How much do you know about data operations?

Generally speaking, data is the most authentic way to reflect the status of products and users . It is used to guide operational decisions and drive business growth. Unlike the position of data analyst , data operations focuses more on supporting front-line business decisions. When used throughout the entire life cycle of product operations , data operations is a skill that uses data analysis to discover solutions to problems, improve efficiency and promote growth.

1. What knowledge do you need to learn for data operations?

1. Clarify the purpose of data analysis

When doing data analysis, you must have a clear purpose, know why you want to do data analysis, and what results you want to achieve. For example: to evaluate whether the effect of the revised product is better than before; or to find the direction of product iteration through data analysis, etc.

After clarifying the purpose of data analysis, the next step is to determine what data should be collected.

2. Data Collection Methods

When it comes to collecting data, the first thing to do is to do a good job of data collection.

The so-called "burying points" means adding statistical codes to the normal functional logic to collect the data you need.

There are currently two mainstream data burial methods:

• The first one: develop it yourself. Add statistical code during development and build your own data query system.
• The second method: using third-party statistical tools .

Common third-party statistical tools include:

Website analysis tools: Alexa, Google Analytics, Baidu Statistics

Mobile app analytics tools: Google Analytics, Umeng , TalkingData, Crashlytics

Different products and different purposes require different supporting data. After determining the data indicators, choose a method suitable for your company to collect the corresponding data.

3. Basic data indicators of the product

• New users: The number and rate of new users added. Such as: daily new additions, monthly new additions, etc.
• Active: How many people are using the product. Such as daily active users (DAU), monthly active users (MAU), etc. The more active users are, the more likely they are to bring value to the product.
• Retention rate : How long do users stay with your product? For example: next-day retention rate, weekly retention rate, etc.
• Spread: On average, each old user will bring in several new users.
• Churn rate: The ratio of users who have lost during a period of time to the number of active users during this period of time.

4. Common data analysis methods and models

Here we will talk about the funnel analysis method and AARRR analysis model

Funnel analysis

It is used to analyze the changing trend of the number of users from potential users to end users, so as to find the best optimization space. This method is widely used in the analysis of various key processes of the product.

For example, this example analyzes the changing trends from when a user enters the website to when they finally purchase a product.

The conversion rate from users entering the website to browsing the product page is 40%; the conversion rate from browsing products to adding them to the shopping cart is 20%, etc. To find out which link has the lowest conversion rate, we need comparative data.

For example, from entering the website to browsing products, if the conversion rate of the same industry level is 45%, and ours is only 40%, it means that this process has not reached the industry average. We need to analyze the specific reasons and then optimize and improve it in a targeted manner.

Of course, the above is an idealized funnel model we designed, and the data may be obtained after aggregation. However, actual user behavior may not follow this simple process. At this point, it is necessary to analyze why users have to go through such a complicated path to reach their ultimate goal, and think about whether there is room for optimization in the middle.

AARRR Model

This is a data analysis model that all product developers must master.

The so- called user acquisition means attracting new users. For an APP, attracting new users means new users downloading and registering; while for many WeChat public accounts , Weibo, and Tieba operators, attracting new users means attracting new fans.

When listing your channels , it is important to note that each channel needs to have a basis, including whether the channel is in line with your target audience, whether the unit price is high or low, and whether the channel can be disseminated secondaryy, etc. Now the channels for promoting APP include:

Acquiring users is the process of attracting new users through various channels. In addition to exchange cooperation, there are other free methods such as posting in major forums, post bars and other communities, and social marketing . Payment methods include but are not limited to the use of search engines , self-media such as WeChat and Weibo headlines, online alliance advertising, offline activities , and Internet TV. Some people also use this special method of growth hacking .

There is a criterion for judging whether attracting new customers is effective - triggering key behaviors. For example, users may not necessarily use an app even if they download it. The key behavior depends on the product. It may be browsing an article, watching a video, sending a message, starting a game , or filling out an email.

A good channel does not mean the channel with the largest number of users, nor does it mean the channel with the lowest cost. Only by constantly exploring user preferences and distribution can we optimize and reasonably determine investment strategies and continuously minimize CAC. The number, quality, and cost of acquiring users through each channel are different, and need to be comprehensively evaluated based on user acquisition cost (CAC), user volume, retention rate, AR PU data, etc.

Of course, in addition to acquiring new customers through external channels, if the user base is large, new customers can also be attracted from the perspective of product design.

First, proactively inform users. There are three ways: APP push messages, EDM emails, and SMS notifications. The time, content, and users of message push can be determined based on user portraits .

Second, passively inform users through splash screen ads, set up obvious entrances, add discount notes to function entrances, set up relevant carousel images on the homepage, etc.; for example, the splash screen ads of Mobike APP show online car-hailing , and various other functions of Didi APP.

Increase activity

Activity refers to the time and frequency of users using the product. Each product has a different definition of activity. For example, Baidu Tieba hopes that users can log in, post and comment every day; online education products are more concerned with users' learning time, number of practice times, etc.

Activity is built on the core value of the product, such as high-quality content, increasingly better user experience , multi-functional needs, etc., to capture users within the first few tens of seconds of use.

There are also some auxiliary means, including activities that meet user needs, a complete user incentive system, a growth system, ways to increase user interaction with other users, and more detailed operations such as the APP's novice guide.

A more comprehensive analysis approach is to list each process of the user from the beginning to the end of using the product separately, and from the user's perspective, constantly look for ways to promote activation. For example, analyze the conversion rate of new functions, the smoothness of the usage process, and extend the user's product usage process.

Of course, we can also screen out high-quality users. If users of a certain channel use the product for a considerable amount of time and launch it a lot, then investment in this channel should be increased. In addition, there are some users who have only launched the product once, and most of these users are passively activated.

In addition to channels, another analysis dimension related to activity is version. But this will create two illusions: users are accustomed to the current products, so they don’t want the products to be updated; users will ask you to add new features.

For example, when Facebook first launched its News Channel in 2006, it caused a huge user backlash. But over time, the product became a core feature of Facebook. Facebook ignored the minority opposition and stuck to its strategy.

We don't want to excite existing loyal users, but we also need to acquire the next million users. It's easier to add features than to cut them. Often users request features that solve small convenience problems rather than true solutions. We need to actively communicate with users, and if the data tells you that the new direction is correct, then ignore the minority of users who speak out.

Improve retention

Users who start using a product and continue to use it after a period of time are considered retained users, and the ratio of retained users to new users at that time is the retention rate.

The user life cycle in each application is a process of contact-use-abandonment or forgetting. During the user usage stage, effective activation methods can also improve retention, but it is equally important to regain users, and there is a common process for regaining users.

First, determine the criteria for lost users; then establish a user loss model, analyze why users are lost, and take appropriate measures to remedy the situation; at the same time, let users know that you are recalling them through EDM, text messages, etc.; finally, use onboarding to familiarize users with product operations again and continue to retain them.

Revenue

Currently, there are three main ways for mobile applications to generate revenue: paid applications, in-app payments, and advertising. Paid downloads are more common in Apple's APP Store. Advertising is the source of income for most developers, and in-app payments are also relatively common, such as in games, value-added services, and self-operated shopping malls. It should be noted that in addition to advertising, Amap's profit model also lies in the combination of its own map data and user data with other fields.

People usually use ARPU (average revenue per user) value to determine the revenue standard. But for an application that has both paying and non-paying users , we also need to look at ARPPU (average revenue per paying user).

Because it involves the proportion of paying users among all users, if the number of paying users is low, then you need to consider whether there is a problem with the product's profitability, including pricing, product features, monetization methods, etc.

When calculating revenue, you also need to consider profit. When calculating profits, there is an indicator: LTV (lifetime value). The user lifecycle refers to the total revenue generated for an app from the first time a user launches the app to the last time the user launches the app. The difference between LTV and CAC can be considered the profit that the app makes from each user.

Self-propagation (Refer)

The rise of social networks has brought greater vitality to products - self-propagation based on social networks. Self-propagation, or viral marketing , comes from viral communication , that is, a host that has been infected with the virus will also be infected with the virus when it comes into contact with other hosts. The K factor quantifies the probability of "infection".

K = (the number of invitations each user sends to his friends) * (the conversion rate of people receiving invitations into new users). When K>1, the user base will grow like a snowball, but most mobile applications still have to be combined with other marketing methods.

In addition to having a good product, it is also important for self-propagation to have an accurate audience and be able to trigger user needs, such as benefits, vanity, scarcity, trial, etc. For example, Didi and Meituan ’s red envelopes are shared with friends; paying users invite friends to try products for free; forwarding to friends’ circles to get gifts, etc.

Taking a successful WeChat 100-day running event as an example, this article demonstrates some adjustable points in the self-propagation process.

1. New user distribution mechanism

We use graded incentives to attract new running KOLs : for every 10 additional people in the team, we will send a group red envelope; when the team reaches 80 people, the team leader can get a pair of running shoes. At the same time, we do group operations in the team leaders’ group every day and post team rankings , such as “Team XX has 80 people” and “Captain of Team XX receives running shoes”, to fully motivate the captains.

2. Conventional sharing mechanism

In the WeChat system, sharing a poster is more eye-catching than sharing a link. The poster with the selling point of "Win an iPhone 8" makes it more eye-catching when users post it on WeChat Moments. At the same time, the sharing process should also be fully guided, such as "long press the picture to send it to a friend."

3. Inducement sharing mechanism

There is a registration fee for the event, so we designed a reward of "Share the event page to your circle of friends after successful registration and get 20 yuan in cash instantly." Because there is a common WeChat group among running users, it is most effective if users share it on their Moments. At the same time, we are also worried that users may choose to make part of their posts visible when posting to Moments, or delete them immediately after posting, so we added the mechanism of "10 people need to click on your share through Moments".

A. Detailed description of the sharing mechanism

B. Revise the sharing title to bring about secondary sharing. Anything that can be digitized can be made into a ranking list. Users will show off their number of registrations, which can stimulate the human desire to compare and show off, thus promoting sharing.

C. Use H5 to design "fake activity pictures and texts". On this H5, you can freely define the number of readings (directly 100,000+), the number of likes and user comments. Through the designed user messages, users are guided to sign up and some doubts are resolved.

E. "10 people clicked to read the reminder"

Each time someone clicks on the Moments, you will be reminded. At the same time, some people shared their Moments but not 10 people opened them, or they shared them to the wrong friends or groups, so every two days, we will send template messages to remind users who have not received the 20 yuan to post again.

2. What needs to be analyzed in data operations?

1. New user acquisition stage : pay attention to the type of user sources: pure new users (first registration) or old users (re-registration); how many users come from patch ads, how many users come from pop-up ads, etc.
2. Conversion stage : focus on conversion rate: 200 users browsed your promotional page, and 100 of them registered. These 100 people achieved conversion, and the conversion rate is 50% (=100/200). Similarly, in addition to the registration conversion rate, there is also the payment conversion rate, etc.
3. Active stage : focus on the user's activity within the product, which is expressed in different forms for different products. For example, Tieba: number of posts, number of replies, etc.; video websites: number of clicks , number of views, etc.
4. Retention stage : focus on the number of users who are retained or lost. For example, there are 300 new users on the first day, and 100 of them are still active on the second day. How many are still active on the third day? What about the fourth day? And so on.

User operation is just one of the functions of operation and runs through the operation of various products. The data indicators that user operations focus on have different emphases in different industries, platforms, etc.

According to the operating platform:

Website Operation:

(1) Traffic volume needs attention:

• PV (page view) is the data generated by accessing the page. If a user visits 5 pages, 5 PVs are generated.
• UV (user view) is the number of visitors to a specific page. No matter how many times an account clicks on a page, the UV is 1 because there is only one visitor.
• VV (visit view) refers to the number of visitors to the entire site. When an account enters a website, no matter how many web pages the account browses, VV is 1, because the website has only one visitor.
• IP: The network IP number for the entire site. You logged into this website using your home computer, and then your cousin also logged into his account using the same computer and visited the same website, but this time the IP address was still only 1, because you and your cousin used the same computer and the network had the same IP address.

(2) Access issues that require attention:

• Bounce rate: There are 300 visitors on the page, but 150 of them don’t like the page and choose to leave, so the bounce rate is 50% (=150/300)
• Second bounce rate: There are 300 visitors staying on the home page, and 150 of them like the website and click to browse the next page, then the second bounce rate is 50% (=150/300). By analogy, there are three-jump rate, four-jump rate, and so on.
• Conversion rate: The ratio of conversions to the final product destination page. If it is e-commerce , the ultimate goal is to place an order, so it is the ratio of new users and users who are converted to the order page. By analogy, there are also payment conversion rate, registration conversion rate, and so on.

(3) Active aspects need attention:

• DAU (daily active user) refers to the number of daily active users .
• MAU (monthly active user) refers to the number of monthly active users .

Related data may include weekly active users, annual active users, etc.

(4) Conversion needs attention: (The conversion here refers only to e-commerce operations. It is different from the conversion rate mentioned above)

• Order quantity: How many orders the user has placed in total
• Payment amount: How much the user paid in total
• Average order value: paid amount/number of orders = average order value. What is needed here is the average amount of money per order.
• Payment rate: The conversion rate that leads to payment

APP operation :

1. Newly added: the number of newly added devices (by phone model); the number of newly registered devices (registered new users.)
2. Active: number of active devices; number of active users
3. Retention:

Next-day retention rate : For example, if 300 new users logged in on the first day and 150 of them logged in on the second day, the next-day retention rate would be 50% (=150/300). By analogy, there is also the three-day retention rate (number of logins on the third day/number of new additions on the first day)...n-day retention rate, and so on.

TAD : For example, 7-day TAD = the number of users retained on the first day + the number of users still retained on the second day… + the number of users still retained on the seventh day

Used to calculate how many days a device has been active within seven days.

(4) Conversion: This also refers specifically to e-commerce, the same as the conversion in website operation mentioned above.

According to the industry of operation:

• Content-based industries: focus on PV, UV, VV, number of posts, page dwell time, number of shares , etc.
• Social industry: focus on the number of posts, number of comments, PV, UV, active percentage, etc.
• E-commerce industry: focus on sales revenue, order volume, average order value , etc.
• Game industry: focus on the number of active users, payment rate, revenue, ARPU (average revenue per user) , etc.

In addition to the two division angles of operation platform and operation industry, there are many other division angles, among which the data indicators that user operations need to pay attention to have different emphases.

3. How to perform data analysis

1. Data Collection

There are two basic principles for a good data source: completeness and detail.

Quan: That is to say, we need to have multiple data sources. We cannot just have one client data source. We don’t have the server data source or the database data source. If you don’t have these data, you may not be able to do the analysis. In addition, big data refers to the total amount, not a sample. We can’t just take data from certain provinces and then start talking about the situation across the country. Some provinces may be very special. For example, the clients in Xinjiang and Tibet may be very different from those in the mainland.

Detailed: In fact, it emphasizes multi-dimensionality. When collecting data, try to collect every dimension, attribute, and field. For example: if you collect information like where, who, and how, you will not be able to jump out of these selected dimensions when analyzing later, instead of focusing on demand at the beginning. Based on this demand, it is determined that certain data will be generated. When a new demand comes later, new data needs to be collected. At this time, the entire iteration cycle will be much slower and the efficiency will be much lower. Try to collect data from the source as much as possible.

2. Data Modeling

Once the data is available, it needs to be processed. The raw data cannot be directly exposed to the business analysts above, as it may be messy and not well logically abstracted. This is where data modeling comes in. First of all, let me mention a concept, which is the data model. Many people may feel afraid of the term data model, thinking that the model is something profound and complicated, but in fact it is very simple.

In the field of data analysis, especially in the analysis of user behavior , a relatively effective model currently is the multidimensional data model , the "online analytical processing" model. It contains two key concepts: one is dimension and the other is indicator.

Dimensions include cities, Beijing and Shanghai, some attributes to the west of the dimension, operating systems, iOS and Android, etc., and the attributes within the dimension. By crossing dimensions, we can look at some indicator issues, such as the number of users and sales, which are indicators. For example, through this model, you can see the overall sales of iOS users in Beijing.

3. Data analysis methods

There are many data analysis methods, such as multi-dimensional event analysis, funnel analysis (a simple analysis has been done in the previous part of the article), follow-up analysis, cross-analysis, etc. Here we will choose a cross-analysis to do a case analysis.

Cross-analysis method: usually combines vertical comparison and horizontal comparison to conduct multi-angle combined analysis of data. For example:

a. Cross-analysis perspective: client + time

From this data, we can see that the number of users on the iOS side is increasing each month, while the number on the Android side is decreasing. The main reason why the overall data has not increased is the decline in data on the Android side.

Next, we need to analyze why the number of new users on the Android side is declining in the second quarter? Generally, the channel dimension will be added at this time.

b. Cross-analysis perspective: client + time + channel

From this data, we can see that the proportion of pre-installed channel A on the Android side is relatively high and is showing a downward trend, while the changes in other channels are not obvious.

Therefore, it can be concluded that the decrease in new users on the Android side in the second quarter was mainly due to the reduction in the A pre-installation channel.

Therefore, the main function of cross-analysis is to segment the data from multiple angles and discover the specific reasons for data changes.

5. How to verify the effectiveness of new product features

Verifying the effectiveness of new product features requires simultaneous consideration of the following aspects:

a. Is the new feature popular?

Metric: Active ratio. That is: the number of active users using the new function / the number of active users in the same period.

The number of users will also be affected by many factors other than the function. Do not judge the quality of the function based on this indicator alone. You must make a comprehensive assessment in combination with the following other aspects.

b. Will users reuse it?

Metric: Reuse ratio. That is: the number of users who return on day N and continue to use the new feature / the number of users who used the new feature on the first day.

c. What is the optimization effect on process conversion rate?

Metrics to measure: Conversion rate and completion rate. Conversion rate is: the number of users who go to the next step / the number of users who go to the previous step. The completion rate is: the number of users who complete the function / the number of users who take the first step.

During this process, the conversion rate and completion rate can be analyzed using the funnel analysis method mentioned in the previous article.

d. Impact on retention?

Metric to measure: Retention rate. The proportion of users who return N days after the initial time, that is, the N-day retention rate. Commonly used indicators include: next-day retention rate, 7-day retention rate, 21-day retention rate, 30-day retention rate, etc.

e. How do users use the new features?

The actual user behavior trajectory is often much more complicated than the usage path we imagine. If the data monitoring platform used can see the relevant data, it can cause us to reflect on why they act this way and whether there is a simpler process to help us make optimization decisions.

No matter it is marketing, product, operation, or boss, everyone will have various data needs, so data operation is actually a very popular position, but it is not so easy to do well, because data is a more complicated thing, and there are many designed factor data indicators. However, as a product operator, you need to deal with data all the time, and it seems unacceptable not to have some data analysis skills, so basic data analysis skills are necessary.