What is a data analysis project and how to do it?

Many students who are responsible for data will want to independently take charge of a data analysis project. This article explains in detail what a data analysis project is and where the crux lies. It is recommended for students who want to learn about data analysis.

"I don't want to run a computer anymore, I want to work on a project." Many students who work on data have this strong desire. It is really bad to run data mechanically every day without knowing what the data is used for. Everyone wants to have the opportunity to be responsible for a project independently.

However, many students are full of doubts:

• What is a data analysis project?
• Why haven’t I seen data analysis projects in my company?
• I searched the codes of Titanic, Taobao e-commerce, and American credit cards on Baidu, and pressed ctrl C+ctrl V once. Does this count as a project?

Let’s start today with the first question: What is a data analysis project?

1. What is a project?

The meaning of a project is to organize human and material resources within a specific time limit to achieve a specific output target. Although this sentence is simple, it brings out the three key dimensions of a project: time, cost, and output quality. These three dimensions are commonly known as the "project iron triangle."

It is called a project because it corresponds to regular work.

• Project: specific time, specific people, specific output.
• Routine work: do it every day, with a fixed group of people, and just get the job done.

You see, it’s the same with railways. If you work diligently all your life to maintain an old railway, no one will know about it. But if you build a new line, you will definitely put up lights and decorations, and the sound of gongs and drums and firecrackers will be heard. We all like to work on projects and don’t like to indulge in routine work. This is roughly the same. In some companies, some departments have many projects and often make big news, while some departments have very few projects and can only swallow their anger.

2. What is a data project?

The cruel truth is: although the bosses are talking about "big data", "artificial intelligence" and "digital transformation", in most companies, data is more like a supply chain. Although everyone says that this thing is important, the future, and the trend, in the end, one, you can't make money for the company, and two, you are doing dirty work to serve others.

This awkward situation means that data is doomed to have a low status. The companies with a slightly higher status are those that can make money directly from data (such as consulting companies that sell data and data services, third-party service companies, and Internet to B products) or wait to make money from data (hiring a bunch of programmers to prove to VCs that they are artificial intelligence big data companies). If the status is not high, naturally there will be fewer projects allocated to them.

3. What is a Data Analysis Project?

The core manifestation of having few assigned projects is that once the data work is broken down, it all becomes daily work .

Yes, they are not the ones who write the code, and they have no idea how much effort you have to put in to “take that data”. This is not as good as the supply chain. At least when people see the mountains of materials, they feel that they are working hard. People look at data more like the old mother at home: “What are you doing in front of the computer every day?”

Therefore, if you want to start a data-related project, there is only one way left: do it all in one go! Package all the above dirty work under a name that has nothing to do with data but sounds awesome (yes, it’s business intelligence).

The most legal, public and visible of these projects is the BI project with a large data screen. The data departments of countless traditional companies win the favor of their bosses by doing large screen projects.

Of course, this situation has changed in the past two years. The barking of AlphaGo has brought hope to countless bosses for artificial intelligence. When faced with a problem that cannot be solved, people always put their hopes on a power that they cannot understand but that others say is magical. In the past, it was the "Book of Changes" copper coins and gossip, and now it is artificial intelligence algorithms. As a result, many algorithm projects have been launched.

In fact, before the self-media hype, there were already many successful cases of using algorithms to improve business, such as identifying default risks through algorithms, improving outbound call success rates, increasing user click-through rates, predicting electricity consumption/call volume, and so on.

However, these applications have very strict data quality and very clear application scenarios. It is not the case that you can just pick up a few pieces of data and throw them into the model to get results. It is not the case that you can just ignore the management system, capital investment, infrastructure, and business cooperation, and as long as the code is run, money will burst out of the screen. As a result, many projects that rashly waded into the muddy waters of algorithms ended in tragedy.

But it doesn’t matter. Soon, everyone found a new power that they couldn’t understand but others said was magical: data center! So a new round of seeing tall buildings rise and fall began in 2020.

4. What is the crux of the data analysis project?

From the perspective of the project's iron triangle, comparing the data with other projects, the crux is very easy to see:

Even numbers themselves are hard to reflect value. For example, in cause analysis, even without looking at the data, the business can guess a few reasons. If you only limit yourself to the business proposing a hypothesis and verifying it with data, then you are no different from a dog catching a frisbee. Even though you work hard, people still think you are just a handyman.

The above are the fundamental reasons why data analysis projects are rarely launched and difficult to succeed. Of course, what is even more terrible is that many newcomers do not realize this (especially those who have just graduated from graduate school and have written several papers with profound names). They show off their charts, models, and codes like a primary school student showing off his newly bought Transformers in class. They are full of "I am so awesome". They ignore that when it comes to transporting goods, the most ordinary dump truck is better than Optimus Prime.

If you want to break through, you have to hold on to the project triangle:

1. Time

(1) Establish a monitoring system in normal times to free up labor from temporary data collection.

(2) Based on daily data, accumulate experience and seize opportunities!

(3) Trigger a sense of business crisis at critical moments and undertake projects for which you are independently responsible.

2. Cost

(1) Do everything possible to improve data quality.

(2) Promote data infrastructure construction when the time is right.

(3) Consider the least amount of data for each project and use the simplest model to solve the problem.

3. Quality

(1) Infrastructure : Do more work from 0 to 1, fill in the gaps, and demonstrate achievements.

(2) Methodology : Establish reasoning logic, think from a business perspective, and act as a coach rather than a teacher.

(3) Suggestions : Propose 100 hypotheses to blow the minds of the business people and take the initiative in outputting results.

From the above, I think everyone can understand the time and cost. The quality part is easier to understand with specific cases.

Author: Down-to-earth Teacher Chen

Source: WeChat official account: "Down-to-earth Teacher Chen"