purchase, and using red envelopes to encourage them to place orders.
Old users hope to use the rights of VIP users to stimulate them to become old users. For example, VIP users can enjoy priority delivery, receive gifts, participate in tasting activities and other activities.

In this way, several user portraits will emerge.

After that, user tags are added through the user's own attributes.

I asked Xiao Ming to export all the orders of all users and make judgments based on the order addresses. For example, if a user uses the same address to receive goods multiple times, then this address is determined to be a frequently used address. Then, based on whether the frequently used address is a company office or a school, it is determined whether the user is a white-collar worker or a student.

The operating strategies for students and white-collar users will be completely different.

For students, cost-effectiveness may be a priority, and you can focus on recommending some relatively cheap snacks to them. Or when doing some fission activities, you can push them first. We also need to consider the student holidays from July to August and during the Chinese New Year, and the school term. Generally speaking, consumption demand is higher during the school term.

For white-collar workers, cost-effectiveness may not necessarily be the priority factor, and the consumption experience may be more important. In that case, we recommend some imported snacks with better taste, or low-fat snacks that are not easy to make you fat. If the consumption scenario is in the company, we also need to consider the scenarios of unpacking packages and eating snacks. Considering that if customers are seen unpacking packages by other colleagues, they may develop a desire to share, we can set up group purchase discounts or recommend larger sharing packages.

At this point, the user tags have been enriched.

Finally, we predict lost users based on their behavior on the APP.

We found from the data that the main reason for the slowdown in growth was that the user churn rate began to increase. There are many reasons for churn, the most important of which is to find the key factors before the time point when users stop consuming.

for example:

• If a user opens the app less frequently over a period of time, they may churn (the key factor for churn is the reduction in the number of times the app is opened).
• If a user has given a bad review, then he or she may churn (the key factor for churn at this time is that the bad review customers are not maintained in time);
• If the logistics time between the user placing an order and receiving the goods exceeds 7 days, the user may be lost;
• If a user hasn’t logged into the app for more than ten days, he may have churned;
• If the amount of a user's first order is particularly small, he or she may be lost. . .

There are so many reasons. List as many possible reasons as possible, and then use machine learning modeling to make predictions. (Technical issues will not be shared here)

It should be noted that all of these are dynamic, so I set the user's repurchase or re-browsing cycle to 7 days (I thought that after buying snacks, they will definitely be eaten within 7 days). Depending on the business situation, try to divide the time period into smaller pieces to make it easier to analyze. Then, according to the data, predictions are made based on the details of user behavior.

With these judgments, targeted recalls can be carried out at different stages.

According to the user's preference for buying snacks

Snacks are divided into meat, puffed foods, casual snacks, meal replacements, fat-reducing foods, etc. Xiao Ming divides them according to the user's purchase preferences, which can be divided into: those who like to eat meat, those who like puffed foods, etc.

It is divided according to comprehensive factors of consumption model.

In addition, you can also use the RFM model (a public model that measures customer value and customer creativity) to divide them (if you don’t know the RFM model, you can search it on Baidu). This is relatively complicated, and I will write a separate article to explain it later.

After reading this case, I believe you must have some feelings.

Then let's look at this set of underlying production ideas:

• The first step is to determine the business goals;
• The second step is to generate data and produce labels.
• The third step is to analyze data and gain insight into users. Insight is not the ultimate goal, because insight is only to get a conclusion or direction;
• Step 4: Start the application. The application brings new data, thus forming a closed loop of data.

The first step is to determine the business goals

Applications drive demand. Many students make a mistake. When they are doing user portraits, they tend to come up with thousands of tags at a time. In fact, this is of no use, because you can't use so many tags at all, and you will be confused by so many tags.

The second step is to run data and produce labels

Data is the core of everything. It is useless to create a lot of labels without data. If you don’t have data, or don’t have much data, your first job is not to build the subsequent labeling system, but to find data quickly.

The main sources of data are what users fill in when registering, as well as their behavior on the platform, whether it is interactive behavior, browsing and clicking behavior, or consumption behavior. Then, a labeling system is established through these behaviors. Of course, some labels exist objectively, and some labels are predicted based on logic.

For example, if gender is filled in, or obtained through WeChat unionID, this is objective existence; but if this information is not available, then predictions are made based on the user's name, like Wang Xiaohong is basically a female, and Wang Xiaohu is basically a male. Of course, there will be errors in the predictions.

Generally, we will establish a basic user tag system through user attributes and behavior data.

There are usually four categories:

The first category: population attributes. For example, gender, age, permanent residence, native place, and even height and blood type are called population attributes.

The second category: social attributes. Because each of us is not an independent individual in society, there must be some connections, such as marital status, education level, assets, income, occupation, etc.

The third category is interest preference. Photography, sports, foodies, beauty, clothing, travel, education, etc. This part is the most common and the largest, and it is difficult to list them one by one.

The fourth category is user behavior. Login duration, number of logins, login time period, browsing depth, price preference, purchase preference, etc. within 3, 7, 15, and 30 days.

Step 3: Analyze data and gain insights into users

Use the original data for processing and establish model labels. For example, as mentioned above, I have built a predictive model to improve the churn rate. When you can understand certain behaviors of a certain type of user, you can predict that this type of user may be about to churn, and you can use various strategies to win them back.

Therefore, based on labels related to marketing and consumption, new customers, old customers, user churn and loyalty, user consumption levels and frequency, etc., all constitute the basis of CRM (customer relationship management). Perhaps everyone is more accustomed to calling it a user/member management and operation platform.

Step 4: Apply tags

It is not enough to have a user management platform, it must also be converted into a product operation strategy. Different labels correspond to different user groups and different marketing methods. The CRM structure will include various common channels for reaching users such as SMS, email, push, etc. It also includes CMS (content management system), through which executives can quickly configure activity pages, activity channels, coupons, etc., and drive data through marketing activities. Once the data is running and a closed loop is formed, the user portrait can become clearer and the labels can become more accurate.

Note: Don't get hung up on technical details. Use simple methods to quickly go through the entire process, and then see which links need to be optimized and deepened. For example, for the above churn prediction, the time dimension can be divided into one day or even one hour, but it is not necessary at all. The key is to complete the entire process quickly.

Author: swimming

Source: Leilei has something to say (ID: leileitalk520)