JD.com shares new understanding and application of enterprise big data

JD.com shares: New understanding and application of enterprise big data

Big data has been very closely linked to each of us’ daily lives.

To give you a random example of a scenario, for example, when I woke up in the morning, I found through the data of my smart watch that the quality of my sleep last night was not very good. I washed my face, brushed my teeth, had breakfast, and walked more than 1,000 meters. I walked to the Liudaokou subway and swiped my card to take the subway. It cost 3 yuan for two stops to reach the Olympic Park. On the subway, I discovered through the JD.com mobile client that a pair of Nike basketball shoes I had browsed before had a price reduction. JD.com proactively pushed this product information to me. , I immediately placed an order and saved more than 100 yuan. I also shared this information with my circle of friends through WeChat.

In this process, I personally produced sleep data, walking distance data, subway card consumption data, subway start and end geographical data, JD shopping data, and WeChat friend circle data, so as a big data producer I So much data was produced all at once. As a big data consumer, when I browse JD.com or the app in the future, the system may recommend to me pillows that improve sleep intelligence, basketball shoes, or other products related to basketball shoes, and friends in my circle of friends see that I After sharing the information, they may also buy because of my sharing.

After enterprises, especially Internet companies, obtain the data we produce, they can cluster, split and predict it through mathematical statistics and mining algorithms to obtain more relevant data. Through these data, Each of us has a labeled description. Such as gender, marital status, hobbies, income, whether you like sports, promotion sensitivity, etc., thus obtaining many attributes of each of us, such as basic demographic attributes, purchasing power, behavioral characteristics, social networks, psychological characteristics, Hobbies, etc.

After companies have mastered this data, how do they use it? Is this data used for marketing, such as precision marketing, precise advertising, and precise product recommendation? Or use these data to refine the company's internal operations and management? Or use these data to improve the production process and guide the secondary research and development of products? That depends on the level of corporate big data practice. If big data is used well, it can truly rise to a strategic level. If it is not used well, big data is just the icing on the cake and a dispensable thing.

According to the clustering thinking of data mining, enterprise data can be divided into internal data and external data, and internal data can be simply divided into financial data and supply chain data (big supply chain concept). Of course, the business operations of enterprises in different industries are very different. For example, in the financial industry, there may be more financial aspects such as investment, financing, and cash management, but very little supply chain data. In the manufacturing or circulation service industry, the data related to the supply chain is limited. There will be more.

Financial data is mainly based on financial statements, especially the three major financial statements, balance sheet, income statement and cash flow statement. Then there is the general ledger. Accounting in the general ledger will involve subjects. If the subjects are not enough, we will also set up auxiliary accounting. Most companies will make budgets every year. Most of the budgets are formulated around financial indicators, or are mainly based on financial budgets. Work backwards into the business budget. Of course, a big part of financial management is money management.

There will be more types of data in the supply chain, from upstream suppliers to downstream consumers, including procurement, warehousing, logistics, production, sales, after-sales and other data. Of course, we can still further refine each link.

In addition, I believe that no company does production or marketing behind closed doors and must actively refer to external data, including national policies, economic environment, stock market conditions, competitors, Prices of main raw materials, etc. The overall architecture of big data

Most companies should have implemented BI systems or report automation systems. If these systems are planned and constructed by Party B, the system solution architecture diagram they develop during the planning or implementation process is nothing more than Divided into three levels and at most four levels.

From bottom to top, the first level metadata layer or data source layer is the data of our business application system, finance, supply chain, human resources, budget, etc.

The second level is called the big data storage layer, which collects the data sources of each level below into a data warehouse. After that, it reaches the third level, the analysis model layer, which builds analysis based on the data warehouse. Model, some solutions even directly omit the analysis model layer, and go directly to the last layer of data display layer to display the data in the analysis model. According to the author's many years of experience in the industry, such an organizational form can at best be called a BI system, but cannot be called a big data system.

JD big data is not a separate system or product. JD big data applications have been integrated into every business application system. Our big data collection platform automatically collects all data to the Hadoop platform regularly and in real time without affecting system or product efficiency and customer experience. With the big data platform as the core, the results will be processed, processed, analyzed and mined. After distribution, it is distributed to various business systems and data products, such as shopping malls, procurement and sales, data compass, pilot, etc. The following figure is for reference only: Enterprise big data application level

Not every company is JD.com, not every company is an Internet company, and not every company's business must require the support of big data. On the premise of meeting their own business needs, can enterprises also play with small data applications? The answer is yes, big data applications can also be divided into levels, and each level meets the needs of enterprises for different levels of data. It is roughly divided into 5 levels, and each level is a progressive relationship.

1. Business monitoring

This is the initial stage of big data application, that is, the traditional DW/BI stage. At this stage, enterprises deploy business intelligence (BI) solutions, which are actually an automated reporting system to monitor the operating status of existing businesses.

Business monitoring, sometimes also called Business Performance Management, refers to the use of basic analysis methods by enterprises to provide early warnings when business operations are lower or higher than expected, and automatically send relevant warnings. Information to appropriate business and management personnel. Enterprise business and management personnel can grasp the business operation situation in advance according to the early warning rules established before, and achieve early warning, helping them to take targeted and foreseeable measures and means to prevent problems before they happen.

There are two most critical points in this stage. One is the design of early warning rules. Frequently used methods include reference methods (comparison of the same period, comparison of similar marketing activities, comparison of peer benchmarks) or indicator methods (brand development, Customer satisfaction, product performance, financial analysis), the indicator analysis method is to choose reasonable indicators. Of course, choosing reasonable indicators here is easy to say, but in fact it takes a lot of thinking to do. Let me give you an example that I encountered before. For example, I was designing a solution for a company engaged in discrete manufacturing. A very important indicator for their performance appraisal in terms of inventory management was the inventory turnover rate or inventory turnover days. This was originally a very normal and frequently used indicator, but this The inventory management of each unit has false shipments and false arrivals. This situation caused the performance indicator of inventory turnover rate to look very good. Later, we considered switching to the moving-to-sales ratio and the inventory-to-sales ratio as indicators to reduce the inventory Indicators and sales indicators are used in combination to avoid false shipments and false arrivals. The purpose of giving this example is to illustrate that when we do business monitoring, indicator selection is very important. We must not only accurately and fairly reflect the business operations, but also avoid artificial fraud.

2. Business insights

Business insights mean that the system does not just provide data reports, but "intelligent" reports or "intelligent" dashboards, which require further prediction and mining based on historical data. Some data that we didn’t know through the previous multidimensional analysis were revealed.

For example, when the author was working on a project for a certain hotel chain in Hangzhou, we needed to make something more interesting based on the operating data of the hotels it has invested in across the country, such as We need to predict the investment in a new hotel based on the decoration investment of previously invested hotels, the current occupancy rate of different grades, the occupancy rate and turnover rate of the hotel's catering department, operating income, costs and expenses, and the situation of competitor hotels in local cities. rate of return and payback period.

In addition, there is DuPont analysis that is often used in financial analysis. Let’s briefly talk about DuPont analysis. DuPont analysis is a model that comprehensively analyzes the financial performance of the entire enterprise from a financial perspective. Its basic principle is that the top is ROE. For ROE, we can decompose it into ROA × equity multiplier, and ROA can be divided into net sales interest rate × asset turnover rate, and then decompose it again, and finally form a tree structure full of financial indicators. Since these financial indicators are calculated through financial statement items, accounting subjects and auxiliary accounting, there is a very urgent logical relationship between them. In this case, we can calculate some technical means to achieve simulation forecasts, such as making next year When budgeting or planning, we will adjust it in advance to what level we want certain financial indicators to reach, and the related indicators will also be linked, such as increasing net profit by 1%, sales revenue, marketing costs, administrative expenses and other To what extent do indicators need to be achieved? This can help us predict in advance and make better planning and budgeting.

Of course, there are many things that can be predicted at this stage. For example, in the retail industry, sales of most categories have sales cycles. Based on the sales cycle, we can predict sales. It can also be based on the relationship between historical users' response to different marketing methods, marketing expenses, marketing products, and marketing effects to more accurately target target groups for targeted marketing, improve marketing efficiency, and reduce marketing costs.

3. Business optimization

Business optimization is still very attractive to most companies, and it is also the goal that many companies think about day and night. In fact, at this stage we can do it step by step, bit by bit. At least enterprises will have the ability to embed analytical technology into business operations. Here is a case we have done for a traditional enterprise before. Like most enterprises, this enterprise also has an ERP system. In the procurement process, we can introduce the supply performance model. Of course, there are factors that may need to be considered in this supplier performance model. There are many factors, such as supply quality, supply efficiency, defective rate, after-sales service and many other factors. When purchasing, purchasing personnel can independently select suitable suppliers based on the supplier performance model. This is an example. In addition The market prices of main raw materials can also be connected to the procurement interface in real time, allowing procurement managers to control the procurement cycle and arrange procurement plans reasonably.

In the retail industry, we all know that there is a strong correlation between products and products, between users and users, and between users and products, just like the example of beer and diapers that everyone often talks about , chocolate and condom examples. Here you can tell me a little bit about how most e-commerce companies do it. We use these products to find the relationship between each two products in the purchased records. This relationship is not equal. For example, when purchasing Users who buy mobile phones generally also buy mobile phone cases at the same time, and people who buy mobile phone cases may not necessarily also buy mobile phones. This shows that there is a relationship between mobile phones and mobile phone cases, and it is a strong relationship. The relationship between the mobile phone case and the mobile phone is a weak relationship. We use coefficients to explain the strength of the relationship here. Therefore, this relationship between commodities and commodities forms a commodity model. Based on this product model, we can better recommend products to users that he has browsed, purchased, collected, and commented on. After talking about products, let’s talk about users. Through similar browsing behavior, search behavior, comment behavior and purchasing behavior, we can find the relationship between users. Based on the behavioral relationship between users, we can recommend to the user some products that other users who are highly related to him have purchased or are interested in. This is a common practice used by many Internet companies to recommend advertising, products, and promotional information.

4. Data profit

Data profit means we often talk about data monetization. One way of data profit is data productization. There are currently many data service companies that can collect mobile games, app usage, user behavior and other data. Through their data mining and analysis technology, the purpose of monetization can be achieved by outputting through the behavior of products or services. In addition, mobile phone manufacturers, such as Xiaomi and Huawei, have hundreds of millions of active users and have first-hand user behavior data on mobile phones, including payment data. There are many aspects that can be realized, and what limits them is their ideas.

In addition, more and more traditional manufacturers are digitizing their products. For example, cars + big data have become Tesla, and home + big data has become smart homes. Of course, there are many examples to be cited here.

5. Business reinvention

Business reinvention should be the highest stage of the big data maturity model. At this stage, some companies hope to use the analysis of customer usage patterns, product performance behaviors and overall market trends to convert business models to new services in new markets, such as JD.com's new business, JD.com Finance, JD.com Intelligence. In addition, we can use our imagination to see which businesses of BAT are developed based on the main business data. Can you think of many?

There are not many companies in China and even the world that truly have big data. We are lucky to have big data across the entire e-commerce value chain. How to tap this gold mine? The only thing that limits us is our own thoughts.

The above is the content shared by the editor about JD.com’s new understanding and application of enterprise big data. For more information, you can follow Global Ivy to share more useful information