The six levels of enterprise construction of large data system from the perspective of bat

This paper divides the construction of enterprise large data system into six levels, but it is not a linear process, and there is a basic relationship between each level, but it does not mean that it must be built on a level-by-layer basis. For example, entrepreneurial companies, in the absence of data research and development capabilities, most of the Third-party platform for data reporting and analysis.

The following picture, is the essence of this article, the following one to explore with you.

Data Base Platform

Basic data platform construction work, including data platform construction, data specification, data Warehouse, product data specification, Product ID, user ID, unified SDK, etc.

Many companies do not use the data effectively, that is, the lack of uniform specifications, product data reporting by development in accordance with their own understanding and habit of reporting, there is no standardized SDK and escalation protocol, and the data scattered in the various departments of the product server, can not build a structured data warehouse.

Do the framework of the data platform, many people will understand that the high technology, in fact, the entire data platform to reflect the value of the company needs the cooperation of various departments, such as the establishment of key data index system, needs to be refined from the business indicators of various departments, and recognized by the business sector. Common key indicators are: DAU, PCU, WAU, MAU, per day retention rate (1-30 days), Cumulative retention rate (7th, 14th, 30th cumulative retention rate), new users, effective new users, active conversion rate, pay conversion rate, income indicators, ARPU per capita incomes, channel effect data.

The following figure is the data platform architecture of Tencent and Ali.

Ali large data Business architecture:

Ali Ladder distributed computing platform overall architecture:

II. Data reporting and visualization

In the first level, the Data Index system specification, unified definition, unified dimension distinction, can be very convenient for standardized configurable data report design, intuitive visual output design, including behavior, revenue, performance, quality and many other data categories.

In the PPT to friends, Thunder, Baidu, Tencent and other companies of the data reporting system for detailed explanation.

Tencent Data Portal

Ali Data Map

Iii. Product and Operation analysis

In the establishment of data platform and visualization based on the existing user behavior, income data, etc. to carry out a variety of analysis, the output of daily, weekly, monthly, various special analysis reports. Common data analysis work is as follows:

1. A/b test for product analysis optimization;

2. The use of funnel model for user touch analysis, such as tips, advertising and other exposure to the active transformation;

3. Income effect monitoring and analysis, including pay conversion rate, channel effect data, etc.

4. Business long-term health analysis, such as from the user flow model, product life cycle analysis of product growth and health;

5. Real-time feedback of marketing promotion activities;

User portraits are also a common form of data analysis, including users such as gender, age, behavior, income, hobbies, consumer behavior, Internet behavior, channel preferences, behavioral preferences, life track and location, etc., reflect the user's various characteristics, in order to achieve a comprehensive understanding of users, targeted to provide users with personalized service purposes, Typically, every six months, a user portrait is analyzed.

The following figure is a common approach to data analysis:

Common analysis tools: Excle,spss,sas,enterprise Miner,clementine,statistica. The more personal use is: Excel and SPSS.

The following diagram is commonly used in SPSS data analysis and Mining methods:

Four, fine operation platform

Based on the data on the construction of a fine operation platform, the main platform logic is mostly for user segmentation, commodity and service segmentation, through a variety of recommended algorithms to optimize the combination of goods and services personalized recommendations. There is also a product data operation system for different product lifecycle and user lifecycle.

V. Data products

The broad sense of the data products are very many, such as search classes, weather classes and so on. This is mainly about the narrow sense of the data products, to bat three companies of data products for example to share.

Tencent: A wide range of communication, homing pigeon

Ali: Data Cube, Taobao Intelligence, Taobao Index, in the cloud

Baidu: Baidu forecast, Baidu statistics, Baidu Index, Baidu Sinan, Baidu actuarial

Several slides are intercepted:

VI. Strategic analysis and decision making

Strategic analysis and decision-making, more than a lot of traditional strategic analysis, business analysis level of the methodology similar to the biggest difference is that data from large data.

Many companies mistakenly put "business operations monitoring layer" and "user/Customer Experience optimization layer" to do things in the business analysis or strategic analysis layer to do. Fu Zhihua that "business operations monitoring layer" and "user/Customer Experience Optimization layer" is more through the machine, algorithms and data products to achieve, "strategic analysis", "Business analysis" more people to achieve. Many companies give people what machines can do, which leads to less efficient discovery of problems.

The suggestion is, can use machine to do the thing as far as possible with the machine to do "the business Operation monitoring layer" and "the user/Customer Experience Optimization layer", on this foundation lets the person to do the human more good experience analysis and the strategic judgment.

In the rapidly changing world of the Internet, in the business of strategic direction selection, data is difficult to predict the direction of business development, if some people say that the general direction of micro-letter is through data mining and analysis, the product managers are expected to laugh. In essence, the data in the fine marketing and operation can play a better role, but in product planning, advertising creativity and other creative things, play a small role. But once the product is creative, you can pass the grayscale test, the data validation effect.