Bi notes-General Bi Process

This figure shows a general Bi process based on personal experience. It is applicable to report and multidimensional analysis solutions, but not to data mining solutions.

Key resources of Bi projects: business experts, Bi developers, and business developers.

The participation of business experts helps to improve the accuracy of requirements. BI developers are mainly responsible for organizing and managing resources related to the Bi process, and business developers cooperate with Bi developers to confirm data and business-related work.

The following is an explanation of the development process:

First, you can easily grasp what users are interested in from reports. Combining with the business system and data structure, you can have a general impression on the topic, A colleague can have a concept of dimensions and measurements that users are more interested in.

However, understanding the business is a process that requires experience and understanding. Different industries have different characteristics. Therefore, the participation of personnel and business experts is important here. In addition, the importance of project-related documents cannot be ignored.

The first four steps must involve bi experience. After reading the report and the system, you can have a general plan for the topic and measurement dimension. If the subject and measurement dimensions do not know what the things are, they cannot be performed here, including subsequent Dimension Modeling.

Gossip: Compared with Bi in China, reports are easy to obtain, because most projects are before Bi, solve the Report on the user's hand first ------ even if you don't want the user to, the report will be inserted to you. Yes, it seems a headache for a lot of reports. What about it, let's see what we can see first.

Model Verification: verifies whether all report requirements can be met based on the created dimension model. Same as above, this step must be performed by persons with Bi experience. If the model does not meet the statistical requirements, re-build the model. Here is an iterative process. The results of each iteration must be precipitated and documented.

Verify the data warehouse structure in reverse order, either manually or automatically by the system. For automatic generation, sqlserver has been supported since 2005. However, it is necessary to manually generate a data warehouse for naming conventions.

Analyze Data sources and SSIS development. It is best to involve developers of relevant modules, because developers have a good understanding of the data structure, SQL skills, and business skills. The purpose of this step is to fill in the data warehouse. SSIS training may be required. However, this step is recognized as the most time-consuming. At the same time, not all statistical items can be explained from the business side. For example, some statistical concepts may never appear in the business system, however, it can be calculated through the basic data combination. Therefore, similar concepts, such as confirmation calculation formulas, require Bi personnel to take up the necessary work to confirm.

At the same time, Bi personnel need to work with business developers to develop incremental data solutions to work with SSIS development. Another good method is that the developer writes SQL and then bi personnel integrates it into the solution using the Bi method. In short, the method is flexible, and the key is to communicate with the developer.

SSAs development, multi-dimensional dataset generation, partition validation, incremental operations, it is recommended that this must comply with SSAs specifications, naming conventions, and so on, this will reduce the subsequent work a lot of trouble.

SSRS and other development. The participants in this step can be determined flexibly, because they need a certain degree of MDX experience and may need to conduct report development training for the team. It should be noted that, even in the current SQL Server version, it is not very comfortable to use Cube as the data source for SSRs development reports, and relevant issues will be elaborated in detail, we also hope that some problems can be solved in the next version of sqlserver.

Data verification is equivalent to the test process. Check whether the statistical data is abnormal. For example, you can use a single SQL query to verify the report data. If any risk occurs, confirm the problem based on the actual situation.

Finally, there is nothing to say about the deployment of the production environment. It is okay to manage the report resources of SSRs. For example, to avoid mutual coverage, we can require report developers not to use shared data sources.

This solution can also be expanded to a multi-dimensional analysis project supported by SSAs. I believe that the main topics of analysis such as reports have been formed, so you can directly use the cube. We believe that multi-dimensional analysis will attract customers' attention. There are many front-end analysis tools. I will not introduce them one by one.

As to whether the data mining experience can be extended on this basis, aspnetx believes that there will be resources that can be inherited, but it is hard to say whether it can meet the needs of data mining, therefore, you have to solve the problem based on your specific needs.

In addition, aspnetx summarizes the four "most" Bi projects ":

The most important part is dimensional modeling. Whether it is accurate or not determines the success or failure of the entire project. experience is also needed here.

The most difficult part: confirm the topic. For a complex business system, this process takes time and requires repeated iterations.

The most tiring part: SSIS development. I am very tired and have to be patient with SQL scripts.

The most important support: the customer's top leaders should remember that if they speak well and encounter problems, they will be able to take the lead, otherwise they will die badly.

This solution may not be the same for different people. The most common is that you should analyze the report first, and then directly analyze the data source for the report content, then, the data source result is used to determine how to create a model. I personally do not recommend this method. This analysis will become complicated and repetitive. Of course, you can summarize the items that need statistics and then analyze them one by one, but do you think it is a better summary to analyze by means of dimensional modeling.

In short, the benevolent sees benevolence, and the wise sees wisdom. You are also invited to provide better implementation suggestions.

Finally, I will also integrate aspnetx into the North float community in April. I would like to take this opportunity to meet more friends in Beijing.