Spring 2000 Information and Decision Sciences Colloquia
¾ All are welcome! ¾

Friday 4/14/00: Mr. Charles Wang, School of Management, Syracuse University, "The Role of Supply Chain Contracts in Channel Coordination for Perishable Goods"

Thursday 4/20/00: Mr. Simon Foster, Partner, PricewaterhouseCoopers, Cincinnati, Ohio, “Trends in Global ERP Implementations”

Friday 4/21/00: Dr. D. R. Mani, Principal Member of Technical Staff GTE Laboratories, Waltham, MA, "Enterprise Data Mining: Challenges and Explorations in Corporate Knowledge Discovery"

Friday 5/12/00: Professor Fred Easton, School of Management, Syracuse University, “Labor Requirements for Finite, Multi-server Multi-class Queuing Systems”

Friday, 4/14, 3:30-5:00, room 001 in the School of Management: Mr. Charles Wang, School of Management, Syracuse University, “The Role of Supply Chain Contracts in Channel Coordination for Perishable Goods

Summary: This talk discusses four types of supply chain contracts for selling perishable goods in a volatile market: (1) returns policy, (2) quantity flexibility, (3) backup arrangement, and (4) contingent claim/options. We show how the supply chain can be coordinated by the pricing and inventory decision in each of the models. After numerical analysis, we find that under certain conditions, all of the supply chain contracts models above are equivalent. Based on this, we provide a general supply chain contracts model for further research.

Mr. Wang is a doctoral candidate in the School of Management at Syracuse University.

Thursday, 4/20, 5:00-6:20 room 010 (auditorium) in the School of Management: Mr. Simon Foster, Partner, PricewaterhouseCoopers, Cincinnati, Ohio, “Trends in Global ERP Implementations”

Summary: An Enterprise Resource Planning (ERP) system is a database and collection of software modules that support the day-to-day processing and storage of transactions (e.g., financial accounting, purchasing, order processing, material flows, human resources, etc.).  The systems also support analysis and planning.  Some of the larger vendors include SAP, Oracle, and PeopleSoft.  This talk will share global ERP system implementation insights gained from first-hand experience.  Mr. Foster has been a consultant with the London office of PricewaterhouseCoopers for 12 years, joining the Cincinnati office about 1 year ago.  For the last 7 years he has specialized in global SAP implementations.  Prior to joining PWC, he spent 5 years in the international hotel business in the area of systems implementation.

Download overheads (1.9 meg PowerPoint file)

Friday, 4/21, 3:00-4:30, room 001 in the School of Management: Dr. D. R. Mani, Principal Member of Technical Staff GTE Laboratories, Waltham, MA, “Enterprise Data Mining: Challenges and Explorations in Corporate Knowledge Discovery”

Summary: With the advent of the information age, corporations are amassing large volumes of data, capturing extensive information about all aspects of their operation. Given the nature and volume of this data, analyzing and extracting useful and actionable information is a challenging and time consuming task. Powerful and automated data mining techniques are increasingly being used to address knowledge discovery in corporate data warehouses.

Appropriate use of data mining techniques is crucial to the success of any knowledge discovery endeavor. With a wide range of database, on-line analytical processing (OLAP), statistical and machine learning tools available to the analyst, choosing the right tool for the right job is more an art than a science.

In this talk we present a framework characterizing the various data analysis tools--especially statistics and data mining--and their appropriate application to corporate knowledge discovery. Using specific business problems--including churn prediction, customer lifetime value modeling, revenue target setting, and cellular telephone usage analysis--as examples, we also demonstrate how machine learning-based data mining tools can be apt complements of classical statistical methods, and show that their combined usage overcomes many of the shortcomings of each separate set of tools, resulting in intuitively understandable and ultimately actionable models for business problems.

This talk includes joint work with James Drew, Andrew Betz, Piew Datta and Brij Masand.

Dr. D. R. Mani is Principal Member of Technical Staff at GTE Laboratories, Waltham, MA.  GTE Laboratories is the corporate R&D Center for GTE and provides advanced research, development, prototyping and consulting services to various GTE Business Units.

Dr. Mani is involved in the Knowledge Discovery and Data Mining (KDD) initiative charged with research, design and development of algorithms and systems for mining corporate data.  In addition to providing consulting and in-house expertise, the KDD project is applying knowledge discovery and advanced data mining techniques to a range of problems including churn prediction, lifetime value modeling, mobile telephone usage modeling and rate-plan analysis, business division target setting, intrusion detection and web mining.

Prior to joining GTE Laboratories, Dr. Mani was a Senior Research Engineer at Thinking Machines Corporation (now part of Oracle Corporation).  While at Thinking Machines, he designed and developed parallel machine learning algorithms for large-scale data mining and knowledge discovery systems.  Apart from other contributions, he was responsible for conceptualizing and implementing a parallel decision tree algorithm in Darwin(TM), a commercial data mining toolset.

In addition to KDD, Dr. Mani's research interests span high performance and distributed computing, and the application of massively parallel processing to artificial intelligence, knowledge representation, and database systems.  He has a Bachelor's degree in Electronics and Telecommunication Engineering from the University Visvesvaraya College of Engineering, Bangalore, India, a Master's degree in Computer Science from the Indian Institute of Technology, Kanpur and a Ph.D. in Computer Science from the University of Pennsylvania, Philadelphia, PA.

Friday, 5/12, 3:30-5:00, room 020 in the School of Management: Professor Fred Easton, School of Management, Syracuse University, “Labor Requirements for Finite, Multi-server Multi-class Queuing Systems”

Summary: To convert demand forecasts (in units characteristics of the service) to labor requirements, many service managers rely on Erlang models like M/M/C/infinity or M/M/C/N. In fact, Erlang C (M/M/C/infinity) is built into virtually all commercial workforce management systems currently on the market. Its ubiquity is due in part to its robustness, its limited information requirements, and (compared with a customized Monte Carlo queueing simulation, at least) the ease by which Erlang models can be implemented in off-the-shelf software. However, these queueing models assume that inter-arrival times and service times are i.i.d. random variables with means 1/l and 1/m, respectively.  When the service system receives and processes two or more customer classes, however, the i.i.d. assumptions may not hold.

To apply Erlang models to staffing decisions under such conditions, some call center experts suggest that managers aggregate the inter-arrival and service time data over all classes. Other experts suggest that managers apply Erlang C to compute staffing requirements for each type of service independently, then sum the results.  However, both approaches tend to distort the required staffing level for a given service level.  In this paper we present a multi-class multi-server finite queueing model to improve labor staffing and scheduling decisions.  Our goal is to increase the accuracy of estimates for labor requirements for systems that serve two classes of customers, yet retain the speed and ease of use characteristics of Erlang models.  We show that the state probabilities can be computed by an efficient recursion or by numeric methods. We then tested the model.

We found that if service time distributions are even slightly bi-modal, conventional M/M/C/N models misstate occupancy probabilities, systematically understating expected queue times, the probability of service denial, and the minimum resource level needed to provide satisfactory service.  This results in unexpectedly high workloads for dispatchers, more blocked calls, and longer than expected queue times for the callers that do join the system. Furthermore, we find the magnitude of the staffing error induced by the M/M/C/N model increases as the means of the two service time distributions grow farther apart.  For example, by varying the ratio of the average service rates for classes 1 and 2 (m1/m2) from 1 to 9 we found that M/M/C/N understated the number of employees required to provide a target service level by over 100 percent.