How to do an excellent data analysis project?

To do a good data analysis project, it is not just "I run some data", but a more convincing output than that. So, what is the first step? Let's take a look at the summary of this article.

First of all, everyone should understand that not all projects need to be held in a large hall with 10,000 people, with banners, and the chairman and general manager taking turns to come on stage to ring the gong. As long as they meet the requirement of "having specific outputs at a specific time and under specific conditions", they are all projects.

Therefore, the key to doing a project is not to seek a name, but to have specific outputs. With the output of specific products, KPI/OKR documents are easier to submit; the boss’s satisfaction with you increases; you have more capital when you are evaluated for promotion; and you have more things to write on your resume when you change jobs. This is what we should strive for. The so-called "excellent" project refers to a more convincing output than "I ran some data".

So, where should we start?

1. Understanding the Service Targets

When doing a project, the most important thing is to figure out the goal; after figuring out the goal, the first step is to figure out who you are serving. This is the biggest difference between data analysis novices and veterans.

Often, newbies who have not yet entered the industry have their minds full of "templates, models, and formulas". They think that as long as they copy the template, they have completed the work. Newbies who have just entered the industry like to say "business" in general. But business is not a solitary, independent individual. Behind the word "business" is a very specific and complex meaning (as shown in the figure below).

Analyzing specific problems in specific ways is the most basic requirement for data analysis and the first step to completing a project. Because these five elements and their specific forms determine the extent to which our data analysis can be done, what it should look like, and what it should look like to meet the needs. The specific relationship is shown in the figure below:

It is very important to clarify the specific issues. In the past, we often talked about how traditional enterprises are and how Internet enterprises are. Today, with the development of channel integration, the boundary between the two is actually becoming more and more blurred. If we do not make a specific analysis, we will often make a lot of jokes.

for example:

It used to be a toC Internet company, but now it wants to focus on toB, and has no idea how to deal with customers;

It is called an Internet product, but its service targets are physical owners, and sales are still done through the most primitive outbound phone calls;

Although it is called the Internet industry, the products that can be operated are still physical products, and the purchase, sales, inventory and profit are all good;

It’s called new retail, but data collection is a mess, even worse than traditional chain stores;

Traditional enterprises can undergo digital transformation, focusing on distribution and fission.

The above complex scenarios cannot be solved by simply saying "I have Internet AARRR thinking". Expecting to use templates will lead to failure. Moreover, after several years of experience, many operations, product managers, and planners have learned basic data analysis concepts. At this time, data analysts who still hold on to PPT templates full of empty slogans such as "SOWT, PEST, 5w2h" will be laid off. Specific problems require specific analysis, which cannot be overemphasized.

Moreover, understanding the situation clearly is very important for seizing opportunities in the next step. If you wait for the business to come to you before doing anything, you are no different from a dog catching a Frisbee (the business proposes a hypothesis, and the data verifies the hypothesis, just like a dog catching a Frisbee). Only when you have a good judgement of the situation can you proactively discover opportunities.

2. Find the right time to exert force

The biggest enemy of data analysis projects is daily work. Therefore, not everything is suitable for project development. Timing is very important.

Often we have to start with the following opportunities in the business department:

• Want to innovate
• Want to improve the current
• I'm completely clueless about my new job
• When faced with a problem
• Three axes were used but no effect

At these moments of opportunity, throw out the system solution and solve the problem independently in one go (as shown in the figure below):

3. Confirm project requirements

After finding the right time to start, negotiate with the specific business party and prepare to start work. Before starting work, you must confirm the project requirements, specifically the project iron triangle:

There are three points to note here:

1. Numbers, models, and reports are not outputs

The business goes from not understanding the situation to understanding it, from having no solution to having a solution, from not knowing how to choose to knowing how to choose, from not being prepared to having first, second and third level emergency plans. This is the output. So don't just talk about numbers without considering the problem. Draw conclusions from the numbers.

2. Don’t forget the time

If time is tight, try to come to a conclusion as quickly as possible; if time is wide, you need to output it in steps. A company is not like a school that gives you half a year to slowly finish your paper.

3. How much rice should be put into the pot?

If the data quality is poor, there is a shortage of manpower, and there is a lack of analytical experience, just stay calm and do it step by step. Don't expect to solve all problems at once.

These three points are crucial to project results. In the past, too many data analysts have been obsessed with the "scientific method", ignoring project management and time investment. As a result, they drew a big pie but fried a small pie, and ended up in disgrace.

Here we should also pay attention to the working method. Confirming the requirements does not mean that you should directly ask the business: "What do you want to analyze?" This way of asking is too passive, and it is back to the old way of catching frisbees. And often the business will answer you in a way that is confusing.

For example: Please help me find a solution (ask you for an implementation plan)

It must be the opponent/the weather/luck... (trying to pass the buck)

I want to analyze user mental resources (there is no data at all)

As long as there is artificial intelligence, users will pay for it (unrealistic method)

Therefore, the reliable approach is to sort out the needs step by step, guide them to the problems that can be solved by data analysis, get to the root of the problem, and objectively solve the problem (as shown below). There are many details to talk about the specific guidance method, and we will talk about it in detail later with specific cases.

4. Conduct analysis

After completing the requirements, the next step is formal work. The specific content is related to the analysis topic, so I won’t go into details here. If the early stage is clear, the intermediate process will naturally go smoothly. Here I would like to emphasize one point: when doing data analysis, remember not to hold back on your big moves. The longer you hold back, the higher people’s expectations of you will be, and the more disappointed you will be in the end.

Therefore, as long as the project duration exceeds one week, there must be a weekly report to inform everyone of the progress; if the duration exceeds one month, there must be a monthly summary to discuss the progress with everyone.

Especially for projects that use algorithms, when the business department hears about the algorithm, they often think it is a magical weapon that has descended to earth, and it will be invincible wherever it goes. Therefore, there are many examples of algorithm projects dying due to overly high business expectations. During the process, the specific algorithm process does not need to be reported to the business, but the difficulties encountered and the expected output conclusions should be communicated frequently to appropriately control business expectations to avoid discovering that the goods are not what they expected at the last minute, and finally being ruined.

V. Work Report

I won’t elaborate on this here. Teacher Chen has updated a series of data analysis reports. You can follow the official account and read them in the menu bar. In short, when reporting, you should consider the identity and purpose of the target audience and make personalized reports based on the project goals. This is the only way to achieve good results (as shown below).

Based on the audience's thinking, even the same data and the same conclusion can be expressed in different forms, which ultimately catches the audience's attention, makes everyone interested, and brings the project to a perfect conclusion.

VI. Summary

Looking at the whole process, we can see that the process of completing a project is the process of applying data methods to corporate practice. Data itself has professional knowledge such as statistics, mathematics, programming, and databases, but a considerable part of it (such as data warehouses and ETL) is to ensure the normal operation of the data itself; a considerable part (such as semantic judgment and image recognition) is used for industrial applications, without considering business understanding and cooperation; a considerable part (such as statistics) is suitable for scientific experiments and agricultural, forestry, animal husbandry and fish research.

A large number of businesses are not scientific problems, but practical problems. How O2O platforms manage merchants, how new media platforms develop local customers, how live e-commerce chooses products, etc., all require combining data knowledge with practical work to output conclusions.

Not to mention, everyone in the workplace is involved in office politics, and the desire to stand out and avoid being blamed. That’s why we have today’s discussion and various ways to promote projects. This is a necessary step for every data student to go from campus to the workplace.

Author: Down-to-earth Teacher Chen WeChat public account: Down-to-earth Teacher Chen