Competition by data analysis

At present, many enterprises in China are facing the biggest challenges of data mining services, such as incomplete data and inaccurate data. Many enterprises have established multiple information systems, including ERP, CRM, and POS.

During the "may" period, which types of sanitary ware need to be sold at a discount? How much discount can the enterprise benefit reach the highest level ?" "Who should we focus on when 3G is about to be activated ?" "Which of the following credit cards can be overdraft and will not be credited ?" These questions cannot be answered by any single information system. At this time, we need to integrate the data of multiple business systems, build a data warehouse, and perform horizontal analysis to achieve accuracy.MarketingAnd precise decision-making.

Zhejiang Telecom and China ConstructionBankIndustrial Bank, Meroni sanitary ware Co., Ltd., and other companies are leveraging the excellent ability to collect and analyze data to find the most valuable customers and accurately predict the product life cycle, so as to achieve accurate marketing and accurate decision-making. How did they do it?

"The essence of Precision Marketing is to group customers, analyze the consumption behaviors and habits of different customer groups, and then design marketing plans based on different consumption behaviors !"

"Scientific fortune-telling"

Last summer, Ningbo Telecom launched the ring tones service. At the beginning, the main Promotion Method of Ningbo Telecom was that the community manager calls each customer through 97 system customer information, and the customer manager calls hundreds of calls every day. Due to the lack of pertinence, the success rate of this "all-out-of-the-box" marketing is only about 7%. Later, after analyzing, classifying, and locating potential ring tones user data, we selected 0.2 million of potential customers who were more likely to purchase 10% for targeted outbound calls. The success rate was as high as 40%. How does Ningbo Telecom Implement Precision Marketing?

This should start with Ningbo Telecom's Mr (re-engineering Marketing) project. In December January 20 last year, the pilot project of China Telecom Marketing re-engineering project Ningbo Mr was officially launched. The core task was to change carpet marketing to Precision Marketing. For a long time, telecom companies have implemented carpet-type marketing, either not offering discounts or giving full discounts to 0.23 million users. How to conduct targeted marketing for 0.23 million public users? Zhejiang TelecomEnterprise InformatizationShen yueming, Deputy Manager of the Department, found that the core obstacle of Precision Marketing is the lack of an accurate customer data warehouse and corresponding customer grouping model. Shen yueming said: "The essence of Precision Marketing is to group customers, analyze the consumption behaviors and consumption habits of different customer groups, and then design marketing plans based on different consumption behaviors !"

As a result, Mao ningfei, deputy director of the marketing center and manager of the marketing department of the Ningbo Telecom Bureau, established a data support team, which is preparing data throughout February. First, they created a Ningbo data view including 243 tables through the data warehouse, then 39 key tables and 337 variables are extracted based on customer characteristics such as user call behavior and data service usage behavior. According to reports, Ningbo Telecom divides customer behavior into two variables: Call behavior and value behavior, and then draws a topology, according to the call behavior data and value behavior data of each customer, the topology graph keeps hitting. At this time, it is found that many customers call behavior or value behavior is very similar, and many points are put together, after the accumulation, a "Hill" is gradually formed, and customers are grouped based on these "hills ,.

Shen yueming and Mao ningfei divide Ningbo Telecom's customers into 11 customer groups, such as the traditional long-distance preference group,EconomicBusiness Group and online surfing group. Then, the customer characteristics and requirements preferences are analyzed based on the Consumption Behavior of each customer group.Data MiningIt was found that some customers in Ningbo have a very high bandwidth usage, are not price sensitive, and the customer turnover rate is also very low. Shen yueming and Mao ningfei called this group an online surfing group. Further data mining finds that the obvious feature of surfing groups on the internet is that, generally around 30 years old, the income is generally high.InternetStrong preferences, strong acceptance of new things, but high requirements on product quality. Shen yueming said: "If 3G business is activated in the future, the online surfing group will be the first customer to be promoted ."

In the process of promoting the ring tones business, Ningbo Telecom uses the Mr tool to conduct Association Analysis on customer data warehouses and "cluster and group" for customers of the same type to form a unified customer view; then, use IDM (intelligenledatamining) and other data mining tools to list multiple smart variables for the customer, such: monthly average fee, intra-region fee, whether the model is specified, whether the SMS function, age, and gender of PHS are available, and then the purchase probability model for various SMART variables is established using the mean algorithm, sort the customer list by probability, and select users with higher purchase possibilitiesTargetThe customer and marketing success rate has been greatly improved. Originally, the marketing staff had to conduct carpet marketing and 20 people's marketing for more than 0.2 million customers.TeamIt takes 200 days, and the success rate is only 7%. After using the prediction model, the marketing staff only needs to select 10% of the most likely to purchase, that is, 20000 of customers for marketing. The marketing team of 20 people only needs 20 days, the success rate is increased to 40%.

In addition, in the process of data mining, Shen yueming also found many valuable customer groups, enabling enterprisesCompetitionThe situation has an accurate understanding. For example, some elderly people only receive phone calls and do not make phone calls after purchasing PHS. For Zhejiang Telecom, monthly rental fees are collected every month. In the traditional view, this group does not make phone calls, and Its ARPU (average income of each household) the value must be very low, which is a restricted business. However, by integrating the data of products such as PHS and landline phones, we found that these old people were using pHs to receive a large number of phone calls from other countries, and mobile phones from Internet networks were used. According to the Protocol, China Telecom can earn 6 cents for every one minute of calls through China Telecom's gateway. If each phone calls 10 minutes and receives 30 calls per month, every PHS can earn 18 yuan a month for Zhejiang Telecom, sometimes at least 18 yuan. For Zhejiang Telecom, these users have a low income but moderate value. Further data mining found that these old people also installed fixed telephone and broadband, and often used Zhejiang Telecom's ipcard for long-distance calls. Shen yueming said: "the ARPU value of these businesses is very high for a single customer ."

Later, in order to promote the strategy of using data analysis to compete, Shen yueming and other data warehouse project teams simplified the data mining process and found more than a dozen key variables from the 337 variables, then some common calculation methods are fixed by using the calculation formula to form a computing method for a specific topic application. The project team calls these key variables "smart variables". For example, the number of complaints to 180 is the smart variable that is most likely to lose customers, because customers will make 180 complaints only when they are unable to endure. Currently, Zhejiang Telecom's data warehouse project and BSS project have summarized 11 package model algorithms. Shen yueming often joked: "These 11 algorithms are scientific fortune-telling ". Now, to design a Business Package for Zhejiang Telecom's render network, you only need to modify the parameters according to the 11 algorithms provided by provincial companies.

Data Mining mainly solves three problems: one is to find the most valuable customers, the most easily lost customers and the most satisfied customers; the other is to analyze their consumption behavior and consumption habits; third, we will launch targeted products and services to achieve a shift from product-centric to customer-centric.

Product Lifecycle

With the deepening of information construction, especially with the completion of data collection in the Financial and telecom industries, China Telecom, banking, retail, and some large manufacturing enterprises can not only rely on data analysis to compete, and they should do the same. Thomas davinport, professor of information technology and management at Babson College, believes that many companies in the industry now provide similar products and use similar technologies. In this era, business processes have become one of the highlights of the last difference that can be created. What analyticdb competitors are good at is to squeeze out all the value from the business process. Therefore, they not only know what products their customers need like other companies, but also know what prices they are willing to pay and how many products each customer will buy in their lifetime, and what will stimulate people to buy more products. Not only do they know whenInventoryInsufficient, and able to predict potential problems in demand and supply chain, so as to maintain a low inventory rate and a high perfect order rate.

Chu shunqi, deputy general manager of the individual financial department of China Construction Bank Qingdao branch, believes that data mining mainly solves three problems: first, find the most valuable customers, the most easily lost customers, and the most satisfied customers, analyze their consumption behaviors and habits, and then launch targeted products and services to achieve a shift from product-centric to customer-centric. The second is to analyze historical data of similar products, determine the life cycle of existing products, and guide enterprises in their attitudes and actions on each product. Third, do a good job in business development prediction. At present, most enterprises often follow the development speed of previous years and multiply the corresponding proportional coefficient to serve as the development goal of the next year. In fact, there is no real basis for doing so, only by analyzing the status of each productStrategyOnly by focusing on the domestic and foreign economic environment can we scientifically develop enterprise development plans.

At present, for the telecom, financial, retail and other industries, in addition to mining customer information to transform from product-centric to customer-centric, we also need to strengthen product life cycle prediction, instruct enterprises to carry out marketing activities at appropriate times and places with appropriate discounts. How can we accurately predict the product lifecycle?

Liu Shiping, president of Jilin Beck information technology company, returned to ChinaEntrepreneurshipBeforeIBMThe company led its data mining team to "dig" from the American Bank of America and Wells Fargo to Canada, and "dig" to Thailand and Singapore in Asia, setting up more than 20 internationally renowned banks.Business IntelligenceManagement System. Liu Shiping said: "Can a product be put into the market to bring benefits to enterprises? What are the benefits? First, we must analyze the historical sales data of similar products and accurately describe the product lifecycle diagram to predict the behavior of new products ."

The lifecycle of a product is dividedR & DPeriod, market acceptance period, maturity period, and recession period. Liu Shiping's practice is to first analyze the sales data of similar historical products in time series, such as how the product is sold in three months, how the product is sold in six months, and when the competitor responds, clarify the "survival status" of similar products in history at each time point, especially the time point of product life cycle transition, and draw a product life cycle diagram. Then, analyze the market reaction within a period of time after the launch of the new product, draw the existing life cycle diagram of the new product, and compare the two product life cycle diagrams to accurately predict the different time points of the new product life cycle turning point, it provides support services for enterprise decision-making.

During the "May Day" period every year, Moroni sanitary ware (China) Co., Ltd. is engaged in a lot of promotional activities, but how much discount the enterprise benefits can be the highest, no one can tell, just follow the feeling. Last yearBi(Business Intelligence), IT andLogisticsManager Peng Yuanhua immediately became "xiangxiao ". Peng Yuanhua said: "Now, before executing a promotion, Meroni analyzes historical sales data of different regions, predicts sales of different products in different regions, and then analyzes the lifecycles of different products, then, combine the consumption habits of different regions to determine the promotion plans for each product in different regions."

Unlike Peng Yuanhua's focus, data mining is more important at the beginning to extend the product lifecycle. With the increasing competition in the market, the product's half-life is getting shorter and shorter, that is, the time from the peak to the recession is getting shorter and shorter. How can we extend the product's lifecycle? At the beginning, compared with the ideal status, the product may have 70% advantages and 30% disadvantages when it was launched, because the new product customers can tolerate these shortcomings. With the product promotion, on the one hand, competitors will gradually launch homogeneous products, and customers will have more options. On the other hand, the customer's own requirements for products are also increasing, and the customer's shortcomings are becoming increasingly unacceptable, this forces enterprises to mine customer feedback during product use and improve product quality in a timely manner. However, the product quality cannot be improved without restrictions. When the product enters the heyday, the Enterprise will continue to improve the product.CostIt may be higher and higher, while the benefits of Product improvement are declining. This means that enterprises need to accelerate the launch of new products instead of continuing to improve old products.

Many enterprises have establishedERP,CRMMultiple information systems, such as pos, can ensure that all transactions or other important transactions are recorded. However, to compete with such information, enterprises must record the information in a standard format, it also needs to be integrated and stored in a data warehouse so that technicians can easily obtain the information.

Source of strength

At present, many enterprises in China are facing the biggest challenges of data mining services, such as incomplete data and inaccurate data. Many enterprises have established multiple information systems, including ERP, CRM, and POs, to ensure that all transactions or other important transactions are recorded. However, they must rely on such information for competition, enterprises must record the information in a standard format, integrate it, and store it in a data warehouse so that technicians can easily obtain the information. In order to implement the Bi system, Peng Yuanhua spent more than a year organizing the data stored in different systems. Peng Yuanhua said: "We should not only integrate the data stored in different systems, but also unify the data format and meaning of different information systems ." Before the implementation of the MR project, Zhejiang Telecom had achieved a big data concentration for the provincial branch network. However, during the data mining process, Shen yueming still encountered a problem of data dispersion.

According to reports, at present, Zhejiang Telecom's data warehouse has 83 sets of information systems of 11 branch networks. Apart from the information systems of the head office, each branch network has six or seven sets of information systems, including the Broadband Network Billing System and PHS billing system. To address the reality of Data dispersion, on the one hand, format conversion is achieved through a unified data model; on the other hand, when combined with topical analysis and other applications, shen yueming and other project team members explored a wide table design data integration method. It turns out that PHS charges a table, broadband charges a table, and fixed-line charges a table, and there is a customer information table in the customer service department. by integrating the customer information on different tables, A wide table is designed, including the customer's name, home address, PHS charges, broadband charges, long talk charges, and municipal calls charges, in the database, all the consumption of a customer in Zhejiang Telecom has become a record.

Shen yueming said: "the ideal situation is that all the products of Zhejiang Telecom used by the family are on your own. The higher the telecom consumption your customer has accumulated, enterprises think that the more you contribute to the enterprise, the more discounts the enterprise provides." Sometimes, a family may use both the fixed phone number of Zhejiang Telecom, PHS, and broadband. But when registering, the family may have one husband, one wife, and one son, because the computer cannot identify the relationship between three customers, statistics can only be collected by three customers. To address this problem, Zhejiang Telecom has specially launched a star family discount package. customers who have installed fixed lines, PHS, and broadband of Zhejiang Telecom will receive two months' broadband fee at the end of the year.

In addition to scattered data, data is inaccurate.CIOThey have to deal with one of the challenges. In the United States, there is a saying: "garbagein, garbageout (whether it is garbage, whether it is garbage )". Liu Shiping said: "data mining is not a waste of money. Because a lot of data accumulated in the Information System is inaccurate, it often takes a lot of time to verify the data during project creation ." When he was working on a project in Singapore, Liu Shiping found that a Singaporean income was 2.8 × 1014. Liu Shiping joked that his income alone could support the whole of Singapore, and his life had a good time. During the project process in China, such obvious data errors that violate the regular rules often occur.

Of course, sufficient data accumulation is required for any data mining project. Otherwise, data mining becomes a "difficult task ". For example, a company may spend several years accumulating data on different marketing methods, and then have enough information to perform a reliable analysis on the effects of an activity. Ddbmatrix is an advertising company under the United States hengxin media group (ddbworldwide,DellIt took the company seven years to hire ddbmatrix to help it build a database that included Dell's presence in print media, radio station, andTV1.5 million records made by various media such as the Internet and cable TV, as well as the sales data of all regions where the advertisements were made before and after the advertisements appeared. This information helps Dell make careful adjustments to each of its media promotions in each region.

Link: complex pre-and post-Work

Since Bi (Business Intelligence) was introduced to China by the end of 1990s, there have been continuous pilot data mining projects or bi software implementation for enterprises, but there are not many successful projects. Why is the data mining service progressing slowly? Why can't I find "gold" when I use the most advanced software in the world "?

Thomas davinport believes that companies such as Amazon and Marriott International use their superior data collection and analysis capabilities to become industry leaders, but they are not just simple digital processing plants, they are both powerful and skillful in applying technology to solve many problems. At the same time, they also focus on looking for the focus that should be focused on, establishing a suitable culture and hiring suitable employees, so as to make better use of the company's continuously obtained data. In the end, it is not only information technology, but also talent and strategy that will bring great strength to these enterprises.

Liu Shiping said that using software to "run" data only accounts for 5% of the total workload of data mining projects ~ 10%. The most difficult part of a data mining project is not data mining, but preparation and interpretation. Before using bi software to "run" data, gisbeck spent a lot of time with the customer to determine the topic of data mining and prepare the data, including data validation; after "running" the data, we also need to explain why such a data operation result occurs based on a variety of non-technical factors, and then provide the customer with policy suggestions based on the data mining results.

In addition to a large amount of data preparation work in the early stage of the project, you must carefully determine the subject of data mining. Liu Shiping said: "Many customers do not know what they want to do or what problems data mining can solve ." Liu Shiping usually spends a lot of time with the customer's management at the beginning of the project.CommunicationAnalyzes the problems that customers urgently need to solve, and then analyzes the information needed to solve these problems, where the information is stored, and so on, finally, the topic of Data Mining is jointly determined based on the customer's business needs and data support capabilities. Shen yueming said: "because the data mining business is still in its infancy, some ridiculous results may appear when you are not careful. Some people who do not understand the data mining business may not trust the project team, cracking Down the enthusiasm of the project team." Therefore, it is best to select a topic with more accurate data resources and easy results as a pilot project for data mining.

Informatization is a "top-notch Project", and data mining businesses require enthusiastic support from management. Thomas davinport believes that if data analysis becomes a means of competition, it will inevitably lead to changes in business processes, behaviors and skills for many employees. Therefore, like any major management change in an enterprise, it should be led by senior managers enthusiastic about quantitative analysis methods. The ideal situation is that the main advocate is the top manager of the company. Former Sarah Libao GroupCEOBarry Bella kept a sign on his desk before he retired, summing up his personal and company philosophy: "We only believe in the words of God, and other people ask for data."