7 steps to create a high-quality data analysis project

How can a data analyst not have a high-quality data analysis project? So how can you make a good project? This article provides 7 steps. Friends who see this article, hurry up and learn it!

At the end of every year, some students sigh: "After a busy year, it feels like all it's are routine data reports, and there's not even a project worth showing off!" So how should high-quality data analysis projects be done?

1. What is considered high quality?

To answer this question, we must first clarify what a "high-quality" project is. In essence, data analysis is a support position, and the quality of work is mainly determined by the department being served. If you work in a company, it mainly depends on the evaluation opinions of the management/business department. If you are interviewing, it is mainly evaluated by the interviewer HR/hiring leader. The key is to understand the other party's needs and hit the other party's pain points.

Students often get confused here, thinking that using a linear regression model (not knowing how to use complex models)/blingbling charts/writing 2,000 lines of SQL to look up a number are considered "high quality", but they ignore whether these things are useful to the business, and the result is naturally a joke.

A few days ago, a classmate came to me in a hurry and said that they had built a model to predict lost users. The operation said: "What are you doing? I don't know how to use it even if it's predicted!" Then the project fell through... This is the typical result of working in isolation.

So how can we address the business pain points?

2. Identify core needs

Data analysis is like the wind that sneaks into the night, moistening things silently. People often don’t think it’s very impressive when there is data, but when there is no data, some people will be anxious. So if you want to find the pain points of the business, it’s best not to force sales: "I have an artificial intelligence AlphaGo model, which is accurate in every test. Do you want to try it?" Instead, first see what the other department is most concerned about and what data they lack the most.

There are four common cases of missing data:

1. There is no basic data, and I am eager to see the data.
2. I have data but I don’t know how to interpret it. I’m so anxious.
3. I have data and interpretation, and I want to further verify my ideas.
4. I have data and interpretation, and I want to make further predictions.

When accepting business needs, you must be clear about the real needs. For example, "user portrait" may be just a word when starting a project. Whether the business is not clear about the current situation of users or wants to do something based on the portrait, you must understand it clearly. If it is not clear at the beginning of the project, it must be gradually clarified in the middle process. Otherwise, if you put a lot of labels and then people question you, "What's the use of this!", you will be dumb...

III. Key points of report-based projects

Report-based projects are the most numerous, but they are most easily overlooked by data analysts. Many newcomers always dislike that they are not technically complex. But in fact, report-based projects are the easiest to achieve results. The key is to do what leaders care about and what leaders can see. When receiving requirements, distinguish the report users and give priority to visualizing the leaders' requirements so that leaders can intuitively feel the data.

In addition, through report-based projects, the cooperation attitude of the business party can be effectively identified. If the business party has a good attitude, then in-depth cooperation can be achieved. Since there is already a business monitoring report, the next step is to analyze the abnormal business trend. First record the abnormal points caused by non-business active behaviors, and then conduct in-depth analysis:

• How big a change is considered abnormal?
• What factors cause the abnormality?
• How to detect anomalies in advance through data

With these accumulations, we can further perform automated exception reminders + problem diagnosis, which can take simple data display to a higher level and lay a solid foundation for subsequent in-depth analysis.

4. Key points of analytical projects

In many people’s original impression, data analysis is to get a bunch of numbers, and then Prajna Mommy will analyze them and tell the business three sentences to help the business earn an extra 180,000!

Therefore, people often have high expectations for analytical projects. But in fact, analytical projects are particularly prone to failure. Insufficient understanding of the business, lack of monitoring data, and lack of experience in abnormal analysis will make the problem analysis superficial. It is common to have "loud thunder but little rain" when doing projects.

Therefore, analytical projects are hatched on the basis of report-based projects, and the success rate is relatively high. If it is found that the business side does not monitor the problem itself enough and has a unclear understanding, it can be returned to the report-based project. After a certain accumulation, if you want to see results, the best way is to first reach a consensus on business assumptions and figure out what the business side is not confident about and what they are confident about. It is easier to disprove than to prove, and it is easier to get results by directly testing assumptions.

If the problem involves too many difficult and complicated problems that are difficult to quantify, there is another solution, which is to transform the problem into a test project. Look directly at what solutions the business side has at hand, and then test which solution works. This can also output a solution to the problem.

5. Key points of test projects

Testing projects are relatively easy to succeed. In essence, testing is also a case of "there is no business data, but we really want to see some data." However, it should be noted that you need to think clearly about what you want to test in advance. The most important thing in testing is to have an early understanding of the factors that affect the results, test the key factors you want to test, and control other interference factors.

Therefore, it is easy to succeed in general page design tests, but it is easy to mess up the test of consumption results. Because there are fewer test points in page design, it is easy to get accurate and stable results. However, there are too many factors that affect consumption results. If you don’t think clearly before doing the test, it is easy to cause the results to fail due to the low comparability between test plans, large differences in the test groups, and failure to eliminate key interference factors.

Therefore, before doing the test, basic analysis work is very necessary. It is necessary to sort out which factors will have an impact and how big the differences are between several sets of test solutions, which can effectively improve the quality of the project.

6. Key points of predictive projects

The key to predictive projects is to confirm the real prediction needs and avoid blindly gambling with "I want 100% accuracy". Not only is it impossible, but it is also meaningless. For example, in the prediction of lost users mentioned at the beginning, if the operation is to invest all resources to recall lost users, then change the goal to prediction: "Which people will naturally return" to save money.

If the operation wants to achieve the maximum effect, the goal can be changed to: "What kind of recall method the user is expected to respond to" so that multiple rounds of push can be made to maximize the awakening of users. In short, it is much more effective to understand the operation plan before taking action than to build a model behind closed doors.

7. A sense of ritual is important

Data analysis projects especially require a sense of ceremony! Because data analysis results rarely become hard cash, and by the time the year-end summary is made, everyone may have almost forgotten it. Therefore, you must have a full ceremony. For example, when the project starts, hold a meeting with the partners, and have a lunch together at the round table at noon.

When reporting at the end of the project, I specifically invited my boss to take a group photo. I tried to put the project results on BI as much as possible, and made a "Data Assets Screen" to be placed in the boss's office. I scrolled it every week to see how much new data was accumulated and how much benefit the business generated. Of course, I had to put up the "Double Eleven Operations Screen", and remember to take photos to record the grand occasion of my colleagues cheering in front of the screen...

There are many specific methods, and you can adopt them according to your own company's style, but there is only one core idea, which is to unite colleagues more, hold more meetings, and use the system more often. Don't just submit a PPT, an Excel, or a CSV silently. The inside has been worked hard, and the appearance must be done well!

Having written this, some students must be wondering: "Then how can we demonstrate the high quality of the project during the interview?" The thinking during the interview is different from the actual work, because there is a cognitive gap during the interview. It is possible that the business party in the actual work likes something very much, but the other company thinks: "This is very ordinary, nothing special!" If you want to impress the interviewer, you need another routine.

Author: Down-to-earth Teacher Chen

Source: WeChat public account "Down-to-earth Teacher Chen (ID: gh_abf29df6ada8)"