This is the real user portrait. Yours is just a list of gender, age and region.

Is it better to just list gender, age and region when making a large amount of data? What is a true user portrait? The author of this article will give you a systematic answer.

A student asked: Teacher Chen, my boss asked me to do user portrait analysis, but I did a lot of data, but was criticized: I didn’t analyze anything? What should I do? Today, I will give you a systematic answer.

01 Wrong posture of user portrait

1. Limited by data, no action is taken

When it comes to user portraits, many people immediately think of basic information fields such as gender, age, region, and hobbies, and then shout: We don’t seem to have this data, so they give up the analysis. But think about it, is it really that meaningful to know the proportion of men? Is it really helpful to know that the proportion of men is 65% or 60%? Not necessarily. There are many ways to label users, don’t limit yourself to some basic information that is difficult to collect.

2. Listing data without any ideas

When many people hear about user portrait analysis, they instinctively start moving user tags from the database and coding them in the report:

• Male to female ratio 3:2
• 20-25 years old account for 40%
• 30% of people logged in within the last week
• 70% of people did not make a second purchase...

As for what to do with this data, I have never thought about it. This kind of analysis result is of course confusing. I asked in confusion: "So what? What about it?"

3. Infinite splitting, no logic

Many people start to analyze data endlessly when they hear relatively specific analysis topics such as "analysis of churned user portraits". They analyze the churn rate by gender, age, region, device, registration time, source channel, purchase amount... They analyze the churn rate by dozens of dimensions. In the end, they only see that some dimensions differ by 5%, and some by 10%. Of course, there is no final conclusion, and the more they look, the more confused they become.

The above problems are all caused by focusing too much on the four words "user portrait" and neglecting the two words "analysis".

As a basic data system, user portraits do not have analytical functions. Simply listing user tags or breaking down user indicators will not serve as an analytical tool. To make good use of user portraits, you still have to follow the analytical routine step by step.

02 Step 1: Transform business questions

User portrait analysis is essentially thinking about problems from the user's perspective.

To give a simple example, if the sales of a newly launched product do not meet expectations, we can think about the problem from the perspective of product management as well as from the user's perspective.

For the same problem, there are two ways of thinking (as shown in the figure below):

Therefore, simply listing a bunch of user indicators (gender, age, region, purchased products, number of logins...) is useless. User portraits are just a tool for analysis. Like other analyses, you must first consider: what is the actual problem I want to solve?

Once you have thought it through and transformed the problem into a user-related question, you can continue to use the user portrait analysis method.

It is important to note that business issues are very complex. Often, one issue may be related to several user groups and several user behaviors.

For example, the example above involves at least three user groups (potential users, lost users, and existing users) and involves many aspects such as user attitude, information reception, purchase process, and usage experience.

Therefore, it is necessary to classify them into categories, sort out the analysis clues and analysis logic, and find the corresponding data. Otherwise, just listing gender, age, and region will not explain any problems. This involves the next two steps.

03 Step 2: Macro hypothesis verification

After transforming the problem, it is very important to test the hypothesis at a macro level first, which can effectively avoid the error of infinite disassembly. If the general direction is not established, there is no need to look at the details. If the problem is still that the new product is not selling well, if you want to verify it from a general direction, you can simply do it as follows:

• If you suspect that the overall environment is bad, then all categories should be affected.
• If you suspect that a competitor is highly competitive, then the competitor should directly affect our product.
• If you suspect that the operation is too poor, then there must be a link in the product conversion funnel that is off.
• If the above assumptions are verified, further in-depth analysis can be carried out
• If none of the above assumptions hold, new assumptions may be needed.

In short, conducting a general inspection first can effectively narrow the scope of suspicion. The smaller the scope of suspicion, the more accurate the subsequent user analysis can be. At the same time, when data is insufficient, the smaller the scope of suspicion, the more you can focus on collecting data, improving data quality, and making accurate analysis.

There can be hundreds of user classification dimensions. If you do a disassembly comparison without screening, there may be differences in dozens of dimensions, and you will not be able to understand the data at all. It is very important to focus on the hypothesis before disassembly.

04 Step 3: Build analysis logic

After macro verification, you can build a more detailed analysis logic based on the verified conclusions. At this stage, the original grand problem has actually been focused on small problems, such as:

Let’s take a specific scenario:

Verified: We are indeed affected by competitors

1. Sub-question 1: What are the needs of the target users?
2. Sub-question 2: How do target users experience competing products? Which needs are most touched?
3. Sub-question 3: How do target users experience this product? Which gaps are fatal?
4. Sub-question 4: What is the difference between competitors and this product in terms of hard functions and soft promotion?

These four sub-questions can all be answered by deeply exploring user needs and behaviors, and the next step is to go deeper.

It should be noted that this part of the analysis requires a lot of research on user attitudes, potential users, and competitor users. It cannot be completed by internal data alone and must be done through external research.

Let’s look at another scene:

Verified: There are indeed problems with the operation of this new product launch

1. Sub-question 1: Which stage does the problem occur in: preheating, release, listing, or promotion?
2. Sub-question 2: A large number of users failed to respond during the launch phase. What went wrong with the ad delivery?
3. Sub-question 3: Why did the sales volume fail to explode during the promotion phase and fail to stimulate the core user demand?
4. ... (You can ask too many questions, just give a simple example)

These problems can be solved in two ways.

On the one hand, by comparing the following different types of users, we can find the differences in details such as delivery, reward activities, purchase categories, and amounts, so as to adjust delivery, marketing, product replenishment and other businesses.

• Core/Normal
• Purchased/Not Purchased
• Reached/Not Reached

On the other hand, by profiling core users, the business can see more clearly the following information about people who really like to buy, allowing the business to capture more of these core users and improve subsequent performance.

• From which channel
• By what subject
• What kind of discount do you need?
• When to place an order

It should be noted that the source of these users, information delivery responses, purchase behaviors, and topic readings can all be recorded through the internal system. Even if we don’t know whether they are male or female, we can still attract them by placing advertisements, posting content, and offering discounts.

05 Step 4: Obtain user data

In the previous step, we have seen that if user portrait analysis really wants to get to know users, it has to rely on multiple data sources, most likely both internal and external data.

Considering that internal data may not be fully collected and external data may have sampling errors, we must make choices and focus on using data. This is why we have been emphasizing gradual verification and narrowing down assumptions. Only when we focus can we collect data.

Generally speaking,

• The more the question is about attitude, experience, or emotion, the more likely I am to use surveys.
• The more behavioral, consumption, and interactive the problem is, the more we tend to use internal data analysis.
• If you want to understand competitors, you can conduct user surveys on competitors or crawl competitors’ online stores.

In the traditional sense, market research and data analysis have their own definitions, methods, and outputs of user portraits. From the perspective of actual usefulness to the enterprise, of course, the more the better.

However, as crawlers, NLP, and tracking become more and more in-depth, with technical support, the utilization of system-collected data has been increasing in recent years.

Therefore, if conditions permit, we should try to enrich internal data. Otherwise, if we rely on research for everything and do not have accumulated data, it will be difficult to do anything in the future.

06 Step 5: Summarize and analyze the conclusions

If the above steps are done well, drawing the analytical conclusion at the end will be a natural thing and will require no effort at all.

In fact, the biggest problems with user portrait analysis are in the first five steps. Lack of hypothesis direction, lack of data preparation, lack of analysis logic, simply listing data, and unlimited disassembly, in the end, you will naturally be faced with a pile of fragmented data entanglement: "What if the male-female ratio is 3:2???"

Of course, user portraits have many other uses, such as supporting new product development, recommendation systems, automatic marketing systems, delivery systems, etc. Analysis is only a small part of it.

Therefore, if you want to do a good job in analysis, you still need to learn more about analysis methods and practice analysis logic.

Author: Teacher Chen

Source: WeChat public account "Down-to-earth Teacher Chen"