Data-driven Operations (v) General analysis methods

In the previous section, we have described the data index one by one that may be used in the operation of the electrical business website, and in this section I will combine the actual work with some simple http://www.aliyun.com/zixun/aggregation/10341.html "> Statistical analysis theory for the data analysis method to do more simple elaboration, I hope to have the experts to participate in sharing.

Phase I: Conventional approach

I think in the grasp of the large operation of the premise of the same, you need to make the following actions on the site data:

1. See Trends

2. Find variation

3. Analysis of reasons

4. Definite countermeasures

(Note: All data analysis is based on the premise that, according to the established business needs, prior to the technical department to provide specific data requirements, otherwise there may be some basic data incomplete and lead to data statistics deficiencies.) ）

See trend: That is, to observe the KPI data of the week, month, season chart curve trend, is up or down; What is the percentage of year-on-year?

Find variation: That is, to find outliers in a single KPI curve, or an anomalous part of an association curve.

Analysis of the reasons: found the exception value, you need to analyze the cause of this anomaly; look at the time node, see the related activities, including internal and external, see the cause of formation, and decomposition reasons, listed, the weight, which is relatively large, which is the possible reasons.

Here, it is often necessary to decompose the data to further split the aggregated data to find the real cause.

The reason for the search is very important, he is the cause of the solution, because found, will have the right decision, resulting in the right action results.

Fixed countermeasures: After the correct analysis of the relevant reasons, we need to give a solution. In general, one reason and one solution can also have multiple solutions, choosing the most realistic and enforceable responses and approaches to action.

And, of course, finally, ethically do it!

Then, after the action, observe again, analyze the chart curve trend and variance value of the key KPI; see if the exception value for the previous period was resolved. If the outliers appear to have good data and have the desired effect, the reason for our previous cycle analysis and the countermeasures to choose are correct, and the next cycle will continue. If the data is still not good, then the reasons for the previous cycle and the choice of countermeasures have problems, but also need to continue to find reasons or replacement of countermeasures and methods of action.

So samsara, cycle by spiral, this is the form of website operation and data analysis.

Example I will not lift, we in the actual work must have a deep experience.

(To be continued)