How to carry out data operations well? Here are 5 tips!

In today's Internet age, almost everyone knows the power of big data . Every business owner emphasizes that we must speak with data, and every student who does operations knows that data is very useful. So, what exactly is the data operation in operation?

This is also the first thing we want to mention in our data article: data is not just those Arabic numerals, its significance lies in the "cause" and "effect" behind it. So it is first and foremost a logical thinking . How can we know what kind of "result" we will get based on what kind of "cause" we sow? This depends on the data. Data is the "result", and previous operational work is the "cause". If we want to change the "result", we must change the "cause".

Let’s first use an easy-to-understand case to explain the steps of data operations:

We usually have to go for physical examinations regularly. After the examination, we will get a physical examination report (with various indicators and data). After looking at the data, we will compare it with the standard data. After the comparison, we will find the healthier and worse parts of our body. Then we will think of solutions based on the worse parts of our body. Finally, we can achieve health for the whole body by regulating or healing those worse parts.

Let's analyze the above steps:

When we go for a physical examination , we begin to do data operations .

The amount of data in each physical examination package is different. Whichever package we choose has completed our data planning work.

When we get a physical examination report , we have completed our data collection work.

By comparing, we discovered our weaker body parts and found ways to improve them , thus completing our data analysis work.

By implementing this method, we ultimately improved the overall physical condition , thus completing a data-driven operation .

And all the data on that physical examination form is our data model .

Let’s look at the following figure to see the steps of data operations:

How can this physical examination help us?

• It intuitively tells us the current condition of our body.
• It tells us which points are good and which points are bad?
• It will lead us to a deeper understanding of an organ in our body.
• It helps us find ways to improve.
• It ultimately makes us healthier.

If we replace "body" with "company", the role of data operations will immediately emerge: telling you the current status of operations, telling you whether the operations are good or bad, giving you a deeper understanding of users, helping you find ways to improve results, and ultimately allowing the company to develop better. This is where DataOps comes in.

Now let’s get into the topic. What should we do?

Let’s take a look at the physical examination case. If we view the physical examination as a data operation, there are two necessary prerequisites for its completion:

1. Know what data to expect;
2. Know what each data means about your body's condition.

Similarly, in our operations, we must first know:

1. What data do we collect?
2. What does each piece of data mean about the operation?

Many students who work in operations, especially some novice operators, feel that there is too much data. It seems that everything is useful, but at the same time they can’t start. In fact, the data is not as complicated as we think.

Before we talk about data, let's first understand two concepts :

1. Data analysis is an extremely large subject that is widely used in all walks of life and fields. It requires statistics, database technology, programming technology, etc. What we are talking about here are just some data operations carried out for users under a company's business system.
2. Data operation is a job with no downward limits. Theoretically, as the amount of data increases and the dimensions continue to be broken down, it can be refined infinitely.

For example: We counted the number of new users in one day. This is the data that ordinary companies would collect. But as the amount of data increases, we can break it down into dimensions:

• For example, we split the dimension region into provinces, then cities, then districts and counties, and then regions...
• For example, the dimension time: we break it down into hours, then minutes, and then seconds...
• For example, we divide the dimensional population into gender, then into occupation, and then into age...
• For example, Dimension Devices: We disassemble them into operating systems, and then disassemble them into mobile phone models...

If we break it down into the above dimensions, we are likely to get an extremely detailed piece of data, for example: at a certain time on a certain day, in a certain community in a certain city, a 25-year-old female, a teacher, holding an iPhone 7 mobile phone, became our user.

As long as we master our object-oriented thinking, it is not difficult to split the dimensions. We only need to judge: which dimension we need to split in our current operation work.

The "big" in big data that we often talk about actually refers to the "big" amount of data + the "many" dimensions.

Next, we need to start planning our data. How many core indicators are there in our operations?

Step 1: We need to find our top-level core data indicators

The so-called top-level data indicators are actually a direction that guides all our operations work to move forward in one direction. It is also the overall strategic direction of a company, which is an indispensable point. So how do we determine our top-level data indicators? If we look back at our “Thinking Chapter | How to Become an Operations Expert (Part 2): Operations Cube”, we will know that, apart from the product, the most important indicators for most companies are: number of users and profit per customer, which is also called AR PU (Average Revenue Per User ).

 As you all know from the previous chapters, except for e-commerce which goes directly from attracting new customers to conversion in one step, in most cases we need to lay the groundwork for retention and activation before conversion. Therefore, before we obtain ARPU, retention rate and activity are very important. Therefore, retention rate and activity (generally using daily active users as the standard, referred to as DAU) are also the top-level core indicators.

Obviously, these are some of our most important data indicators. In this way, we identified the four most commonly used core indicators at the top level: user volume, retention rate, activity (DAU), and average revenue per user (ARPU). In fact, the above four indicators also correspond to our four steps of attracting new customers, retaining existing customers, promoting activation, and converting existing customers.

Step 2: There can only be one top-level data indicator

Although the four indicators of user volume, retention rate, activity (DAU), and average revenue per user (ARPU) are all very important and must be addressed simultaneously in operational work, we must also choose a leader among these four at different stages of operations . There is no clear selection criteria for this, and it will be adjusted according to the actual situation of the company (industry, model, capital support behind it, etc.). Generally speaking, the core indicators in the early stage will be the number of users, in the mid-term will be retention and activity, and in the later stage will be the average revenue per user. If a startup company is raising funds and its product is in the A round, then the above indicators may correspond to the B round, C round and IPO round respectively.

Step 3: We start to break down our indicators

We know that the four indicators of user volume, retention rate, activity (DAU), and average revenue per user (ARPU) are our core indicators, but if we only look at these four indicators, it will be biased in many cases, so around these four, we also have some important core indicators.

1. Number of users

Those who have read the previous articles know that our user base comes through channels , and the channels come from: our own product channels, external free channels , and external paid channels. Regardless of which of the above, what we do is: let certain content be exposed on a certain channel, let certain users see it, and finally acquire users.

So, without considering channel costs, such as owned channels and free channels, we can break it down into the following two data indicators: UV (unique visits) and new user conversion rate . The new user conversion rate = the number of new users this time / UV (number of unique visits).

When considering channel costs, we must consider another indicator: the cost of acquiring new users.

for example:

• Channel A: Invested 10,000 yuan, brought in 5,000 traffic , converted into 1,000 users, and the conversion rate is 20%.
• Channel B: invested 2,000 yuan, brought in 3,000 traffic, converted into 450 users, and the conversion rate was 15%.

Judging from the above data, although channel B is lower than channel A in terms of traffic and conversion rate. We will still give priority to channel B because its cost of acquiring new users is relatively low.

Of course, we cannot judge the quality of a channel solely by the cost of acquiring new users. We must also comprehensively consider the subsequent retention and activity of these users, otherwise we will fall into the hands of the wool party .

In this way, we have three more core indicators in this link: UV (unique visitor number), new user conversion rate, and new user acquisition cost.

2. Retention rate

The retention rate is relatively easy to understand. It refers to the proportion of new users who remain after a certain period of time. Since retention rate is a linear concept, we generally break it down from the dimension of time. We have three more commonly used core indicators: daily retention rate, weekly retention rate, and monthly retention rate. Generally, we mainly look at monthly retention, but daily retention and weekly retention are very important for some high-frequency used products.

3. Activity (DAU)

In fact, there is no clear definition of "activity" for different companies and products. Especially for some low-frequency products, such as travel apps, most people will only open them when they want to travel. At this time, some high-frequency auxiliary outputs are needed to support them. Refer to "Structure | How to Become an Operations Expert (IV): A Picture of Operations"

Daily active users are also a linear concept, but they cannot be viewed alone. They must be considered in combination with daily active user growth . Here is an example of why:

The daily active users of a certain product on the first day were 10,000, and the daily active users on the second day were 11,000. On the surface, the data looks good.

But in fact, there were 2,000 newly added and active users (DNU) on the first day, and 2,000 newly added and active users on the second day. Therefore, the daily active user growth on the first day is 2000/10000=20%, and the daily active user growth on the second day is 2000/11000=18.2%. In other words, daily active users are indeed growing, but the growth rate has slowed down.

 We all know from physics knowledge that acceleration is a very important thing. When the acceleration begins to decrease, although the speed is still increasing, it is also showing a declining trend, and we need to be vigilant. If one day the daily active user growth is 0, it means there are no new active users. When Facebook's Mark Zuckerberg went to meet investors , all the data he showed were lower than those of his then-rival Myspace, except for one: daily active user growth . Based on this data, investors believed that Facebook would surpass Myspace within a year, and this was indeed the case. Therefore, daily active user growth is also one of our core data indicators.

Here, we have two more core data indicators: new and active DNU on the day and daily active growth (DNU/DAU)

4. Average Revenue Per User (ARPU)

ARPU is a time-bound value, usually measured in months. It is generally understood as the average monthly income of each paying user , and the formula is total monthly income/monthly number of paying users. Therefore, ARPU must be viewed together with the number of paying users. For example:

• Company A: Monthly revenue is 10,000, with 990 free users, 10 paying users, and a monthly ARPU value of 1,000.
• Company B: Monthly revenue is 50,000, with 500 free users, 500 paying users, and a monthly ARPU value of 100.

 From this example, we can see that although Company A’s ARPU value is much greater than Company B’s, it is definitely not as good as Company B’s because its paying user conversion rate is not as high as Company B’s. Therefore, we have one more core indicator here: paid user conversion rate.

In addition, we know a concept: every product has a life cycle, some are long, some are short. Example:

• Product A: Its life cycle is three months and its ARPU is 1,000 yuan/month.
• Product B: Its life cycle is 1 year and its ARPU is 800 yuan/month.

Obviously, product B is more attractive than product A, so we need to introduce another indicator, user lifetime value (LTV). That is, the value contributed by users during the entire life cycle of the product. For example, in the above example: A is 3,000 yuan and B is 9,600 yuan.

Whenever we talk about revenue, we have to think about cost. Do you remember the box in our operating structure? It has an entrance and an exit. Therefore, return on investment (ROI) is also a core indicator that we must grasp.

Okay, up to this point, we have listed the most important core indicators in the entire operation. Of course, we can continue:

Step 4: Continue to split according to actual needs

The data indicators in the first three steps are almost the core indicators required by any company. Starting from this step, there are basically no standards. Different industries, different companies and different stages can continue to be refined unlimitedly. for example:

For new users who need to register, we can divide them into: guide page conversion rate, registration page conversion rate, completion page conversion rate, etc.

Channels can be further divided into reach, click-through rate , etc. according to the channel type (CPM, CPC, CPA).

The churn rate can be divided into: pre-payment churn rate, post-payment churn rate, etc.

In terms of activity, we can divide it into: light activity rate, moderate activity rate, heavy activity rate, etc. according to the set activity standards.

We can divide the transaction into: collection conversion rate, order conversion rate, transaction conversion rate, etc.

We can also continue to refine based on the above dimensions such as time dimension, regional dimension, device dimension, user dimension, etc.

To a certain extent, the advancement of science and technology also means the continuous refinement of data. Perhaps this is the future of artificial intelligence .

Well, at this point, we have listed all the core data indicators we want, as shown below:

The reason why we explain each of the above data indicators separately instead of explaining other data is because the above indicators are directly related to the survival of a company and are the most important core indicators in our operating system . In our operating system, these indicators belong to three-dimensional space indicators, while the others belong to two-dimensional space or one-dimensional space indicators. As for the four-dimensional space of operating indicators, we will describe it in combination with analysis methods in the next article, so stay tuned.

Next, let’s think about this with a small question:

Title: We are currently operating a product, and its operating data is as follows:

The daily, weekly, and monthly retention rates of new users of the product are 80%, 40%, and 20% respectively, and then stabilize at 20%

We define the top 20% of retained users as active users.

We have a paid product that can bring 10 yuan profit per transaction

The life cycle of this product is 1 year

All active users log in once every 2 days on average, and 1 transaction is generated every 3 times

All inactive users log in once every 6 days on average, and generate 1 transaction every 5 times

At the beginning of the third month of the product, we prepared a soft article release. According to convention, the registration conversion rate of this soft article is 5%, and the investment amount this time is 100,000 yuan.

question:

1. When the number of readers of this soft article reaches a certain level, can we basically ensure that the investment will be recovered?
2. Assuming that another channel quotes a CPC price of 1 yuan per time, what is our estimated ROI on this channel?

(Note: If your calculation time exceeds 5 minutes, then you are probably using the wrong method.)

You can think about it for a while, and we will announce the answer at the end of this article.

Let’s continue with the content of this article. Can the picture above be considered as our operational data system?

No, not yet. Even if we refine many dimensions, it is only a data model and cannot meet the requirements of a data system. Because the above data is processed data and is at the presentation layer.

What does it mean? Let’s look at the data above, such as conversion rate, retention rate, activity rate, etc. These are actually the data we want subjectively, rather than the data that can be obtained directly. In other words, what we want is processed data, and in order to obtain this data, we must have the original data at the access layer.

Let's take an example:

We held an event with two pages, the guide page and the registration page, which were launched on channel A. We now want to know the new user conversion rate of this activity on channel A.

The new user conversion rate we want is a processed data, which will not be directly told to us in the original data. Therefore, we must design the original data based on the processed data. As shown below:

Based on this table, we can know that 4 people came to channel A. Through their behavior records, we can know that 1 of them finally completed the registration, and the new user conversion rate is 25%. Here, user source, user, behavior, and time are the raw data, and the final new user conversion rate is the processed data.

The original data is generally obtained through the following aspects:

(1) Setting by user attributes and behaviors

For example, the user’s age, gender, and what actions the user has taken.

(2) Setting by product functional segmentation

This is mainly aimed at some Internet products such as APP. The data of each subdivided function in the entire product process, such as collection, comment, coupon collection, etc., are also our original data.

(3) Calling external data

Generally, it is to call the API interface of external products. For example, by calling the WeChat interface, we can obtain various data on WeChat. A more crude method can also be achieved by exporting and importing reports.

If we want to build our final data system, we need to first determine our processed data, and then build the original data behind it based on these processed data. Only in this way can data planning be truly completed.

Finally, let’s look at this picture:

We collect raw data from the access layer, process it, produce processed data, present it, and then analyze it through analytical means to identify problems and form effective operational solutions, which are then implemented to generate a new round of raw data. This is a data-driven closed loop.

Okay, now we have truly completed all the data planning work in the early stage of data operation. But it is worth mentioning that data planning cannot be completed in one step, and in fact it is difficult to complete it in one step. We should build the data system as perfect as possible as early as possible, and continuously improve and optimize it during the operation process. I hope this article can also help you complete your own data planning.

Finally, let's use the solution to the above small question to complete this article:

• First, we need to recognize that the daily retention and weekly retention here are pitfalls.
• Calculate LTV:
  1. Assuming there are 100 new users, there will be 20 retained users, of which 4 are active and 16 are inactive.
  2. The average output of active users is 1 transaction/6 days, that is, 5 transactions/month = 50 yuan/month. Since the product life cycle still has 10 months, the output is 500 yuan/person, and 4 people means 2,000 yuan.
  3. The average output of inactive users is 1 transaction/30 days, that is, 1 transaction/month = 10 yuan/month. Since the product life cycle still has 10 months, the output is 100 yuan/person, and 16 people means 1,600 yuan.
  4. The total amount is 3,600 yuan, divided by 100, and the LTV is 36 yuan/person

The total investment is 100,000 yuan, which means the number of new users needed is 100,000/36 ≈ 2,778 (people)

Based on a 5% conversion rate, 2778/0.05=55560 (reading volume)

(1) That is, the effective reading volume must reach at least 55,560 to ensure that the investment of 100,000 yuan can be recovered in 10 months.

(2) The CPC unit price is 1 yuan per time. According to our conversion rate, our new user acquisition cost is 20 yuan per person. Combined with our above-mentioned LTV of 36 yuan per person, our return on investment is ROI=36-20/20=80%.

summary

Data is the key and prerequisite for data operations, and data planning is also the first step in data operations. Of course, now that we have our data, how can we effectively display it and use it reasonably to ultimately produce results? This requires talking about the scenarios in which data is used and the methods of data analysis.

The author of this article @志远 is compiled and published by (Qinggua Media). Please indicate the author information and source when reprinting!

Product promotion services: APP promotion services, information flow advertising, advertising platform