about the author
Simba
IBM senior business analyst
IT workplace veteran
Lifelong learner
This article is about 5,000 words. It takes 8 minutes for skimming and 15 minutes for intensive reading. Welcome to come back anytime.
Introduction
"A simple number has no soul, but the business behind it is fresh." However,
every time we encounter a new business scenario, we must overthrow the data analysis knowledge we have learned?
I now know how to measure my own products, but how are others' products measured? What can be improved in my product measurement method?
This is a question that lingers in the mind of the product manager Xiao P. Xiao P wants to make more friends for data analysis, so he joins a group of "data people's private land" (yes, this is a soft advertisement, click on the top The data person’s reserved place to pay attention to the official account, or scan the code at the end of the article to join the group), here is found someone is asking the question of the choice of data analysis "tool". Correct! "Tools", since there is no way to check all products, but I can check most of the data analysis tools on the market, and the ability of data analysis tools is a highly abstract product measurement method. So Xiao P started his physical (steal) inspection (acquisition) journey, exploring the high abstraction of product measurement methods from concrete tools-that is, the "routine" of data analysis.
As the product manager of an international B2B SAAS platform, Xiao P's tools are also all over the world:
Google Analytics-can be regarded as the bigwig of website analysis.
Amplitude-This is known as a Silicon Valley unicorn, but it is not well-known in China.
Baidu Statistics-the equivalent of Google Analytics in China.
Shence Analysis-Only established in 2015, but the market is growing rapidly.
Zhuge IO-so-called "in-depth analysis of user behavior data in business scenarios".
Growing IO-a leading one-stop service provider of overall digital growth solutions in China.
The "routine" of data analysis
After several weeks of experience journey, Xiao P summed up his "travel strategy": data analysis is ultimately to help the business understand several philosophical issues:
Who are they? ---- Know your users in all directions.
Where do they come from? ----How to implement the product Go To Market strategy to bring the product to the market.
What did they (coming to you) do? ----Measure and improve product capabilities from user behavior.
Where are they going? ----Analyze the flow and predict.
For example, in the B2B SAAS e-commerce platform that Xiao P is responsible for:
Who are they? — The buyers and sellers on the platform.
Where do they come from?-This platform advertises in different channels. We need to know which channel has more people, where customers are better, and which channel is more cost-effective.
What did they (come to you) do? — The customer came, what pages did the buyer read, and did they buy in the end?
Where are they going? —Is the user process smooth, do most customers go through the process as you designed, do they jump to some external sites, and some tools can also predict the likelihood that customers with certain behaviors will purchase the product How big.
It doesn't sound difficult, but there are many pitfalls in further exploration. Next, let's take a look at how to define these philosophical issues in data analysis with Xiao P.
1. Who are they? Know your users in all directions
The performance in google analytics/Growing IO神策/ is as follows:
It is easy to see that these tools will focus on several indicators related to
users : the number of active users
Number of new users
X-day retention rate
First look at what is a "user"
The common ones are as follows:
Treat the device as a user unit. For example, the concept of users in Google analytics is defined in terms of Cookie ID. The same user who logs in on the mobile phone, computer, or pad will be regarded as different users.
Regard logged in members as users. If we look at the user according to the first definition, it may be extremely misleading for some products. For example, a customer views an email notification on the mobile phone and needs to reply to the email. The email content is on the PC, so he turns on the computer , Replied to the email. At this time, if there is an indicator to evaluate the rate of email responses, then the user on the mobile phone will not have the action of replying to the email, and the user on the PC will not have the action of checking the email notification. At this time, it is necessary to use the logged-in user as the user to count the rate of email responses.
Defined based on the customer organization. Note that this is a customer, not a user. It is common in B-end products. For example, if a company buys Slack, a company that has 100 people, these 100 people are product users, and the company itself is the product customer. For a while, Slack defined the number of new/active accounts as the North Star indicator. The number of accounts here is defined on a customer basis.
After reading the concept of "user", let's look at "time definition"
Define time period
For example, daily activity, new additions, and retention on the 7th, what about this "day"? There are generally two types:
one according to the natural day
One is based on 24 hours as a day.
For example, an App is generally used at the beginning of work (similar to an attendance check-in tool). We pull a new activity on the first day, and the user logs in at 8 o'clock in the morning at work time the next morning, although it is not satisfied at this time. Hour, we still need to define this user as day1 retained user, then it is calculated according to the natural day. If there is no such restriction on the use of a specific time, or an internationalized product, we can define a day as 24 hours.
Let's look at "active, new, and retained."
How to define "active, new, retained"
active
Generally speaking, "active" generally means that as long as the user has access records, it can be considered active. For example, the definition of active in google analytics is:
"ActiveUsers: the number of unique users who initiated sessions on your site or app on a specific day." (The total number of users who have generated session records on a specific day).
In Shence analysis, users can also customize "active" users as users who satisfy certain behavior records. For example, on the creator side of content tools, we can define it as "there is a record of published content". In social tools, a communication record is defined as active.
Add
This concept is closely related to the user's definition. If the device is regarded as a user unit, if the same user logs in with different devices, or logs in with different devices, but the browser cache is cleared, the cookie ID will be updated, then it is also counted as a new user. The same goes for the other two user-defined cases.
Keep
In layman's terms, retention is the process of pulling in new ones. You will find that after a period of time, there will always be a loss of users, and retention is a measure of how many users stay.
Little P is thinking that my product is a B2B platform, the seller and the buyer are both corporate customers, and the buying behavior is also low-frequency, and the retention rate does not seem to be suitable for my product. He went to ask a boss who kept his focus.
Big brother: "Why do you want to look at retention?"
Little P: "I want to see how many customers are left for your product"
Big brother: "Why do you want to see how many customers are left?"
Little P: "Look at my product Is it attractive enough?"
Big guy: "Then, what if the retention rate is not high?"
Little P: "Thinking gets higher."
Big guy: "How to get higher?"
Little P: "Uh, uh, this is necessary See specifically..." Little P planned to tell the boss about the product methodology, but was stopped by the boss.
Big Brother: "Stop, retention. In addition to measuring the attractiveness of the product, it is more important to help you discover two elements:
1. Core customers, that is, which customers are more likely to stay, and whether the follow-up should strengthen the service work for core users.
2. Aha Moment, that is, what behaviors are important behaviors that promote retention and conversion, and whether support for this behavior should be strengthened in the future. You can look at Facebook's case of finding aha moment to promote growth.
Always remember, Data is not the end, but the starting point...."
Little P Daigo empowered: "So for the product of the B2B platform I am responsible for, through retention analysis, I found that those who come frequently, although they may be few, can study their Behavior to enhance the improvement of certain functions. It can also help discover whether there is an aha moment that facilitates the order."
Big Brother: "Smart! And in the second half of the Internet, retention will become more and more important. The past pirate growth model AARRR has gradually evolved into the current RARRA growth model, which means that you must first retain, polish the product, and then detonate. As the first important growth factor, Retention is already very important."
Little P: "Wow, I learned a lot of new knowledge."
Just after Xiao P figured out why he wanted to keep it the next day, a small partner in charge of operations came to Xiao P. We have a SAAS seller here who made such a request. They want to know who applied for a trial of SAAS products. Of users, how are they trying it out, and see how they can increase the conversion rate from the trial to the paid version. By the way, their product is a CRM system specifically for sales. You should use it daily. Think about how to do it.
"Isn't this just watching retention?" Xiao P secretly overjoyed. He immediately went to see the definition of retention in the data analysis tool. The following is the definition method given in the Shence demo:
Therefore, in the retention analysis, we need to define the following parameters:
1. Define an initial behavior, that is, what the new behavior is. In this example, the small P is defined as the "application trial" behavior.
2. Define a follow-up behavior, that is, how to determine "the user has stayed". In this example, the small P is defined as "login to access the CRM software"
3．Define a time frequency, by day or week. Operations said that this CRM system is a "daily" tool for sales, so we look at the retention of each "day".
4. Define a time period, that is, how many days to observe the retention. The trial period of this software is 7 days, so define it as 7 days.
After defining these, how is retention calculated? Take a chestnut: There is an activity that encourages trial for 2 days. Let's look at the calculation of the 3-day retention rate.
· On October 1, 1,000 people tried it out; on the first day, 300 of these 1,000 people logged into the CRM software; on the second day, 200 people logged into the CRM software; on the third day, 150 people logged into the CRM system.
· On October 2, 1500 people tried it out; 400 people logged in to CRM software on the first day; 200 people logged in to CRM software on the second day; 150 people logged in to the CRM system on the third day.
Then:
· Day1 retention rate is (300+400)/(1000+1500)
· Day2 retention rate is (200+200)/(1000+1500)
· Day3 retention rate is (150+150) /(1000+1500)
Assuming that after the calculation, we get such a chart, we can see that the user retention dropped sharply on the first day, and then stabilized, we can analyze what the stabilized batch of users did, and there is nothing left. The batch users did not do anything, can it help discover the core users and aha moments?
Half a year later, the B2B platform that Xiao P is responsible for introduced retention analysis tools for the SAAS sellers on the platform. Everyone praised: "What we want to do, your platform did it for us, and it can finally help us see what it is Why the conversion rate of our product’s trial-to-pay conversion is low. Look at what the remaining customers do, and in turn, you can see what behaviors can make customers stay. It’s great!”.
The pride in Xiao P's heart spontaneously arises.
The three indicators are clear, and the small P also found that there is an important function in the major tools, called "user grouping".
User grouping:
Classify them
When users come, different users need to be treated differently, and users need to be grouped, which is also a prerequisite for refined operations. Let's first look at the performance in Zhuge IO and amplitude.
Zhuge IO:
Amplitude
The definition of user group in Zhuge analysis is based on: done or not done, added to...when, active in...when, user attributes satisfy.... In addition to the similarities above, there is also a "Have prospensity" in amplitude. What is prospensity?
We first classify users into groups based on characteristics, roughly divided into attribute characteristics, behavior characteristics, and prediction characteristics. These three characteristics are progressively advanced according to the depth of analysis.
Attribute characteristics: Mainly objective factual characteristics of customers, which will not change with our analysis methods. The C terminal has gender, age, name, and place of residence. B-end products include company size, industry, etc.
Behavioral characteristics: part of the user behavior itself, but also contains statistical characteristics for the behavior, such as how many times a certain page has been visited, when was it active, etc. The definition of these characteristics is closely related to "what the user did" Connect, don’t worry, we can describe the behavior in detail later.
Predictive features: "have propensity" corresponding to Amplitude. In Amplitudesupport, the method of obtaining propensity is described as follows: Predictions use past behavior to predict future behavior. Amplitude starts by looking at users who were in the starting cohort two periods ago, and will then identify which of those users did vs did not perform the action one period ago(a period can be set to seven, 30, 60, or 90 days). Next, Amplitude comparesthose two groups of users along three sets of variables. See the original text for details (https://help.amplitude.com/ hc/en-us/articles/360049161832)
For example, in the e-commerce website, we divide the customers who paid to place orders and the users who did not pay to place orders in a certain period of time in history. The behaviors, behavior attributes and user attributes of these two types of customers analysis. We use this behavior to train a linear regression model, and then use this model to predict the likelihood of customers placing orders and paying.
It has to be said that the intelligent forecasting function must be a charming attribute of data analysis tools in the future, and tools with such capabilities are also in the forefront of analysis tools. The data analysis tool itself does not own the data, but it is very commendable to provide such capabilities.
Of course, user grouping is just the beginning. After all, the purpose of grouping is to serve the business. It is the ultimate goal to analyze different groups of people and take actions (such as sending pushes and improving products for core groups).
So far, we finally know who "they" are, so where do they come from? You need more information.
2. Where do they come from?
In other words, traffic is commonly referred to. Let’s take a look at the display of traffic in Google Analytics and Baidu Statistics:
The reason for comparing Google Analytics with Baidu Statistics is that both are search engines, and both can use the advantages of their own search engines to tell customers: what keywords customers often search for come to your website. As shown below:
In fact, the initial formation of Google analytics also stems from a demand within Google: how do we prove to customers that Google advertising brings traffic to customers. For customers, they also need to know whether advertising is useful, and where they can get more high-quality customers, so as to help formulate the strategy of product marketing. There is a role in Silicon Valley’s occupational classification called PMO (Product Marketing Owener). Similar to domestic operations jobs, traffic is their most concerned content.
The traffic itself is easier to understand, that is, where your users come from, but you still need to distinguish a concept, namely Source and media. For example, if you made an advertisement on CSDN, and someone recommended your website inside the CSDN website, then you need to know whether the advertisement is working or the recommendation of others is working, so in google analytics, just With the concept of source and media.
Source: the origin of traffic
Media: the medium of communication
To figure out the distribution of traffic, Xiao P firmly remembered what the big guy said "data is only the starting point, not just the end point", then where can this starting point lead us? If you find that there are too many customers from a certain channel, yes Shouldn't we increase the promotion of this channel? Xiao P found that the bounce rate and conversions of different channels were also provided in Google analysis, and the conversion rate analysis chart was also provided in the channel analysis of Shence.
It seems that we should not only look at the number of passenger flows brought by channels, but also the quality of passenger flow. The bounce rate and conversion rate mentioned above are two important indicators for measuring channel quality. They are defined as follows:
Bounce rate: There is a unified definition in the industry, that is, how many people come to the page (here more of the landing page) and leave without any action. For example, if someone advertises at the subway entrance saying "50% off in a store," you go into the store as soon as you hear the 50% off, but when you enter the door, you find that it is a tobacco and liquor store, and you never smoke. Drinking, in this case, you will basically turn around immediately and stop exploring. Your behavior is just a record. There are many reasons for the high bounce rate. However, if multiple channels are compared, some channels have a high bounce rate and some channels have a low bounce rate, then the reason for the high bounce rate is mostly that the product placement channel is inappropriate. If multiple channels have high bounce rates, it may be that the landing page itself is not good. It’s worth noting that in most cases, the bounce rate is generally used to measure the landing page, and is not suitable for measuring the quality of all pages.
Conversion rate: How many people successfully converted after they arrived. Whether the customer’s needs are urgent and true depends not only on what he browses, but also on what he does. The success of the conversion is the actual behavior of the customer (such as paying and registering). If multiple channels are compared, and some have high conversion rates and some have low conversion rates, then it can also be used as a delivery channel for high quality. Therefore, the definition of conversion rate must be supported by behavior. For example, in Shence analysis, you can define the behavior of funnel transformation.
This reminds Xiao P that they just started to promote their B2B platform a long time ago. They used Google search and offline SAAS vendor summits to promote it. Obviously, the channel quality of offline summits will be much higher. Of course, the cost of offline summits will be much higher. But no one can tell which is more cost-effective. If you can do such a bounce rate and conversion rate analysis at that time, you can give a rough cost and benefit.
So far, Xiao P probably figured out who the user is, where the user came from, what the user did next, and where the user went. Little P will take you through this "exploratory journey of data analysis", so stay tuned.
[To be continued]
Reference
Google analytics support document: https://support.google.com/analytics/?hl=en
Amplitude technical documentation: https://help.amplitude.com/
Shence Data Official Website: https://www.sensorsdata.cn/
Zhuge IO official website: https://zhugeio.com/
Baidu Statistics Official Website: https://tongji.baidu.com/web/welcome/login
Growing IO official website: https://www.growingio.com/
The private place of a data person is a big family that helps the data person grow, helping partners who are interested in data to clarify the learning direction and accurately improve their skills. Follow me and take you to explore the magical mysteries of data
1. Go back to "Data Products" and get <Interview Questions for Data Products from Big Factory>
2. Go back to "Data Center" and get <Dachang Data Center Information>
3. Go back to "Business Analysis" and get <Dachang Business Analysis Interview Questions>;
4. Go back to "make friends", join the exchange group, and get to know more data partners.