In a blink of an eye, 2 months have passed since Q4, and there are only 20 days left in 2020. I don’t know how well your KPIs have been completed? In addition to the year-end KPI, everyone is also having a headache for the year-end work report. Most people will accumulate a lot of data on their hands, and they are eager to use these data to show off their work results and let the boss shine.
However, once you open these reports, you will find it difficult! "I don't understand sql!" "I don't know python!" The lack of technical language and tool skills has become a stumbling block for many technical novices to start data analysis.
In fact, technical ability is second. For business data analysis, mastering data analysis methods and clarifying business analysis ideas are the key.
Wheat has sorted out commonly used data analysis methods and tools, so that everyone can easily embark on the road of data analysis. Not much nonsense, let's get into the car with wheat!
First, let's talk about the method
1. Contrast
"No comparison, no analysis". The comparative analysis method is also called the contrast method. It is the most common and most basic analysis method
in data analysis. If we lack comparison in the evaluation and reporting of the data, it will not be able to explain the effect. Still bad.
Example: "How to judge whether an online event meets the standard?" One of the methods is to compare the logged-in user index before and after the event
Commonly used comparison methods include: time comparison, space comparison, and standard comparison. There are three kinds of time comparisons: year-on-year, ring-on-month, and fixed-base ratio.
2. Segmentation
Segmentation analysis is the basis of analysis, and the information value of index data under a single dimension is very low.
When some preliminary conclusions are obtained, further details are needed, because in the process of using some comprehensive indicators, some key data details will be obliterated, and changes in the indicators themselves also need to analyze the reasons for the changes. The subdivision here must be subdivided in multiple dimensions.
Segmentation analysis is a very important method. Asking more why is the key to getting a conclusion, and step-by-step splitting is the process of constantly asking why.
** 3. Classification **
The goal of classification analysis is to divide a group of people (or things) into several categories, or predict the probability that they belong to each category.
For example: "Among JD users, who will place an order in 618?" This is a typical two-category problem: buy or not buy.
Classification analysis (based on historical information) will produce a model to predict which category a new person (or thing) will belong to, or the probability of belonging to a certain category. The result will be in two forms:
Form 1: All users of JD.com are divided into two categories, either will buy or will not buy.
Form 2: Each user has a probability of "will buy" or "will not buy" (obviously these two are equivalent). The greater the probability of "will buy", we believe that the user is more likely to place an order.
4. Clustering
The goal of the clustering task is: given a group of people (or things), without specifying the target, see which people (or things) are closer together.
Pay attention to the essential difference between clustering and the above classification and regression: both classification and regression have a given goal (whether to place an order, whether the loan is in default, house price, etc.), clustering does not have a given goal.
Example: Is it possible to divide the purchase records of a batch of users into several types? (Snacks crazy, electronics enthusiast, beauty expert...)
5. Frequent set discovery
The goal of frequent set discovery is to find people (or things) that often appear together. This is an example of the famous "beer and diaper". This example is too easy to extend, so I won't give an example anymore.
6. Causality
As the name suggests, the goal of causal analysis is to find out the relationship between things that affect each other. Develop two plans to randomly divide users into an experimental group and a control group. Users in the experimental group conduct single-variable interventions on product functions or marketing incentives. The control group runs naturally without any intervention. After a period of time, the data performance of the two groups of users is counted. Evaluate function or incentive effect.
Example: The reason for the increase in advertising effectiveness is that the advertising content is good? Or is it delivered to more accurate users?
The most common method here is A/B test
Second, let's talk about tools
How to choose tools? Of course, on the basis of meeting our analysis needs, the simpler the better, our own job is better to complete the business work, data analysis is only an aid, there is no need to waste too much time in learning tools.
Here, Wheat recommends a tool that is quick to start and powerful in data processing- Smartbi Excel fusion analysis !
It is a self-service data analysis tool that combines Excel and BI organically . It perfectly combines the advantages of Excel and BI. It has 4 key features:
- Operate on Excel to save learning costs. There is no need to export data and then perform analysis to improve efficiency.
- Simply check it to perform data analysis, and complete data processing with the mouse, avoiding the trouble of complex function formulas and no need to memorize function formulas by rote. Not only supports chart analysis, but also supports the use of pivot tables.
- Solve the problem of data sharing. Reports are shared via web links, data is updated in real time, and each user can only see the data within his own authority.