2023 CDAS Summit, is being held soon.

Before the summit, we have invited the CDAS summit speaker, Mr. Zeng Jin, a data analytics industry bully, to join our CDA licensee interview. Mr. Zeng can say hello to everyone.

Guest:
Hello everyone, hello Mr. Hailong.

Let me briefly introduce myself to you. In fact, I have quite an origin with our CDA. I am a CDA LEVEL 3 Data Scientist. I was in charge of data analysis and data products in some internet companies such as Tantrum and Go.com.

Then I was mainly responsible for the construction of metrics system, BI system, experimental analysis and user profiling, and so on.

Today, I am very glad to be here to share with you about the data analyst position and data empowerment related content.

Question 1:
As a senior analyst with more than 10 years of experience in data analysis, can you tell us what aspects of data analysis and data science are used to empower business in Internet companies?

Guest:
In my personal work experience, there are two major aspects of data empowerment in Internet companies.

The first major aspect is actually data diagnosis.

The so-called data diagnosis is based on the construction of a set of reasonable indicator system, we go through the data to find problems, diagnose problems, to help find the bad development of the business chapter. This is called data diagnosis.

In fact, I summarize two words in data diagnosis:

The first word is demolition. Big problems are split into small problems, and difficult problems are split into simple problems.

Like the usual principle of the pyramid.

There is also funnel thinking, and then these actually belong to the splitting thinking.

The second word is actually than. It refers to various comparisons.

For example, we can compare with ourselves through the Pareto principle. Find out the 20% of these small points that can really solve 80% of the problem.

Also you can compare with other similar points according to the industry.

So that the data indicator system is built on the basis of a demolition of a ratio. This is the data diagnosis aspect.

Another aspect, in fact, is the data strategy application.

Our enterprise will generally accumulate a lot of data, these data are terabytes, so such a large amount of data if not formed strategy, not formed application that is very unfortunate.

Or we can rely on the data to build a user profile and do some differentiated strategies for users. And then a thousand people with a thousand faces, so that our business can have better results.

Question 2:
Can you give us an example? Let's say the business data diagnosis process, yesterday's daily activity it fell by 30%, today's conversion rate turnover is not more than 1%.

GUEST:
Business diagnosis is a very, very important point in the daily work of data analysts.

Just now I also mentioned that the first step for business diagnosis is actually demolition, the second step is to compare.

We actually use the idea of dismantling as I mentioned earlier to deal with these two problems of Mr. Hailong.

For example, if the DAU dropped by 30% on a certain day, we should first think about how to split this 30% drop into different simple dimensions or problems.

Let's say we can actually use the formula of DAU to break it down.

DAU = DNU + retained users + returning users.

So DAU is composed of these three parts.

After the first level of decomposition, we have to see how many new users, how many retained users and how many returning users contribute to the 30% drop in DAU. For the reasons of these three different users, we can actually further deconstruct them.

If the proportion of new users contributing to the decline is relatively large, we need to look at which new channel is declining the most, then go to do further disassembly.

If the decline in retention data is relatively large, then what is the one thing we need to do at this time?

The first thing we need to do is to see which one of the population retention is declining more, or which mobile platform, iOS or PC, is declining more. Through this way of decomposition, we can approach the truth and the important answer step by step.

In this way, we can understand where the 30% drop in DAU is most specifically attributed to.

This is Mr. Wang's first question, and let's look at the second question, which is the conversion rate.

The conversion rate from page browsing to purchase order is less than 1%, how to solve this?

Still the first step is to split, but here the split we may have to use another tool, is the funnel split.

We can think, from browsing the page to the order, in fact, can be divided into a total of four steps.

The first step is to browse the page, the second step is the product details, the third step is to place an order, the fourth step is to pay.

Probably after these 4 steps, each step it is actually a funnel. The lower the conversion rate of the link of the funnel, the more we need to focus on the link point.

Let's say we have a low conversion rate from browsing to the detail page, the reason is most likely that the recommendation algorithm or recommendation strategy is not effective. The user won't click on the product because it's not pushed to the product that the user likes more.

If the algorithm is fixed and optimized, the conversion rate will be improved.

Another example is if an order has been placed and not yet paid for.

In this case, the first or the user may not be particularly generous, this time may give him a subsidized coupon to assist him to place an order, then the other may be our order interface to the payment interface is not some bugs, there will be some problems. If once fixed, this conversion rate also went up.

Because funnel analysis is a multiplicative relationship between its links, the conversion rate of each link is a multiplicative relationship. So every link you optimize will lead to a very big increase in your conversion rate.

Moderator:
In fact, after listening to Mr. Zeng's introduction, I heard two words, the first word is "demolition", and the second word is "comparison".

That is, you first demolition to find the reason, and then on the basis of this reason you go to the comparison to say which of the score is the largest reduction, on that basis we then go to repair your rules, the indicator system to choose a reasonable or not, directly affect the criteria for this judgment.

Question 3:
Then how do we build a scientific indicator system?

GUESTS:
This question is indeed a very important issue.

In fact, we may think that the index system is a very basic thing, may not be particularly difficult, but in fact it is a very important thing for our data analysis.

If there is no metrics system, all the analysis will be built on the basis of a building in the air.

In a normal enterprise, there is a model to build a metrics system from scratch, which we call the OSM model.

O is the goal. The first step is to find our goal, commonly known as the North Star indicator. All business actions need to be implemented around this North Star indicator.

For example, Taobao, it must be GMV as the North Star indicator. Facebook, for example, is a social network, so its North Star indicator is a user volume.

What kind of metrics are suitable for Polaris metrics?

That must meet two conditions:

The first is to meet the user value. That is to say, this Polaris indicator must be able to reflect the user's love for the product or the business, which is the first aspect.

The second aspect is that it must meet the business value of our company. That is, if the Polaris indicator rises, the business will be able to earn more money.

The second step is strategy. In fact, we need to set up a lot of strategies to achieve the goal of improving the Polaris indicator. This strategy is the key to developing our metrics system.

For example, if my company makes DAU as my North Star indicator, there are only three strategies for me.

The first one is to pull new users, that is, to let a large number of new users come to my product to experience it.

The second strategy is to promote retention, so that users stay.

The third one is to recall, and this time our indexes may be built around the new user DNU and the retention rate of the retained users, as well as the number of returned users.

M in OSM is actually measurement, which means we need to find measurable and easy-to-operate metrics to consider whether the strategy I mentioned earlier is well executed.

This is how we build a metrics system.

In addition, there are some very famous metrics in the field of growth hacking.

For example, the Pirate's Law, which is also the basis we can refer to when building this metrics system.

Question 4:
You also mentioned this aspect of data-enabled business strategy application, that is, can you give a detailed example or explain the content of the strategy application?

Guest:
In fact, there are many applications of data in our Internet enterprises.

First of all, I would like to give an example of user profiling.

In fact, we all know more about user profiling. It can bring a thousand people to the enterprise's strategy, and provide a more personalized experience for users.

For example, we can get in touch with Jitterbug and Racer around us, through the user profile and recommendation strategy is able to let us experience that kind of immersion.

And then there are video sites, for example. By analyzing user profiles and historical data, we can know which celebrity you like more.

The benefit of using portraits is that we can get a larger product revenue for a very small cost.

Another concept that is often mentioned in this growth hack is the magic number.

If a user can complete such a number under the guidance of our operation system, he will be able to witness the beauty of our product and business, and he will stay and continue to use our product.

For example, Facebook, their analysts found that new users added 10 friends within 7 days of registration, so they will stay in their platform for a long time.

With such magic number, the operation or product colleagues can design some punch card activities, and then or design some product functions to guide users to complete the magic number, and eventually become our loyal users.

Another scenario is AB testing.

For example, if a product manager thinks that our copywriter may bring more conversions by designing in this way, then he has no quantitative support, he just comes to such a conclusion based on his own experience and knowledge of the product.

However, if the data analyst helps them to use the AB test, they can tell him that you can increase the conversion rate by 3% with this copy, which can give him the results and basis for iteration in a very intuitive and quantitative way.

Question 5:
In the past few years of working in the field of data, what do you feel most deeply about? What are the achievements and frustrations you have encountered?

Guest:
There are two points that are relatively deep.

The first feeling is that data analysis or data science, this position is actually a lifelong learning position, and it is not possible to achieve overnight.

Whether you work in the first few years, you need to have this learning mindset.

For example, recently came out ChatGPT, and then also very hot. Basically, it can replace the trend of search engine in the future. If you do not learn such things, do not maintain a long-term learning mindset, is very easy to fall behind.

The second point of human and human differences, enterprise and enterprise differences is very huge.

I'll give you an example of the differences between people, in fact, the data analyst position, it is easy to learn difficult to fine positions.

It's very easy to get started, you may be able to become a junior data analyst through Excel or through some very simple tools like SQL.

But you have to really be able to influence the business, to reach a higher level of analysts, in fact, it takes a lot of effort to achieve.

The second is that the difference between enterprises is also very huge.

Some enterprises have terabytes of data in their databases, but no applications. So there is a gap in the process from data to application.

This gap is a thrilling jump, and it requires the cooperation of product managers, operations, data analysts, and business leaders to make this digital leap.

Just now Mr. Wang mentioned what the happiest or most fulfilling thing is, that must be their own data insight and data analysis conclusions for the business use, can play a real role.

In fact, the biggest frustration is that we have analyzed half a day, and finally did not use it, did not play a real business effect.

Therefore, as a data analyst position, his happiness and anger is actually closely related to the business applications.

Moderator:
Recently, Mr. Zeng also shelved a new book "Data Analysis Practice - Methods, Tools and Visualization", what kind of opportunity did you have to inspire the writing of this book?

Guest:
In fact, I wrote this book is closely related to my personal work experience.

As I mentioned earlier, I am actually not a data analyst from a scientific class. I am actually a liberal arts student.

So I came from the position of data analysis, I know what a data analyst needs to do from junior to senior level.

So I hope to write such a book about practice, especially about junior and intermediate data analysts, or non-us data analysis industry, want to understand this piece of content of some readers. This is the original intention of writing this book.

The first aspect is that I hope this book can be read to understand, and there are 260 pictures, after all, pictures or the fastest way to convey information, so the whole book to do the first point.

The second is to open the book can be used, there are more than 50 cases in this book, are the first-line Internet companies
The book contains more than 50 cases, which are the most direct and most in touch with the frontier of such cases.

The main object of coverage is actually three types of people.

The first category of people, just mentioned that junior and intermediate data analysts want to make their own theory and tools really applied to the business of this type of people.

The second category is actually a part of product managers or operations who want to help themselves improve business results through data empowerment in other positions.

The third category of this book is also more suitable for students in the school, the future students want to engage in data analysis work. Because there are many practical and tools and theories that can be used directly in the interview.

Moderator:
In fact, like Mr. Zeng said today, we have talked about this content in this book to everyone like this book can go to Jingdong and Dangdang to search. Thank you all for watching.