The students who do data are most afraid of being sprayed: "What you do is useless!
" Ten "what you do is useless", at least seven are related to high expectations,
especially those who don’t understand data principles. The higher the expectations
, this is how the tragedy came. How to break? Let’s explain the system today
Problem scenario: A
retail company has started the Internet transformation, has launched a micro mall, and built CDP, and has a little collection of user information on WeChat (limited to user ID + shopping behavior + scattered interactive behaviors). Now the leader in charge of the WeChat Store found a data analyst and said: We have everything, we are just short of an inscrutable analysis. I hope you can make a valuable user profile model to improve your business.
Question 1: After listening to the speeches of the following five leaders, do you think they "do not understand the data" in order of degree...
A: We have everything, we are short of advanced modeling.
B: You want to do artificial intelligence big data Analyze the user portrait model
C: Our daily mall reports are just a few numbers back and forth, and I can’t see what
D: The retention effect of the micro mall is not good, and the user repurchase is low, so we need to analyze the problem
E: For example, buy Users who have used skin care products can see how often she can repurchase
1. What does not understand the data
Thanks to various online articles and online courses, now when you mention data, new people are full of brains: excel, sql, python, tableau, hadoop, spark, sklearn, tensorflow, then since these five leaders do not know a line of code Write, then the answer should be: A=B=C=D=E=0, only I am the master of data...
Or some newcomers think: the model is the most powerful, then the leader of the model must be the most powerful. So it’s B≥A≥D≥C≥E. The leaders have said that they want big data artificial intelligence user portrait models. Seeing that I don’t like him, it turns upside down...
So wrong
Production data is the business of the data department, but application data is the business of each department, and even many senior businesses understand the meaning and usefulness of data better than the cousin who runs the number every day. So from the beginning, you can't use technical capabilities to demand business departments, but you have to see if these people understand how to use data.
The biggest bottleneck of application data is, of course, not modeling, but collecting data. Without good data quality, all analysis and modeling are nonsense. The quality of data is closely related to business processes, construction time, and investment. If the business side is always anxious to launch new functions, do not carefully bury the points, and do not carefully clean up the data, then the data is a mess.
A large pile of shit is also very big, but it is still shit and cannot be turned into rice. This is basic common sense. There is no data, no analysis, this is also common sense.
So don't listen to the business side saying "My data is huge, it's all there." If he doesn't understand what data he has, his analysis ideas are basically bullshit.
Therefore, what can really determine how much data the business side knows is how much he knows about the collected fields, how much he pays attention to the data indicators, and how much he thinks about the problems behind the data indicators.
-
Know the field ≥ Know the indicator
-
Understand the indicator ≥ Understand the data problem
-
Understanding of data issues ≥ Understanding of business issues
-
Understanding the business problem ≥ "It's just a user profile model"
So the true answer to this question is: E≥D≥C≥B≥A
If the leader/customer/colleague you are facing does not know exactly what fields and indicators are in your system, but you really want you to build user portraits and accurately predict models, then you have to be careful! This is no different from expecting you to refine the elixir of life.
For example, in this scenario, although it is a micro mall in name, the only real data collected is user ID + transaction behavior, which is no different from offline store data. A slightly more complicated model cannot be built, and user characteristics are even more impossible. Talk. At this time, instead, leader E mentioned that it is still feasible to make predictions based on certain trading behaviors. In the face of several others, lowering expectations is the right way.
Question one is finished, come to question two
Question 2: After listening to the speeches of the following five leaders, do you think their "expectations are too high" in order of degree...
A: If the data analysis is done well, then the performance will definitely be good.
B: The user portrait model can be accurate Marketing, greatly improving the user purchase rate
C: At least some analysis is necessary, I don’t know, but a very important question
D: User repurchase reasons, you need to analyze it in depth, to figure it out
E: First calculate the repurchase cycle Look
Thinking
test
Yi
Fen
Zhong
Without thinking about it, it can be seen that the expected value of E is the lowest. Then the question is, is the requirement for ABCD four-digit high? Data drives business, and data is productivity. Isn’t it what people are talking about?
Two, what counts as too high expectations
First of all, data-driven business, this sentence itself is not wrong.
But the subject is: the boss uses data to drive his business. As a user, you can never pay for the code written by a certain programmer of a certain company. You will only buy a specific product at a specific time, at a specific location, and at a specific price. A specific need.
Therefore, if data is to be effective, it must be combined with promotional channels, product configuration, price positioning, and promotion methods. These require a systematic operation, rather than a few lines of code by a programmer. Therefore, all behaviors that count on hitting a few lines of code to achieve soaring performance are considered high expectations and unrealistically high.
Secondly, can we analyze the problem without data.
Answer: Yes, absolutely. In fact, data is the slowest way to analyze problems. You have to collect data, clean up, store, and calculate. The Prophet of Chunjiang Plumbing Duck, people in the front line of business can always feel the problem more quickly and more carefully, and find some answers based on experience, insight, testing and other means.
Therefore, it is very unrealistic to rely on data analysis to analyze important issues that are completely unknown to the business. If this is the case, it can only mean that the company's business is all idiots, and the data collected by these idiots is naturally unbelievable.
Once again, how many reasons can the data analyze?
Answer: The subjective reasons are almost impossible to analyze. People's behaviors are not consistent with their inner thoughts, and there are often duplicity or wanting to suppress first. At present, data cannot record people's hearts, and it is very flawed to use behavioral inferences. Secondly, user behaviors are often scattered on various platforms, and a single platform records incomplete data. Therefore, in addition to complaints, returns, participation in group purchases, registration of out of stock, and payment appointments, other behaviors are also difficult to point to a certain inner thought. Therefore, when analyzing the cause, it is very likely that it can only fall into a certain event or a certain behavior, and the real cause is difficult to derive.
So in summary, the ontological answer is A≥B≥C≥D≥E. Students can simply remember the following order of difficulty:
- Directly improve performance ≥ Suggestions to improve performance
- Recommendations to improve performance ≥ Precisely predict trends
- Precisely forecast trends ≥ in-depth analysis of the cause of the problem
- In-depth analysis of the cause of the problem≥Assess the status of the problem
- Assess the status of the problem ≥ understand the status
The more difficult it is for the business side, the more you need to calm down and do it step by step, especially to gain the participation and support of the business department, and integrate the data into the business process to see the effect. Some students will ask: how to integrate data into the business process. If you are interested, follow the WeChat official account of [Ground Qi School]. In our next article, we will use data analysis to improve the sales performance of the WeChat mall as an example. How to do it, stay tuned.
Author: Chen grounded gas, micro-channel public number: down to earth school. A data analyst with ten years of experience has launched a series of data analysis courses and has more than 20,000 students.