User churn data analysis tips!

A classmate asked: How to analyze user churn? The data on user churn rate can be calculated, but what happens after it is calculated? Just looking at the data, it seems that there is no reason for the loss. We only know that the user has not come for X months, and we don’t know what to do with this. Today I will give you a systematic answer.

1. Common Mistakes in User Churn Analysis

Mistake 1: Trying to retain every user. This is the most common mistake in operations, and many newcomers fall into this trap. If you don’t shop, you will get coupons; if you don’t log in, you will spin the wheel. As a result, the funds were wasted and a bunch of profit-seeking freeloaders were created. In fact, user churn is inevitable, and there is no 100% retention in the world. Every business must focus on its core users. When talking about user churn, what we really need to do is to put the churn rate in a cage and control it at an acceptable level.

Mistake # 2: Trying to understand every reason for churn. This is the most common mistake in analysis, and many newcomers fall into this trap. Users don’t like it? Didn't we do it well? The opponent is too strong? The user has no money? ——In short, I want to give everyone a reason. But there was no data at all, so they just stared at each other in confusion. In fact, we don’t need to and we don’t have the ability to list all the reasons. Same as the previous point, we just need to control the controllable factors and reduce obvious errors.

▌Error 3: Focusing on churn and not activity is like being wise after the event. This is another common mistake. Start the analysis only after the churn rate has actually increased. As a result, the matter is done and the users have all left, so there is no point in analyzing it. Churn rate is a relatively lagging indicator. Before the user is "lost" in the data, he may have already run away and has not been active in the past few months. Therefore, the churn rate should be viewed in conjunction with the activity rate. Pay attention to events that affect user activity as early as possible, and closely track the activity rate of core users to avoid doing useless work afterwards.

2. Basic Ideas for User Churn Analysis

The goal of user churn analysis is to put the churn rate in a cage, so in terms of data, we first focus on the churn rate trend, especially on three types of problems (as shown in the figure below).

Churn rate is a concept relative to activity rate. Although we usually define lost users as "users who have not logged in or purchased anything for X months", real loss may have already occurred when the user is already inactive.

In order to better identify churn problems, we often use both natural cycle and life cycle methods, combined with activity rate. Natural cycles often point to event-based problems (because events occur on natural dates), and life cycles often point to system-based problems (poor business performance, short user life cycles, or breakpoints).

3. Event-type problem analysis method

Negative events can lead to user churn. For example, out-of-stock, price increase, system bugs, user complaints, big promotions by competitors (which we haven’t done yet), etc. This type of event is the easiest to identify. As reflected in the data, the activity rate of the user group affected by the incident will drop sharply after the incident, and N months later, the churn rate will begin to increase.

This way we can focus on the facts and it will be easier to see the results. It is also easier to prescribe the right medicine when designing retention methods. Finding out the real reason that makes users unhappy is more likely to retain users than simply giving them coupons.

Note: Positive events can also increase churn. Especially user acquisition, activation, retention, awakening, etc. Simply stimulating non-consumption soft indicators is most likely to cause false prosperity.

Objectively speaking, as long as there are promotional activities, they will attract arbitrageurs, and this type of users has a high inherent churn rate.

Subjectively speaking, in order to create good-looking data, operators will also reduce restrictions and leave room for arbitrage. As a result, the effectiveness of positive activities is often reduced. For example, when a new user registers, the user life cycle churn rate generated by new user acquisition activities is likely to be significantly higher than that of normal new users (as shown in the figure below). After N months, the churn rate of this group of users is bound to be high.

Therefore, when doing activities, you have to consider the relevant consequences in advance. Positive events are different from negative ones. We still have to do what we should do, and we just need to evaluate it comprehensively. Although the final result may be something that planners and operators do not want to face, what is actually being tested here is everyone's moral integrity.

4. Systematic Problem Analysis Method

If a systemic problem occurs, it only means one thing: our business is worse than that of our competitors. At this point, diagnosing business problems and improving business performance are the core. The diagnostic method can refer to the user life cycle theory. [ User life cycle, this key part is forgotten in the book... ]

The reasons for user loss in the entry, growth and maturity stages are different, and the focus of the analysis is also different. In order to save space, here is a brief summary as shown in the following figure. Students who are interested can click on the "Reading" button in the lower right corner of the article. If the number exceeds 60, we will share it with you later.

When dealing with systemic problems, different focuses are considered at different stages.

▌Entry stage: Generally, there will be no difference in improvement during the entry stage. During the entry stage, users have not actually experienced the core selling points we provide, so we need to improve the process indiscriminately to allow users to experience the core selling points as much as possible.

In the Internet industry, people often focus on the black minute (the minute from downloading to registration) and the process of novice tutorials. In traditional industries, emphasis is often placed on welcoming customers and allowing users to experience and try out the product as quickly as possible.

▌Growth stage: After entering the growth stage, it needs to be treated in different ways. After entering the growth stage, marginal users and freeloaders will be eliminated, and user value will begin to differentiate. Non-core users should be allowed to leave. Trying to retain them is just a waste of money and will also devalue the brand due to frequent discounts.

At this time, we should pay special attention to the loss of core users, the decline in the activity rate of core users, the shortening of their life cycle, and the decline in the proportion of core users among new users. These are all big problems that need to be carefully sorted out and resolved. It is possible that action has already been taken before the churn rate actually increases.

Systemic problems may not be solved in one step, but rather are a continuous iterative process. It’s possible that we can diagnose the problem, but the solution is not easy to use and doesn’t improve the data. Therefore, if you find a systemic problem, you need to:

In the end, we see that our user retention curve is getting closer and closer to our competitors, and the churn rate continues to decline. At this time, we can say that the problem of systematic churn has been solved. This may require many experiments and attempts, so you need to observe and record well and fight a protracted battle.

5. Analytical methods for persistent problems

Ongoing problems are often the most difficult to solve. Because in reality, data such as churn rate, activity rate, and retention rate often show irregular small fluctuations rather than large and continuous growth.

This is the real useless problem: no matter how little attention is paid to it, the leader will always ask about it. I want to take care of it, but I have no idea. There was even a case where the churn rate increased for a few days, but it dropped back before the analysis report was even written. It was really embarrassing.

The order of processing is event type > system type > continuous type. Because single major events are easiest to identify and can be easily seen through data. At the same time, a series of events is often the root cause of systemic and persistent problems, and identifying specific events can also help deal with other problems. Systemic problems are relatively easy to handle if the business side is experienced and can find suitable benchmarks.

The most difficult problem is the persistent problem. Often the change in churn rate will not last to be particularly serious, but will fluctuate repeatedly in a small range (as shown in the figure below). In the absence of experience and data accumulation, it is difficult to fully identify these small fluctuations, so they are solved at the end.

If the problem really cannot be solved, set up observation indicators and track them first. When you reach a certain level, you may be able to find clues.

VI. Differences in handling churn in different business types

Because the churn problem is highly related to the business, the direction of churn analysis for different businesses is also different. From a broad category perspective, there are two most important distinguishing dimensions.

▌Expensive low-frequency products VS cheap fast-moving consumer goods.

The more expensive the product (cars, houses, large home furnishings, weddings, etc.), the longer the user's decision-making process is, the more they tend to judge in advance, and there is no such thing as repeat purchases. This type of business user decision-making has a clear window period, and the closer it is to the deadline, the more likely the user is to make the final judgment.

Therefore, user churn for this type of business is a countdown hourglass. When you first come into contact with a user, you must understand the user's status: what the user's needs are, which competing products have been compared, and whether bargaining has begun.

This way we can roughly judge how much time we have left. This will help you better seize transaction opportunities and follow up quickly. Instead of being foolish and introducing and following up step by step without considering the needs, the opportunity will be lost.

Users of fast-moving consumer goods, or consumer products with a high purchase frequency (such as clothes, shoes, and mobile phones), are naturally less loyal and their attitudes can be easily changed by popular trends and promotions. It is entirely possible to adopt a strategy of retaining employees without any gaps. Anyway, even if the user doesn’t buy this time, he will come back to buy after a while.

Therefore, when dealing with such products, Internet companies often distinguish between two types of loss retention: platform loss and product loss.

As long as the user remains on the platform, continue to wake up. Traditional companies often use seasonal changes, new product launches, periodic celebrations, holiday events and other means to activate users multiple times. In short, as long as the user value is big enough, don’t abandon or give up.

▌Traditional industry VS Internet industry.

The amount of data accumulated by the two during the user life cycle is different. The Internet industry has a lot of data, which can often record the entire process of users from clicking on the promotion link - landing page - registration - browsing - ordering.

Therefore, the funnel analysis method is often used to see at which steps the lost users get stuck, and to identify the problem points for improvement. Especially during the new user registration stage, there is often indiscriminate optimization.

Traditional industries often only have consumption data, so users can only be measured by consumption frequency and consumption interval. Generally, after users consume n times, those who don’t like the product will leave, while those who like the product will continue to buy. This is the so-called magic number. For more information about magic numbers, please read this article [What is the magic number? How to find it with data analysis】

By comparing the size of the magic number, you can know the gap between yourself and your opponent. As for the behavioral aspects such as user arrival at the store - welcoming customers - experience - service - evaluation, there is no data at all, which needs to be supplemented through market research and other means.

The main reminder here is that there are huge differences between businesses, although the definition of churn can be defined as no login/no purchase in XX months. However, the actual loss scenario may have already occurred, and the key actions to stop the loss may not have data records. It is more effective to think of solutions based on specific business needs than to use mechanical codes.

VII. Summary

Many students find the user churn problem difficult to deal with. On the surface, it is because there is little data on user churn and we don’t know what users are thinking.

But in essence, the reasons that lead to user churn are related to many factors such as user life cycle, user segmentation, user decision-making process, user growth path, new user conversion process, user experience, user MOT, and the influence of competing products.

Any topic here can be put into a separate article. Once you understand all of this, you will basically understand the entire user operation process. Essentially, user churn analysis is difficult because few people who do the analysis understand the business of user operations.

The answer is: I don’t really know. Or even: I don’t know anything at all. You ask him what he knows? He only knew how to calculate the churn rate data, and then make a bunch of cross-tabulations based on indicators such as user age, gender, registration channel, purchase frequency, etc. Then I stare at the 1%, 2%, or 3% differences in the data sets and wonder: What do these differences mean?

The above is a joke. In short, analysis is not just about running data and pulling up a table, but also about getting deep into the problem and finding the real root cause of the business problem. This article is already very long and there are some incomplete details, which I will slowly fill in later.

Author: Down-to-earth Academy

Source: Down-to-earth Academy