The text is mainly explained from the data analysis process.
First, clear analysis of objectives: clear goals and focus points, understand that the ultimate goal of analysis is to promote product improvement; give a look at the appearance of a beautiful data report to deal with leadership checks is not what we want, to address user pain points
Second, data collection:
1. Key indicators:
DNU (Day New User) / DAU (Day Live) / WAU/MAU (Weekly/Monthly Live) / UV (Independent Visitor) / PV (Page Views) / ROI (Production Ratio)
AARRR model: Get users / Promote / Promote retention / Gain revenue / Viral spread
Process conversion rate (minutes/hours/days), retention (function/app overall, etc.)
2, the data can come from external, internal, user, etc.
Third, data processing:
This way of processing data is sour and sour (high efficiency), write the sql expression to the desired data, export the excel format, and perform the function processing on the data in excel to make a histogram, funnel chart, pie chart, polyline Diagrams, progress bars, etc., are more intuitively displayed to the reader;
Sql learning website w3c, to know the simple query conditions to write, complex statements can directly ask the group big data colleagues;
Excel's chart display form Baidu, the basis of comparison
2, python, R language
High-level data processing methods, it is recommended to have free learning, which will inevitably improve the efficiency of data processing.
Fourth, data analysis methods
1, segment users
The overall data change trend without subdivision can only feedback a trend, and can not give us a reference. We need to aggregate user scenarios, requirements design logic, business changes to subdivide the data;
As a product manager, staring at PV every day does not allow PV to be targeted. What we need to do is to analyze the various factors affecting PV.
Perfectly solve the conversion rate in one or more segmentation scenarios to improve the overall process conversion rate of the product.
— For example, is the quality of the users entering the headline distribution channel high enough? What is the behavior change of this part of the user in the app? Has this part of the user improved the overall process conversion rate?
— For example, does the user who reads the information every day trade orders higher than the number of users who do not read the information? Do you want to guide users to read the information?
2, comparison data
It doesn't make sense to look at a piece of data in isolation. You need to find a reference to compare it and get a conclusion that is better or worse. The data will only have its special meaning after comparison.
You can compare historical data, competing data, and market data;
Common comparison methods: year-on-year / ring ratio / fixed base ratio
In addition, each piece of data should be specified before the analysis of the data indicators, such as a reduction of N percentage points is considered to have a significant impact, the number of functional retention on the next day has improved the success of M-person calculations; the data should be analyzed in a certain dimension.
Fifth, find the problem
The process of statistical data is exhausted, and by this step you can see the hard work~~
Data analysis will find some outliers, such as the conversion rate is abnormal compared to last week, the transaction data has fallen sharply, etc.
Finding data anomalies is a process. To find the cause through the problem, let us enter the sixth step.
Six, give a solution (analysis reason)
It is difficult to find the cause of the problem only from the data level. We need to find a solution to the in-depth investigation of the problem.
1, combined with the user scene
After all, a problem involves product-related functions. Analysts need to be familiar with the iterative optimization of functions, design logic, functional value, etc. The more they understand, the more comprehensive analysis will not be biased;
Then list the flow of the function, emphasizing that the process is the backbone, and if each function is not functioning smoothly, it must be a problematic function;
Each item of the process is checked item by item, and then it needs to be analyzed in combination with the user scene.
— For example, a button, a prompt, etc. in the second step does not understand the design principle well, and is a product design problem.
— For example, in the login process, the number of failures is 30%, and the reason for the failure is mainly caused by which error code, and then the code is further investigated.
2, combined with business changes
Some companies are business-led, and a decision of the business may cause large fluctuations in product data;
Nothing more than business and investment, understand the recent changes in the business, and the changes in the data will be clear;
Don't use a single vote to make a partial approach, find a few more people, and see if everyone's feedback is the same;
The investment will deceive people, but the data will not be the basic quality of an analyst.
3, combined with competing products
At present, the function of observing mainstream competing products is also very helpful for giving solutions.
Seven, create personalized reports on demand
1. Report structure:
Report overview: data selection time, report purpose, report content, data feedback (insufficiency and revenue), solution
Data Trend: Put the chart made by excel here, so you must use the chart, it is intuitive; the exquisite form is the icing on the cake, the report is based on feedback, and the problem is clear.
2. Report form:
I like to send the report to the WeChat group in the picture, and click it to see it. It can also be made into a word, so that the reader has to open the view and there is a cost, and it needs to be downloaded and opened. This step is irrelevant.
WeChat (neal): 17521187214
Welcome to discuss product related, you can also go to Qudi together~