|
...business information industries audit business processes and business process management systems to reduce revenue loss and enhance efficient resource allocation. For example, service companies use activity monitoring in a continuous, iterative effort to fine tune operations ranging from claims processing to inventory replenishment. Government agencies now use it to identify and thwart money laundering schemes and bioterrorist attacks. Other examples of activity monitoring include churn detection, credit card or insurance fraud detection, computer intrusion detection, network performance monitoring, and monitoring of news stories (Fawcett and Provost, 1997; DuMouchel and Schonlau, 1998), where thousands or even millions of time series streams have to be monitored simultaneously.
Although activity monitoring has recently begun to receive significant attention from information industries, similar problems and solutions were identified and developed in manufacturing industries, a long time ago, where they are referred to as Statistical Process Control (SPC). SPC methods classify the root causes of process variations as either a common cause or a special cause, and its basic objective is to quickly detect the occurrence of special cause variation (or process shifts) so that the process can be investigated and corrective action taken before the quality deteriorates and defective units are produced. SPC techniques are routinely used for on-line process control and monitoring and they are highly successful in manufacturing applications (Woodall et al., 1997; Montgomery, 2001). Montgomery and Woodall (1997) provide a comprehensive panel discussion on SPC and multivariate methods are reviewed by Hayter and Tsui (1994) and Mason et al. (1997). Recently, Jiang et al. (2003) successfully applied univariate and multivariate control chart techniques to monitor the stability of market segmentations.
Although the principles of SPC can be applied to service industry tasks such as business process monitoring, little research has been done on the application of SPC methods to the monitoring of customer activities so that appropriate marketing campaigns and service customizations can be developed. This paper develops a SPC framework for activity monitoring to allow business planning and forecasting in telecommunications industries. The proposed SPC approach monitors customer profile evolution through a set of state space equations to capture dynamic changes in profiles and detects abnormal events that deviate significantly from a customer's historical profile based on statistical testing principles.
It is shown that the success of the proposed framework critically depends on efficient and robust profiling algorithms to quantify each customer's business trend and volatility. Due to various business and operational reasons, outliers and change points tend to be widely spread in customer databases. To model and track thousands of diversified customer behaviors, it is important to develop simple and unified robust modeling tools, which can accommodate different behavior patterns including business changes, structural breakdowns, as well as unnecessary operational errors. In this paper, in addition to some well-known exponential smoothing methods, we propose a dummy change-point model based on Bayesian Model Averaging (BMA) to estimate customer profiles robustly. Both the simulation and industrial examples from the telecommunications sector show that the dummy change-point model is more resistant to business changes and observational outliers than other smoothing methods when estimating customer profiles.
The aim of this paper is to evaluate SPC-based activity monitoring methods that make use of different profiling algorithms to detect abnormal events such as customer churns and frauds in service industries. The rest of this paper is organized as follows. Section 2 presents an overview of the Business Activity Monitoring (BAM) concept and applications. Section 3 proposes the concept of a three-stage SPC-based activity monitoring framework which consists of customer profiling, monitoring/updating, and event diagnosis. Section 4 introduces a dynamic process model for the implementation of this activity monitoring framework. Section 5 discusses several well-known smoothing and outlier resistant methods for time series profiling. A dummy change-point model based on BMA is proposed to accommodate frequent change points and outliers. Section 6 compares the performance of the profiling algorithms in a simulation example and demonstrates the application of this activity monitoring framework to telecommunications industries. Conclusions are drawn in Section 7 and future research directions are highlighted.
2. BAM
The service sector is currently highly competitive in that it offers numerous competing services and products. For example, in the battlefield between service providers, customer churn is critical for many service companies including telephone service providers, credit card providers, banking, internet service providers, and cable service operators, etc. Customer churn refers to the tendency of customers to cease doing business with a company in a given time period. The annual churn rate in the wireless telecommunications industry is reported to range from 23-46% (Anon, 2000; Anon, 2002) while that in the internet service industry is 21-63% (Anon, 2001; Kolko and Gordon, 2002). Tremendous revenue loss thus occurs due to this customer churn and consequently considerable research effort has been focused on how to predict churn activities before and after their occurrences (see e.g., Lu (2002) and Neslin et al. (2004)).
Many service companies analyze corporate performance by looking at customer activities in a rear view mirror: crunching historical data to link results with causes, develop new promotions and resolve inefficiencies and problems after the fact. To maintain business competitiveness, it is crucial for these companies to develop BAM modules which help track customer profiles and continually refine business models based on feedback that comes directly from knowledge of operational events. By analyzing business activities such as churns in real time, companies can make better decisions, more quickly address problem areas, and reposition the organization to take full advantage of emerging opportunities.
Real-time BAM usually begins with profile modeling which is sometimes difficult due to the complexity of business activities and profile recording processes. For example, in a typical telecommunications company, many customer records of telephone usage exhibit diversified business trends with serial correlations over time and involve frequent business changes brought about by technology innovations, product substitutions, and various financial/accounting adjustments. In addition, the business recording process may often be interrupted by erroneous observations, e.g., flat stretches, bursts of activity, and outliers, which make smoothing and profiling techniques a necessity to aggregate huge transactional data sets, reduce data storage space, and speed up information retrieval from databases.
For illustration, Fig. 1 presents nine representative examples of monthly customer telephone usage patterns from October 2001 to September 2003 in a telecommunications company. Although detailed daily telephone usage is available for each customer, it is always more meaningful to store and analyze aggregated monthly usage as shown in Fig. 1. Moving average and a linear regression fits of the data are also plotted with the observations to act as a reference. The three graphs on the top panel from left to right represent customers whose usage remains approximately constant (or locally constant), constant contaminated by outliers, and constant with behavior changes. Similarly, the second panel represents customers whose usage exhibits trend/local trend with/without outliers and change-point contaminations. The bottom panel shows several more complicated patterns where customers may be new, or undergoing significant changes recently, and/or of high volatility with mixtures of change points and outliers.
The contaminated observations in the database may contain important business information for a company. While outliers may be caused by fraud, recording or billing errors, change points often correspond to marketing campaigns, churn activities, and/or even consolidation of internal customer accounts. If these outliers and change points can be detected promptly, the company will benefit from reduced business errors, improved customer retention, and improved fraud detection. Moreover, detecting these events is also crucial to adapt...
NOTE: All illustrations and photos
have been removed from this article.
More articles from IIE Transactions
A distribution-free tabular CUSUM chart for autocorrelated data., March 01, 2007
Failure event prediction using the Cox proportional hazard model drive..., March 01, 2007
Variable EWMA run-to-run controller for drifted processes.(exponential..., March 01, 2007
Homogeneous redundancy optimization in multi-state series-parallel sys..., March 01, 2007
Element substitution algorithm for general two-terminal network reliab..., March 01, 2007
Looking for additional articles?
Search our database of over 3 million articles.
Looking for more in-depth information on this industry?
Search our complete database of Industry & Market reports by text, subject, publication
name or publication date.
About Goliath
Whether you're looking for sales prospects, competitive information, company
analysis or best practices in managing your organization,
Goliath can help you meet your business needs.
Our extensive business information databases empower business
professionals with both the breadth and depth of credible,
authoritative information they need to support their business
goals. Whether it be strategic planning, sales prospecting,
company research or defining management best practices -
Goliath is your leading source for accurate information.
|
