How to analyze user activity index?

This is a very boring physical job, but it is the key to achieving results. Without these details, any "thinking method", "underlying logic" or "core rule" will not work. Only fortune tellers can tell the truth by shaking coins. People who do data analysis are actually no different from bricklayers.

01. Common issues regarding user activity

1. There is no unified definition of what is active.

User registration and payment are very clear actions, and generally there is no mistake. However, there are often different opinions on what "activity" is, such as:

1. Is successful login considered active?
2. How many clicks after logging in is considered active?
3. Login to complete a special action?

Of course, different definitions can be used for different goals and different businesses. But the premise for using these definitions is to have a unified caliber. All departments must reach a consensus: those who have XXX behavior are considered active. The most common problem is that not only is there no unified caliber, but new terms are constantly invented, making historical data inconsistent. In the end, the meeting ended with everyone talking at cross purposes.

2. Get bogged down in details and worry about every day’s fluctuations.

Students who have looked at the activity rate and active number indicators know that these things fluctuate too much on a daily basis. Almost everything, big or small, will have an impact on activity rate. Sometimes, before the report analyzing the decline in activity rate is submitted, it increases back on its own. As a result, I am obsessed with “Why is it 1% higher/lower again!” every day. I am exhausted but still haven’t found the root cause.

3. Look at problems in isolation and spend money to make them active.

Although it may not be clear why the activity rate dropped, the method of how to increase the activity rate is very clear. Sign in for 7 days to get benefits, sign in for 30 consecutive days to get benefits, log in to win a lottery up to 888, plant trees for 20 days to get a bag of kiwis... Therefore, often before the cause of the problem is found, short-term measures to increase indicators have already been implemented.

As a result, one problem is solved but another one arises. The activity rate is high, the conversion rate drops, the conversion rate is high, and the cost is exhausted...

Indicators such as registration, activity, and payment can never be viewed separately, just like when we comment on a person's figure, we report all three measurements together. Otherwise, if you want a 36D girlfriend, would you like one with a 36-48-52 figure? Not only do you have to want it, you also have to be happy.

02. Core issues of user activity analysis

I always give the example that working with data but not understanding the business can lead to all kinds of problems. However, in user activity analysis, it is precisely because those doing business do not understand the data that the above chaos occurs. The operations department has not thought deeply about the meaning of active indicators, nor has it analyzed the relationship between active indicators and registration, payment, and conversion. Simply because “this is my KPI”, “KPI indicator falling syndrome” arises, and people are impatient to take action without waiting for a clear analysis. In the end, they always treat the symptoms and not the root cause, and it is very troublesome.

If you want to break the impasse, you must first solve a core problem: What does being active mean to us? Apart from very liver-intensive games like Onmyoji and DOTA Legend, do a large number of other Internet applications really require users to stay here every day? Apart from super apps like WeChat, is there really another app that users can’t replace and check every day? (As shown below)

Essentially, Internet applications focus on user activity, just like traditional businesses focus on customers visiting their stores.

• Activity is the foundation of everything and must be paid attention to
• You can’t just be active without conversion, you have to connect them together
• There are too many influencing factors, so we must focus on the big ones and ignore the small ones, and focus on action

When talking about user churn analysis, we said: The purpose of churn analysis is not to eliminate the churn rate, but to control the churn within a controllable range. When analyzing user activity, the principle is similar: the purpose of doing activity analysis is not to force users to check in every day, but to provide stable support for payment and referrals.

03. Basic ideas for user activity analysis

1. Setting standards: What new, active, and paid structures are needed for the current business?
2. Finding patterns: What should a regular active trend look like?
3. Check for abnormalities: distinguish between normal changes and abnormal changes
4. Cause tracking: Track and analyze abnormal changes
5. Make a plan: Make a plan based on the severity and urgency of the problem

The most important thing, of course, is to set standards. As a business party, you need to have a psychological judgment: how many active users do I need and what is the activity rate I need. And we cannot look at one indicator in isolation, we must pay attention to the overall shape of AARRR.

Common methods for setting standards

The standards are mainly based on three points:

1. Business characteristics: Different businesses require different numbers of active users and activity rates.
2. Development stage: Generally, new apps tend to gather more people (increase DAU) and only make paid conversions when they reach a certain level.
3. Competitive strategy: Different strategies mean different requirements for activity and payment.

For example, the three most basic strategies (as shown below):

Please note that competitive strategy is at the heart of standard setting. For example, the traditional view is that financial services are low-frequency businesses. However, when making a financial app, you can incorporate financial news, financial education, local food and drink guides, movie information and other consumption-related things into it, turning a low-frequency application into a high-frequency application.

Therefore, the general characteristics and development stages of similar businesses are only for reference. More importantly, it is the inner voice of the business: "We want to make an XXX application. Compared with the products on the market, my goal is XXX."

This requires that operations must have their own business judgment capabilities and a clear understanding of their own direction. Data analysts only play an auxiliary role, providing data such as business characteristics, development stages, and competitive product data for reference.

This is why Mr. Chen complained that the issue of activity is "done haphazardly from the top". Indeed, many companies have no idea about their operations and just know how to mechanically complete KPIs as long as the data meets the standards. When the data does not meet the standards, they try to blame it on the data analysts' lack of insight, the competitors' inflated numbers, or our insufficient funding. If the standards are not clear, there is no way to talk about the subsequent analysis.

Common ways to find patterns

There are three types of rules

1. Policy rules. After the policy was released, it generated a huge response.
2. Laws of nature. From January to December of the year, there are many factors that affect activity (as shown in the figure below)
3. Operation rules. Common operational accidents (product out-of-stock, system downtime, misleading publicity...) and operational measures (lotteries, sign-ins, interactive games) can all trigger changes in active data.

These specific policies, events, and operational actions are the fundamental reasons for changes in indicators. Therefore, before analysis, you should first collect a large amount of internal and external events and think about the problem based on these events. Instead of just talking about numbers and saying illogical things like "because it rose in the past three days, it rose today" or "because it rose on the previous Friday, it will rise this Friday".

After finding some obvious patterns, they can be used to make qualitative predictions and predict indicator fluctuations based on when they will occur in the future. It can also be used for explanation. For example, when an indicator fluctuates, if a corresponding event occurs + a corresponding fluctuation pattern, then it is most likely a regular change. Doing this can save a lot of analysis time, instead of doing a lot of analysis and then being criticized: "I knew it a long time ago" or "That's how it is."

Common methods for checking abnormalities

If you encounter an abnormality, pay attention to:

1. Amplitude: Is the daily fluctuation large enough?
2. Sustainability: Is there a trend of continuous increase or continuous decline?
3. Regularity: Is there a regular, planned fluctuation?
4. Correlation: Do the associated registration and payment indicators fluctuate in the same way?

Note that not all fluctuations are worth chasing. Large-scale, continuous, irregular, and those that affect other indicators should be given priority. Occasional fluctuations are normal, but it is necessary to record the time of occurrence and observe the trend so that it is easier to trace the source when the problem worsens. By doing this, data analysts do not have to get stuck in endless entanglement and it is easier to find the real abnormal problems.

Common ways to find out the cause

Confirm that it is an abnormal fluctuation. There are three common forms:

1. Event type: one-off, sharp decline
2. Continuous type: starting from a certain point, it continues to fall
3. Systematic: Small volatility, but always worse than competitors

First determine what type of problem it is and then find out the cause. The difficulty of tracing the causes is: event-based > system-based > continuous. An incident that occurs once is easiest to trace to its source. The answers to systematic differences can be found through competitive product analysis. On the contrary, persistent problems are the most troublesome. They may disappear on their own, they may be the aftermath of a major event, or they may be deeper-seated problems.

It should be noted that we often say DAU=DNU+DOU (daily active users = daily new users + daily active old users, generally new registered users are directly counted as active). Often systemic problems will affect DNU. If the user is not properly guided within the time of T+1, T+2...T+N after registration, it will lead to user inactivity or even loss.

DOU is often related to events, such as seasonal promotions, awakening silent users, new product launches, etc. Therefore, when tracking the cause, you can observe it separately. For new users, pay attention to the process from registration to first payment, and for old users, tag them and pay attention to their response to the activity (as shown in the figure below)

Common ways to plan

However, there is no such part. This part belongs to the operation scope and is a business action, which is not within the scope of discussion in this article. The planning mainly depends on the business capabilities of the operation. As data analysis, the support that can be provided is:

1. Prioritize issues
2. For urgent and important issues, indicate the source of the problem
3. Provide ROI analysis results to support past methods of improving problems
4. Waiting for operations to come up with ideas and provide temporary support

Finally, I would like to emphasize one sentence: good methods are designed, not calculated. Data analysis can only be used to evaluate the pros and cons of past methods, and at most to predict the user response rate to XX product, but no more. To truly implement it well, the operations team still needs to practice their internal skills.

After reading it from the beginning, we will find that data analysis methods are not mysterious at all, but more about:

• Collect events in large quantities and in detail
• Use data to describe, evaluate, and summarize events
• Use logic to deduce the impact of events and use data to verify assumptions.

This is a very boring physical job, but it is the key to achieving results. Without these details, any "thinking method", "underlying logic" or "core rule" will not work. Only fortune tellers can tell the truth by shaking coins. People who do data analysis are actually no different from bricklayers.

Author: Down-to-earth Academy

Source: Down-to-earth Academy