How is a good data analyst tempered?

In recent years, it is no accident that Internet companies have more and more demand for data analyst positions.

In the past ten years, the Internet industry in China has grown wildly by relying on demographic dividends and traffic dividends; With the increase of traffic acquisition cost and the decline of operational efficiency, this extensive business model is no longer feasible. Internet companies urgently need to achieve refined operations through data analysis, reduce costs and improve efficiency; This also puts forward higher requirements for data analysts.

This paper will share with you the evolution of data analysts, the value system of data analysis, the four necessary abilities of data analysts, seven common ideas and practical analysis cases.

First, the past lives of data analysts

Before introducing data analysts, let's take a look at these historical figures and see how they are related to data analysts.

Famous "analyst" in history

The six historical figures shown above (from left to right and from top to bottom) are Sean, Guan Zhong, Xiao He, Bin Sun, Guiguzi and Zhuge Liang. They are all famous counselors in history, and some of them have also served as prime ministers. They read widely and have a unique vision. They discovered many laws by summarizing a large number of historical facts and successfully predicted many events in practice. Through the practice of "historical statistics-summary analysis-prediction of the future", they have created great value for their own organizations, which is the predecessor of "data analysts".

Now, what are the necessary skills of a data analyst and how to become an excellent data analyst?

Second, the value pyramid of data analysts

A complete enterprise data analysis system involves many links: collection, cleaning, conversion, storage, visualization, analysis and decision-making. Among them, the work content of different links is different, and the time consumed and the value generated are also far from each other.

Data analysis value pyramid

The data analysis system of Internet enterprises has at least three aspects of data: user behavior data, transaction order data and CRM data. Engineers collect data from different sources, and then unify them on the data platform through cleaning and conversion. Then professional data engineers put forward data from the data platform. These tasks occupy 90% of the time of the whole link, but the value generated only accounts for 10%.

The data analysis on this pyramid is closely combined with the actual business, supporting the business decision-making of enterprises in the form of reports and visualization, covering all front-line departments such as products, operations, markets, sales and customer service. This part only takes up 10% of the whole link, but it can produce 90% of the value.

An excellent business data analyst should be value-oriented, closely combine the practice of product, operation, sales and customer support, support all business lines to find and solve problems, and create more value.

Three, the four essential abilities of data analysts

Four basic skills of data analyst

1. Global View

One day, the product manager found me and asked me: Hello, can you help me look at the data sent by the new function of the product yesterday? Thank you. I will say, OK, I'll give it to you right away! But I asked politely: Why do you need these data? The product manager replied: Oh, the new function went online yesterday, and I want to see the effect. Knowing the purpose of the product manager, I can extract and analyze the data in a targeted manner, and the analysis results and suggestions will be more operable.

Many times, data analysts can't just count and get stuck in various reports. An excellent data analyst should have a global view, take a step back and ask more questions when encountering analysis requirements, and better understand the problem background and analysis objectives.

2. Professionalization

Data scientists in enterprises model and predict the user churn, and the final prediction accuracy of the user churn model is over 90%. The accuracy is so high that business analysts can't believe that after testing, it is found that one independent variable in the data scientist's model is "whether the user clicked the cancel button". Clicking the "Cancel" button is an important sign of user churn, and users who have done this action will basically be shoddy. Using this independent variable to predict the loss has no commercial significance and maneuverability.

Data analysts should demonstrate their professionalism in the industry (such as e-commerce, O2O, social networks, media, SaaS, mutual funds, etc.). ), be familiar with the business process and the meaning behind the data in her/his industry, and avoid the data jokes above.

Step 3 imagine

The business environment changes faster and more complex, and the influencing factors behind a set of business data are unimaginable to ordinary people. Data analysts should use their imagination, make bold innovations and make assumptions on the basis of work experience.

Step 4 trust

Take the sales position as an example, a salesperson must first establish trust with users; If users don't trust you, it's hard for them to trust or buy your products. Similarly, data analysts should also establish good interpersonal relationships with colleagues in various departments and form certain trust. Colleagues in all departments trust you, and they may be more likely to accept your analysis conclusions and suggestions; Otherwise, get twice the result with half the effort.

Four, seven common data analysis ideas

1. Simple trend

By obtaining the trend in real time, we can understand the usage of products, which is convenient for the rapid iteration of products. The number of users, access sources and user behavior are of great significance to trend analysis.

Minute-level real-time trend

Comparison with the trend of weekly cycle

2. Multidimensional decomposition

Data analysts can decompose indicators from multiple dimensions according to the analysis needs. Such as browser type, operating system type, access source, advertising source, region, website/mobile phone application, device brand, app version and so on.

Multidimensional analysis of access user attributes

3. Conversion funnel

According to the known transformation path, the transformation of the whole and each step is analyzed with the help of funnel model. Common transformation scenarios include registration transformation analysis and purchase transformation analysis.

Funnel analysis shows the wastage rate of each step of registration.

4. User grouping

In the refined analysis, it is often necessary to analyze and compare the user groups with certain behaviors; Data analysts need to take multi-dimension and multi-index as grouping conditions to optimize products and improve user experience.

5. Check the path carefully

Data analysts can observe the user's behavior trajectory and explore the interaction process between users and products; Then find the problem, inspire or test the hypothesis.

Analyze the user's behavior law by carefully looking at the path.

6. Residue analysis

Retention analysis is to explore the relationship between user behavior and return visit. Generally speaking, the retention rate we refer to refers to the proportion of "new users" who "return to the website /app" within a period of time. Data analysts find the growth point of products by analyzing the retention differences of different user groups and users who have used different functions.

Retention analysis shows that users who "create charts" have higher retention.

7.A/B test

A/B testing is to test several schemes in parallel at the same time, but only one variable of each scheme is different. Then the best scheme is selected through some rules (such as user experience, data indicators, etc.). In this process, data analysts need to select reasonable grouping samples, monitoring data indicators, post-event data analysis and evaluation of different schemes.

Verb (abbreviation of verb) is a practical case of data analysis.

A social platform has launched an advanced payment function, which is pushed to the target users in the form of EDM (direct mail). Users can directly click on the link in the email to complete the registration. The registered conversion rate of this channel has been between 10%-20%; However, in late August, the registration conversion rate dropped sharply, even less than 5%.

If you were a data analyst in this company, how would you analyze this problem? In other words, what factors may cause the EDM conversion rate to plummet?

An excellent data analyst should have a global vision and professionalism, proceed from the business reality and synthesize all possibilities. Therefore, the possibility of a sudden drop in EDM registration conversion rate is as follows:

1. Technical reasons: ETL is delayed or failed, resulting in the loss of front-end registration data and a significant drop in registration conversion rate;

2. External factors: whether there are holidays at this time node and whether other departments have sent promotional emails to users recently may dilute users' attention;

3. Internal factors: whether the copy and design of the email have changed; Whether the arrival rate, opening rate and click rate of the mail are normal; Whether the mail registration process is smooth.

After one-by-one investigation, data analysts locked the reason in the registration process: the product manager added the content of binding credit cards in the registration process, which led to a significant decline in the user's willingness to submit registration and a significant drop in the conversion rate.

Behind a seemingly simple conversion rate analysis problem is the embodiment of the ability of data analysts in all aspects. The first is the technical level, the understanding and understanding of ETL (data extraction-conversion-loading); In fact, it is a global view, with a clear understanding of the business at the seasonal and company levels; Finally, I am professional, and I know the process and design of EDM business like the back of my hand.

The power of practicing data analysis is not achieved overnight, but constantly grows and sublimates in practice. An excellent data analyst should be value-oriented, look at the overall situation, be based on business, be kind to others, and drive growth with data.