Cheng Jing is a CDA member and the author of the best-selling book Xiaobaixue Products. He has 13 years of experience as a product manager of top Internet companies. He has worked as a product manager in Baidu, Meituan, Alibaba and other large factories.

I. Data-driven business analysis
Comparison between data-driven and human-driven
Manpower-driven advertising placement: Take the purchase of cheap traffic as an example. After the business finds cheap traffic, it needs to notify the operation to carry out trial placement. After the operation analysis, it needs to find the financial department to apply for funds. After the placement, it is also necessary to analyze the effect and write a report. The process is cumbersome, dependent on individuals and inefficiency is low, and it is prone to problems such as high communication costs and high decision-making risks.

Data-driven advertising placement: Any company must build a traffic monitoring system, automatically monitor the traffic quotation, automatically carry out small traffic trial placement after finding cheap traffic, select the best channel according to ROI and automatically place it. At the same time, use artificial intelligence to prepare materials, and finally automatically summarize the results, which is more stable and efficient than human-driven, and experience can be inherited.

Key elements to realize data-driven

North Star Index: The premise of data-driven business is that the target can be quantified and driven. The North Star index of different products or companies is different. For example, social platforms pay attention to the number of active users, and Meituan pays more attention to the number of order completions at a certain stage. After determining the North Star index, the data can make decisions based on the index and clarify the key direction of the business.

Business process modelization: Business processes must be summarized, universal and reusable. Take the e-commerce order processing process as an example, each link should be solidified, so that the data can play a decision-making role in the corresponding links. For example, when choosing a courier company, decisions can be made according to user preferences and purchased items. At the same time, the business process is not immutable. It should be optimized according to the actual situation. For example, Pinduoduo simplifies the shopping process. How to judge whether the process is patterned? It depends on whether the newcomer can complete the work according to the document.

II. The role and construction of data platform

The role and construction of the data platform
Reason for construction: Take Tencent Sports as an example, there is a lot of duplicate work in data processing for different businesses such as football members and basketball members. The data platform can process data into semi-finished products, improve data processing efficiency and reduce repetitive labor.
Platform stratification: data platform includes data collection, collection of user behavior data, and integration of third-party data; data cleaning includes processing dirty data and related conversion backup; data processing, that is, visual analysis, and the establishment of decision-making models; for example, deciding whether to buy a platform traffic and number again based on historical advertising data According to the application, it provides decision-making suggestions, abnormal alarms, self-help analysis tools, etc.

Problems and solutions faced by data platforms
Data division of labor: Data attribution is unclear, and there may be conflicts between the business department and the data central station due to data ownership. The solution is that the data is shared, and both the business department and the data central station have the right to obtain the required data to avoid data monopoly.

Resource problems: The data mer may reject the data needs of the business department due to limited resources. The solution is to support co-construction and open the data interface, so that business departments can obtain data by themselves in case of emergency.

Explainability of algorithm results: The prediction results provided by the algorithm team may be difficult to explain, leading to conflicts with the business department. The two sides need to negotiate to determine whether to focus on interpretability or effect. If it focuses on interpretability, the algorithm provides a simple model but is not responsible for the accuracy rate; if it focuses on effect, the algorithm needs to be responsible for the result.

Data security issues: There are data security risks in the data middle, such as employees may obtain and leak sensitive data. The solution includes establishing an approval flow and restricting personnel from accessing data beyond permissions; individuals try not to touch the original data and realizing business needs through the result data; desensitizing some data, such as hiding the middle digits of the mobile phone number.

III. Sharing of actual combat cases

C-end case - Hema users pull new

Target user positioning: target users are determined through white boxization and black boxization. White boxization is interpretable information based on the user's economic conditions, gender, etc. For example, the typical users of Hema are women over 40 years old with good economic conditions and are in charge of buying vegetables, but Shanghai male users also consume more in Hema. Black boxization refers to the use of personality algorithms, which can screen out accurate users, although inexplicable.

Transformation path design: The location of Hema stores is concentrated in the place where potential users are concentrated, encouraging online purchases by issuing coupons, and then using text messages to accurately reach nearby potential users; also increasing the promotion, selecting potential communities to set up stalls according to the data, improving the new effect, and the data can help the experience be copied to the new city.

Data processing method: introduce the potential user model and expand the potential user pool according to user characteristics; judge whether multiple accounts belong to the same user through data association, such as device ID, email address, etc.

Business data analysis is an important test point for the first level of CDA data analysts. If you want to know your data analysis level, you can find simulation questions in the CDA certification applet for testing.

B-end Case - Site Selection of State-owned Enterprise Service Point

Project background and problems: The opening of service points in state-owned enterprises involves multiple departments. The leaders of the planning department, the inspection department and the site selection department hope to open a store. External business negotiations may have objections to rent issues, which leads to difficulties in decision-making.

Data-driven solution: analyze the pain points of each department, provide decision-making data for leaders (such as point flow, number of potential users, etc.), provide data support for business to apply for rent adjustment, solve problems of all parties, win trust, and promote cooperation.
[Summary]

Interpret the essence of data-driven business in detail, compare the disadvantages of manpower-driven, and emphasize the importance of Polaris indicators and business process modeling. In-depth discussion on the construction, role, problems and solutions faced by the data platform, and share practical cases of site selection of C-end box Malaxin and B-end state-owned enterprise service points, so as to help students better use data to promote business innovation and development in work and study.