About CRM-based information technology channel preference models network management applications _ Papers

Paper Net: Abstract: The Article in the CRM system for customer behavior and preferences of in-depth analysis, based on information technology based on the customer's channel preferences for in-depth discussions. Paper to China's telecom industry, for example, a typical case of selected enterprises, its call center, messaging and web channels, channel preference analysis, the conclusions of the telecom operators to enhance customer relationships, improve business performance, improve customer data processing techniques offer a good guide.

Keywords: CRM, channel preference, telecommunications

A case study method and model description

Western scholars of the origins of the long case study, Kyburz-Graber (2004 case study will be divided into three categories: descriptive case studies, case studies and exploratory case study causal explanation of this study is exploratory case study in the hope that the domestic a telecom operator's CRM practice management process for the study, analysis of customer's preference for the channel in this case for a carrier company, a municipal branches, has more than 30 million users, mainly engaged in mobile communication and data services .


This completed the basic telecom operators in the various operations of production information system, further complete the data collection and building a customer interaction history data covering all the central data warehouse system as an enterprise-class business data platform for their Marketing depArtments fully use this information centralized management platform for various depths of the customer data mining Research for the channel preference analysis of the Research program are as follows:
1 channel type as four categories: Office, call center, messaging, web site.

2 customer channel preference models, the use of data mining, Artificial neural network technology (such as detailed later, calculated using four channels for each user business tendency score.

3 main sources of data including customer service center call records, text messages requesting records, website channels operating records, business support system, work orders, service use data, and the basic properties of the customer data.

4-time analysis window data, the use of channel preference models for each user channel analysis using the propensity score, and through channels of user authentication window for model validation. Analysis window: the historical data used to analyze the characteristics of the time span needed 3 months of historical data; validation window: channels for user authentication, requiring 2 months of historical data.

5 The case study method using the technology of Artificial neural network (ANN. Artificial neural network Research and development originated in the 1940s, to imitate the human brain is a nonlinear mapping system structure and it does not depend on accurate mathematical models, and show adaptive and self-learning function .1943, the French psychologist WSMcCuloch and W. Pitts proposed the first mathematical model of neurons, creating a natural science of human history a new science of ANN Research.

Will continue to test artificial neural network prediction results are consistent with the actual situation. To the actual situation does not meet the input and output data as a new sample, the sample of the new dynamic learning and dynamically change the network structure and parameters, so that the network adapt to the environment or predict structure and parameters of the object itself changes, so that the prediction network model has more flexibility. In the ANN implementation process, often require large amounts of data to generate sufficient training and test samples mode set to effectively train and evaluate the performance of ANN, which is just based on the data warehouse and data mining tools can provide. As the ANN and complementary strengths of both data mining, neural networks for data mining of practical significance and practical value of artificial neural networks in data mining has the advantage of: For the strong capacity of noise data, the high accuracy of data classification , as well as a variety of algorithms can be used to extract rules.

Artificial neural network used in classification, clustering, feature mining, forecasting and pattern recognition, Neural network model can be broadly divided into the following three types: (1 feed-forward networks: the perceptron, back-propagation network model and the functional representatives, mainly used for prediction and pattern recognition fields, (2 feedback network: the Hopfield discrete model and continuous model represented, mainly for associative memory and optimization, (3 self-organizing networks: the adaptive resonance theory: ( Adaptive Resonance Theory, ART model, represented mainly used for cluster analysis.

In this case application, is mainly used for feed-forward network to predict the probability distribution of multi-variable, because this goal is to use several channels to the user the possibility to predict the level.


Second, case studies and analysis of the implementation of

1 business planning considerations. The habit of channel preferences of customers to use the analysis of great significance, can provide strong Marketing support through customer behavior analysis, looking for customers to choose the channel preferences, Marketing channels to provide customers an optimal path . not only help to optimize the channel resources, reduce Marketing costs, can increase the success rate of Marketing, improve customer satisfaction.

Telecom customers can now use mobile communications service channels, including Office, telephone customer service, text messaging, Web sites, self-service terminals, which Office features the most complete service, but the cost is the highest, the most widely used telephone customer service, almost every customer have experience with telephone customer service, most services also provide a channel for the telecommunications company customer satisfaction is very important, therefore, the focus of Research and application is one of how to play the advantages of telephone customer service, with limited resources, more high-value services customers and business, to reduce low-value customers and business occupancy ratio of customer service resources while focusing on the development of electronic channels to focus on improving the penetration of electronic channels to cultivate the habit of users to use electronic channels to guide the user from traditional sources (Office, telephone customer service to electronic channels (SMS and websites) change.


(2) specific technical implementations:
(1 Data preparation: data analysis based on business understanding as well, select the following variables to build a model based variables; (detailed list in Table 1
(2 data quality analysis: the basis of the variables after preprocessing the data quality analysis to remove low-quality variables,
(3 data exploration: by visualization (Visualization tools and statistical analysis and other methods to display and explore the availability of various variables, to obtain the model input variables from which to understand the importance of variables and business rules
(4 data processing: According to the mining tasks, data collected from the central data warehouse, data marts to generate a dedicated tap. Basic data processing are: a summary of combined data source, perform data exploration sampling; through artificial neural network (ANN for model scoring, resulting in the overall model and model validation of technical solutions in two key areas: the modeling process: the classification of preferences for the channel to find a suitable basis for forecasting variables, the model converges faster and better help ; model application process: application of the minimum length principle, control hidden nodes, in order to achieve the best fit. other SAS software tools to implement the model with the scoring.

3. To achieve the specific application case. According to telephone, web, SMS and Office channels every score 10% of users, taking the user's score value of each channel, average revenue per user (ARPU), and the length of the information in the design of network Marketing programs.

(1 of preferences telephone customers through teleMarketing call centers for Marketing, complete messages need to thank customers, also introduced the convenience channel and Web site information-rich features.

(2 pair preferred site customers, SMS alerts the user login through online business for the business offers, the client log when the online business Marketing recommendations, taking into account the development of a higher user base as a response to the Marketing of seeds customer site, the continued preferential incentives.

(3 messages of preferences of customers, Marketing recommendations through SMS, SMS for business to give preferential terms, to remind customers to try a more extensive use of information channels - websites, and provide sites for the simple guidelines.

(4 of preferences Office customers via SMS to remind customers to the latest Office also recommends that customers use the telephone channel, and then guide customers through telephone sales channels outside the hall, and consider incentives to attract these customers to their business use . Links to free download http://www.hi138.com 4 channel model analysis and verification.

(1 call center channel preference analysis in the call center Marketing activities, phone calls outside the preferred target customers with telephone preference of the customer base, followed by no obvious channel preferences of customers, the Office again preferences of customers, customer preferences for Office, you can join in teleMarketing when recommended to the nearest Office user information.

Preferences through the telephone channel customer analysis, summarized the impact of telephone customer service channels, the most obvious preference for the first 10 parameters shown in Table 2.

Among them, call the customer service number, the net length, the total charge number of minutes, whether VIP customer, call the customer the average length of time, make voice sets the number of call forwarding number of the seven factors on the customer's phone preferences have positive effects, that is, customers The larger the value of these parameters, the preferred telephone channels more likely, while the number of sites operating the business type, message number of operations, Web site logins these three factors have a negative impact on the telephone preference, and is affecting the opposite.

To "call the customer service number" for example, T statistic basic significantly (P-value less than 0.05 level of significance, namely, "call the customer service number" on the dependent variable has significant explanatory power, the parameter estimate was 0.102 3, that is, in other controlling for other variables unchanged, compared with the occurrence of several "call the customer service number" increases.

From the telephone channel model validation point of view, the number of the top 10% of users had significantly more, so choose the top 10 percent before the model scores of customers as the preferred target phone from Figure 2, curve, models score top 10% of customers cover the actual proportion of customers with a telephone channel preferences to more than 30%, so the model to enhance the rate of more than 3 times, the instructions to select the top 10% is to meet the current requirements.

(2 messages channel preference analysis of customer preferences through the SMS channel analysis, summarized the most significant impact on preference for SMS channel the first nine parameters: the type of business operation in which the number of SMS, WLAN using the number of minutes, whether to use Chinese secretary, roaming charges minutes to four parameters, have positive impact on the message preferences, when in net long, the number of sites operating the type of business, the total number of billing minutes, call the customer service number, whether to use voice mail messaging preferences on the five parameters negative impact.


As the message the way to conduct business more easily accepted by young people, and in the net length is usually larger customers and old customers, they are more accustomed to using the telephone, the possibility of using SMS is relatively small, so the comparison can be seen in the nets length on the telephone channel is positive impact on the SMS channel is a negative impact.

Model was validated on the SMS channel, almost all of the SMS business model for the user is within 20% of the score, the use of SMS for business user's score is very high, very precise model coverage, models score top 20% of the user several calls for coverage of 100% SMS users, the model to enhance the rate of nearly 5 times shows the preferred model to evaluate the SMS channel users have the ability messaging preferences, with good predictive ability.

(3 Web channel preference analysis of channel preferences of customers through the website analysis, summarize the impact of preferred site of the first 10 channels of the most significant parameters: the number of business types including site operations, data services using the number of species, whether to use Numbers 3 parameters have positive impact on the site preferences, and call the customer service number, the total number of billing minutes, the average length of call customer service, order the number of WAP services, whether VIP customers, the number of SMS operation type of business, buy song ring tones IVR number seven parameters Web preferences negative impact.

Parameters in the top 10, the site preferences affect only three positive parameters, the negative factors affecting the seven, because the parameter settings and select properties now come from customers and use the phone information, the content is usually Site Actions do not have much relevance, many of the data associated with the site is currently difficult to get into the system, another reason may be that Web site Office of the relatively late emergence of time, can provide services less. for the skilled use of site for business Users can provide targeted Marketing to become the company's website business use "seed customers", through their circle of contacts to influence other customers to enhance the site for the number and proportion, to reduce the pressure on the telephone channel, making phone Marketing Center's resources can be put to more productive activities.

Web channel models score top 10 percent of customers through the Web site for the actual verification was significantly higher than on the list of other clients, that the model's accuracy rate is relatively high.


Third, the case study summary

Master channel preference work effectively with limited resources as possible to serve more high-value customers and business, to reduce low-value customers and business customer channel resources occupied by the ratio of the same time, focusing on the development of electronic channels, to develop guide the user from the traditional channels (telephone customer service to business offices and electronic channels (Web sites and message transformation, for telecom operators, you will be able to produce key performance improvement.

Use of data mining data warehouse and then break through the technical difficulties can limit effectively build high-precision models. To build model-based variable selected properly can produce very good adaptability and flexibility in popularity, reflecting covering different provinces, municipalities regional differences from the actual case, but also verify the application of information technology that is superior CRM capabilities, once established, the standard model and technical solutions implementation mechanisms will be easy for operators to facilitate their promotion to create a significant performance.

References:
1. Armstrong, G, Kotler, P. Marketing: An introduction, 2005.

2. Eisenhardt, K. Building Theories from Case Study Research. Academy of Management Review, 1989,14 (4:532-550.

3. Kyburz-Graber, R. Does Case-study Methodology Lack Rigour? The Need for Quality Criteria for Sound Case-study Research, as Illustrated by a Recent Case in Secondary and Higher Education. Environmental Education Research, 2004,10 (1: 53-65.

4. Geertz, C. The Interpretation of Cultures. New York: Basic Book, 2000.

5. Kotler, P, SH Ang, SM Leong and CT Tan. Marketing Management: An Asian Perspective. Singapore: Simon & Schuster, Prentice Hall, 1999.

Links to free download http://www.hi138.com