National Academies Press: OpenBook

Managing Innovation: Cases from the Services Industries (1988)

Chapter: Operations Research and the Services Industries

« Previous: Using Technology for Competitive Advantage: The ATM Experience at Citicorp
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 115
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 116
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 117
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 118
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 119
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 120
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 121
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 122
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 123
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 124
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 125
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 126
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 127
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 128
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 129
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 130
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 131
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 132
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 133
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 134
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 135
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 136
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 137
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 138
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 139
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 140
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 141
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 142
Suggested Citation:"Operations Research and the Services Industries." National Academy of Engineering. 1988. Managing Innovation: Cases from the Services Industries. Washington, DC: The National Academies Press. doi: 10.17226/765.
×
Page 143

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Operations Research and the Services Industries RICHARD C. LARSON A dispatcher is using a color-graphics computer workstation to design efficient routes for 14 trucks that tomorrow will carry liquid nitrogen to customers in the Northeast. After consulting a computer printout, an operations engineer has just directed crews at 6 dams in a 12-dam hydroelectric system to increase outflows by 10 percent for the next 3 days. The chief executive officer of a major forestry products company is using a video-game-like system to understand better how mill operators in the field can increase profitability by improving their log-cutting . . . activities. A social service volunteer in Atlanta, Georgia is using two card files to assign drivers to vans and vans to routes to deliver "meals on wheels" to elderly and handicapped individuals. A vice president of a large railroad is scrutinizing a consultant's report that recommends an ambitious $1 .5 billion capital investment program over the next 5 years. What; do all these activities have in common? In each case the individual mentioned is a consumer of a product of operations research. Representing a quantitative knowledge-based service industry, operations research has established a foothold in corporate America, both in the goods-producing and non-goods-producing (service) sectors. Despite the growing importance of the field, as demonstrated by numerous documented case studies, relatively little is known about it outside its own "inner circles." As illustrated above, the material product of operations 115

116 RICHARD C. LARSON research can assume numerous forms, making the field ill-defined to out- siders. Operations research can be highly mathematical and is often embedded in less-than-transparent computer software or mathematical models. As a result, reactions of fright, suspicion, distrust, insecurity, and irrelevance are not uncommon. In many ways, operations research is a perfect example of an emerging class of technologies (often provided in the marketplace as a service) that might be characterized as decision-aiding "software technologies." Decision- aiding software technologies range from spread sheet programs to complex optimization algorithms specially tailored to a specific application. The professional fields that provide such software and services include computer science (especially with recent advances in "fourth generation" languages, relational data base systems, and "expert systems"), various branches of engineering and operations research/management science (or, "OR/MS". Despite an impressive array of successful implementations, the market penetration of OR/MS in services industries in the United States is low, raising several questions related to operations research. How does a firm choose to invest in ORJMS and what determines success or failure in im- plementation? Does low market penetration reflect difficulties in evaluating the likely returns to investment? Does it reflect difficulties in translating the analytically rigorous academic discipline of OR/MS into application? This chapter, therefore, has two purposes. First, to provide an overview of successful applications of OR/MS in services industries. Second, by ex- ample and with some admitted speculation the paper addresses the ways in which investment decisions are made about OR/MS applications and the ways in which those investments are evaluated. To get to that point we must first backtrack, spending a little time on a description of the field and a brief review of its history and some of its major accomplishments. OPERATIONS RESEARCH: BACKGROUND Decision-aiding Technology Operations research focuses on developing improved procedures for plan- ning and operating complex systems. To distinguish it from other fields having similar objectives, operations research tends to utilize the scientific method to discover the "laws of physics" of the system under scrutiny. By a process of trial and error not dissimilar from that of a physicist who is both an experimentalist and a theoretician, the operations researcher attempts to develop an accurate mathematical abstraction (i.e., mathematical model) of the system. By manipulating the model, the operations researcher tries to discover improved ways for operating the system. Operations research does not exclude inputs from social scientists and organizational theorists, and a

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES 117 large number of the founding members of the Operations Research Society of America (ORSA) were from these fields (in the year 19521; however, the mathematical model seems to remain a central feature of operations research studies and products. Most operations research models contain decision variables whose values are to be "optimized" subject to certain constraints. If one views a decision as an irrevocable allocation of resources, the values of the decision variables represent a particular allocation of resources. The optimization objective may be to maximize profits or to minimize costs or to maximize customer sat- isfaction, or it may be multidimensional, including two or more such objec- tives. In any event the desired goal of an operations research enterprise is the identification and implementation of improved decisions (i.e., allocation of resources). Operations research is a state-of-the-art technology. It uses the latest sci- entific knowledge from such diverse fields as mathematical programming (i.e., computer-based optimization of mathematical functions subject to often complex constraints), stochastic processes, graph theory, and computer sci- ence. Since its focus is on improved decisions, we may regard operations research as a decision-aiding technology and thus admissible to a discussion of technologies in services industries. Institutionally, operations research is carried out by consultants, in-house technical groups, university professors, and-ever more frequently soft- ware firms. Operations research is itself a services industry. And, as will be argued below, its services are most often sought by other services industries, including transportation, finance, government, health care, education, and consulting. Brief Histor'? Operations research was identified as a field of scientific inquiry and named during World War II. The initial important work, focusing on radar utili- zation, antisubmarine warfare, and other military operations, was done by two groups of diverse scientists, one (in the United Kingdom) under the direction of the physicist P. M. S. Blackett and the other (in the United States) under Philip M. Morse, also a physicist. The group in the United Kingdom christened the new field "operational research" Esee Morse (1977) and McCloskey (1987) for more details]. Among the numerous accomplish- ments of these groups was the creation of "search theory," a new integrated mathematical formalism combining ideas of probability, geometry, and math- ematical optimization and used initially to deploy planes and ships to find enemy submarines. Search theory has subsequently found wide application elsewhere, including design of search strategies to find lost items over vast

118 RICHARD C. LARSON areas; for instance, it was instrumental in helping search parties to locate the wreckage of the Shuttle Challenger crew module. After the war there was considerable interest in developing and applying methods of operations research to problems of the private sector and the nonmilitary public sector. Considerable momentum was given to this effort by the simultaneous developments in computers and algorithmically oriented mathematical optimization, arising initially as "linear programming" with the celebrated "Simplex method" due to George Dantzig. Subsequent ad- vances in algorithmic optimization, many developed at the Rand Corporation, included dynamic programming (Bellman, 1957), various special forms of linear programming (see Dantzig, 1963), and network flows (Ford and Fulk- erson, 1962~; the methodological developments fit nicely with the concurrent technological advances in digital computation. The ORSA was founded in Cleveland in 1952. The first academic programs in operations research were established at Case Institute of Technology and (under the direction of Philip M. Morse) at Massachusetts Institute of Tech- nology. The first Ph.D.s were awarded in the late 1950s. Although the field coalesced as a result of the war effort and subsequent developments, important components of operations research reach back prior to 1940. Thomas A. Edison, serving during World War I as head of the U.S. Naval Consulting Board, used statistical and gaming ideas to develop im- portant early results in antisubmarine warfare (Whitmore, 19531. Queuing theory, which focuses on the development of mathematical models of waiting lines, had its roots in Denmark during the period 1910 to 1915 when the Danish telephone engineer Erlang used probabilistic reasoning to develop the first queuing models to help engineers determine the capacity of telephone switching systems. Graph theory, which has been used extensively by op- erations researchers to model transportation networks, is rooted in the efforts of the Swiss mathematician and physicist Leonhard Euler who in 1736 at- tempted to route a parade over the seven bridges of Konigsberg (now Ka- liningrad) in such a way that each bridge was crossed exactly once; in developing the initial important results of graph theory Euler proved that such a route did not exist (and he showed how to design a minimal length parade route that crossed each bridge at least once) (Larson and Odoni, 1981, p. 3851. With regard to linear programming and the Simplex method, the now famous vertex-to-vertex descent method was at least suggested by the mathematician Fourier in 1826 and additional important early work was done by other mathematicians: Farkas in 1902, von Neumann in 1937, and, es- pecially, Kantorovich in 1939 [see Dantzig (1963, Chapter 2) for details!. As this brief history shows, operations research uses techniques and ap- proaches from many disciplines. A persistent problem for the field has been a labeling one, in which it has often proved difficult to determine how operations research is distinguished from various branches of applied math

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES 119 emetics, physics, or engineering, or (more recently) computer science and artificial intelligence. Prize-winning Works The Lanchester Prize of the ORSA is given annually to the book or paper (in the English language) judged to be the most outstanding contribution to operations research during the year of publication. Since the award's incep- tion in 1954, 39 publications have been honored as Lanchester Prize winners. Using a rough categorization scheme, 15 of these publications have been general in nature and 24 have focused on applications in a particular industry or service category. Of the 24 applications-driven winners, 14 (or 58%) are clearly directed at services industries. These include applications in trans- portation (Leslie Edie, in 1955, was awarded the first Lanchester Prize of ORSA for his 1954 work on traffic management over the bridges and through the tunnels of the Port Authority of New York), banking transactions, uni- versity operations management, urban systems, library management, com- munications, criminal justice, logistics, postal operations, and health care provision. Of the remaining ten applications-focused winners, only two were prompted by problems from the manufacturing floor; the other eight were directed at less easily categorized problem areas: search theory, inventory management, military operations, mining exploration, and purchasing poli- cies (data provided to the author by ORSA). Each year The Institute of Management Sciences (TIMS) selects up to six finalists for the Edelman Award for Management Science Achievement. Each entry is judged on its use of operations research and management science techniques within an implementation context in which real dollar savings or service level improvements or both are reported and verified by external referees. Of the most recent 28 finalists, 19 (or 68%) are clearly in services: transportation and logistics, 8; financial planning and asset management, 3; marketing and sales, 3; urban services, 2; work force planning, 1; postal services, 1; general corporate planning services, 1. Only 3 are motivated directly by concerns in manufacturing. The remaining 6 focus on water systems, inventory, and energy systems. This review of OR/MS prize winners was not meant to imply that operations research is unimportant in the goods-producing sector. Many services such as logistics are often "services in support of manufacturing." In specific manufacturing processes, ORIMS can provide design and analysis tools to assist in the engineering design and operations control of those processes. A Roadmap In the following three sections I have selected cases from three general areas of OR/MS application to illustrate the variety of contextual settings for

120 RICHARD C. LARSON ORIMS, the driving forces behind the decision to invest in OR/MS, the types of products that emerge, some broader organizational impacts of OR/MS, and estimates of the return on investment in OR/MS. The first and third topic areas are directly focused on services industries: logistics and work force planning. The second is "production-related services," selected to indicate how OR;MS provides technical engineering service in the goods-producing sector. Most of the cases are drawn from the open literature, particularly the flagship OR/MS applications journal Interfaces. Several were nominated for the prestigious Edelman Award, thus no claim is made that the sample is "random" in any sense. On occasion the discussion is augmented with information provided to me by the authorts). The space-constrained limited descriptions of these cases do not give a full picture of the institutional and organizational factors that come into play in ORIMS work. Thus I have included, as an appendix, a more detailed case in the area of work force planning. It focuses on the scheduling of emergency telephone (911) operators in New York City. Following the cases I offer a suggested set of conclusions and discuss the problem of estimating the value added by OR/MS, particularly focusing on how managers make the decision to invest in OR/MS. ILLUSTRATIVE CASES IN DISTRIBUTION AND LOGISTICS One of the most successful areas of application of operations research has been in improving operations of spatially dispersed systems. Usually the problems focus around issues of transportation, deployment of vehicles, location of facilities, design of service territories, and inventory management. "Distribution/logistics" is the label we assign to these types of problems. Tactical Planning A recent well-publicized example involves the efficient routing and sched- uling of trucks delivering industrial gases (nitrogen, oxygen, and argon) to spatially dispersed customers. In the production and distribution of industrial gases the major costs are due to electricity (to separate the gases from the air) and to distribution. Typically distribution costs amount to 30 to 40 percent of total direct costs. In day-to-day operations a dispatcher matches customer orders with available trucks and drivers, tries to identify other customers who may benefit from a delivery, and attempts to devise cost-efficient routes for the trucks while not violating any one of myriad constraints (e.g., De- partment of Transportation-imposed constraints on maximum allowable driver time per trip). The potential number of combinations is often enormous, and operating in manual mode the dispatcher must rely on experience and intuition to devise the trip assignments.

OPERATIONS RESEARCH AIID THE SERVICES INDUSTRIES 121 In 1983 Marshall Fisher (University of Pennsylvania) and his colleagues devised an operations research procedure based on mathematical program- ming techniques that allowed the testing and sorting of thousands (perhaps millions) of trip combinations; the new procedure produced solutions typically 10 to 12 percent less costly than those produced manually. In implementation at Air Products, Inc., the reported savings were reduced to approximately 6 percent, still representing substantial dollar volume when projected over the entire corporation. This work was honored by TIMS, which awarded the authors the Edelman Award for excellence in the practice of management science (Bell et al., 19831. The work has been favorably reviewed in the Wall Street Journal, the New York Times, and elsewhere. fAlso see Bodin et al. (1983) for additional references in vehicle routing and scheduling.] The "product" of the Fisher et al. work is a complex computer program that must be executed at least once daily to devise the next day's trip as- signments. The success of the implementation in Air Products, Inc., is due in part to corporate commitment from top management to quantitatively based, computer-implemented tools for improving operating efficiencies. The record of success and failure of similar attempts indicates clearly that a necessary condition for implementation success is the existence of a broadly based constituency within the organization (including top management) sup- porting the effort and able to maintain the delivered product after the oper- ations researchers have left the scene. Not all operations research products in the logistics area are sophisticated computer programs. In 1983 John Bartholdi and Loren Platzman (Georgia Institute of Technology) studied the problem of "meals on wheels." In this application a charitable organization in the inner city delivers meals daily to elderly and infix individuals in their homes. The organization is staffed with a combination of volunteers and near-minimum-wage employees; re- sources are scant, and a state-of-the-art computer for solving complex math- ematical optimization problems is out of the question. Yet distribution costs represent a large fraction of direct costs of operation, and even casual ob- servation of operations revealed that then current methods of distribution were far from optimal. Ingeniously, applying some ideas from the mathe- matical field of "space-filling curves," the authors devised a scheme for assigning drivers to vehicles and vehicles to routes that (1) only required two card files (no computer); (2) produced solutions vastly superior to pre- vious solutions; and (3) naturally included certain operating realities, such as, the fact that the number of drivers showing up for work on a given day is a random quantity. As in the Air Products case, the procedure must be performed daily, but it only requires 15 minutes; before the procedure was introduced, 3 hours had been consumed daily to construct routes. The devised operations research procedures are now used widely throughout the United States for delivering "meals on wheels" and in a variety of commercial

122 RICHARD C. LARSON endeavors as well (package delivery, supplying fresh pastries to restaurants, servicing banks' automated teller machines) (Bartholdi et al., 1983; Bar- tholdi, private communication, 19871. Strategic Planning Whereas routing of industrial gas trucks and vans for meals on wheels required vastly differing computational power, each represents an example of tactical or operational planning in a logistics setting. Many of the successful operations research ma plementations in logistics have been of this type, nary ely involving near-ter - .^ decisions in an operational setting. Yet perhaps a more powerful area for application in logistics is in strategic or long-term planning. Rails Consider the case of the Boston and Maine Railroad. Between 1977 and 1982 the Boston and Maine Railroad made extensive efforts to improve its operating perforce ance, especially in the areas of freight service, terminal control, and freight car utilization. In the arena of long-haul transportation, railroads achieve their competitive advantage by using a single locomotive to pull a great many freight cars. The "operating plan," the most fundamental control at the disposal of the railroad, governs the movements of cars and trains (blocking policy, train schedules, and dispatching policy). Procedures for developing and modifying operating plans are extremely important aspects of the railroad control system. Specific problems arise in establishing con- sistent standards for yard, train, and system performance, due in part to availability of only aggregate information and to the limitations resulting from analyzing each train and yard somewhat independently, despite their clear interdependence. As reported in 1986, Carl D. Martland and his colleagues (MIT) developed the "Service Planning Model" (SPM) that estimates the service and cost impacts of alternative railroad operating plans. Using available data on traffi flows over the network, costs, operating constraints, and parameters describ- ing the proposed plan, SPM estimates yard performance, trip times, aggregate perforce ance per user-defined traffic categories, and numerous types of costs. As a consequence of analyses conducted with SPM, major changes were made in the organizational structure and decision-making processes of the company, as well as in physical facilities and information systems. Savings attributable to this effort amounted to more than $3 million annually, or roughly 3 percent of total operating expenses (Martland et al., 19861. Ac- cording to Martland, "The Service Planning Model in and of itself did not cause the benefits, but provided an impetus to create an effective interde- n~rtm~.ntn1 Planning process" (Martland, private communication November rim r~-^^^^^^^= rip - ~ 19874. Operations research has produced other major strategic planning impacts

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES 123 in the rail industry. As another example we consider the Canadian National (CN) Railway, Canada's largest railway. In 1984 CN Railway's total traffic volume was 174.3 billion gross-ton-miles, which generated C$3.S billion in revenues, resulting in a net income of C$304 million. In the late 1970s traffic volumes through 1990 were projected to double on CN Railway's already congested single track main line. Faced with complex constraints that were physical, financial, and operational, CN Railway embarked on a "Plant Expansion Program" (PEP) whose implementation would handle the in- creased traffic of 1990 while maintaining existing levels of service. Proposals called for capital expenditures during the 1980s of C$2.2 billion, of which C$1.3 billion would provide double track in some congested links of the system. After analyzing state-of-the-art line-capacity methods, CN Railway de- cided to develop detailed simulation models to estimate the capacity of trans- port on segments of the line. An important component of the analysis was the Signal Wake Model that determines for a given configuration of signals the minimum train headway that can be maintained as a given fleet of trains follow each other in the same direction over a specified track layout. Another component was the Route Capacity Model that estimates train delays by simulating operations of trains over a rail line under specified track main- tenance activities. The resulting analyses produced a package of cost-effective improvements for capacity expansion, which included a combination of con- trol technology (closely spaced signals) and strategically located sections of double tracks. The major cost savings of the analysis was the identification of 128 miles of track, originally slated for expansion to double track, that with extra signaling could remain single track until after 1990. This allowed CN Railway to defer C$350 million in capital expenditure beyond 1990 (Welch and Gussow, 19861. Banking Banking is a major (financial) service industry that is not usually associated with logistics. However, as the case of BancOhio demonstrates, logistical concerns can play a key role in efficiency of banking operations. Partially due to the relaxation of branch banking restraints, U.S. banks are increasing their branch networks. The wider geographical dispersion of bank branches can complicate the check-processing function resulting in the need to determine (1) how many operations centers should be used and where should they be located; (2) which branches should be served by each center; and (3) what costs and performance measures should be included in evalu- ating alternatives? From an operations research point of view the flow of checks through a bank can be viewed as a "pipeline inventory model." Items are input at various entry points (banking offices), flow through the processing pipelines and exit in the form of outgoing cash letters dispatched to clearing banks.

124 RICHARD C. LARSON The problem becomes complex due to external time constraints imposed on outputs of the system (clearing deadlines). Transportation is a significant component of the system as checks are moved from receiving branches to encoding sites to capture sites. In January 1984 the BancOhio network consisted of 266 branches repre- senting 42 individual banks. Checks were encoded in 31 of these locations and eventually transported to one of two capture sites (Columbus or Cleve- land) for computer processing and clearing. Management felt that centralizing processing facilities would achieve economies of scale, and initiated an op- erations research analysis to determine the validity of their views. The analysis used a check-processing simulation model (CHECKSIM) to generate efficient transportation routes for the messengers who pick up checks and deliver them to the processing center. The simulation was run for each processing center configuration under consideration. The analysis showed the expected result that consolidation would produce economies of scale, but perhaps even more importantly, that even greater savings could be accrued by moving certain ancillary support functions to consolidation centers. In fact, the savings in transportation and encoding efficiencies ($287,000 per year) were dwarfed by savings associated with transferring certain retail and operations functions ($1,381,000 per year). The total identified savings rep- resented approximately 9 percent of then current operating expenses. In implementation the most difficult problems were associated with reassign- ment (and displacement) of personnel (Davis et al., 19861. According to Davis, the OR/MS implementation effort lasted 270 days and cost BancOhio $80,000 (Davis, private communication, November 19871. Urban Services The use of operations research in logistics is not confined to the private sector. Let me briefly cite two projects that I recently directed involving agencies of New York City. The first was with the New York City Department of Sanitation (DOS). In 1981 DOS was confronted with imminent closings of major in-city land- fills, resulting in a projected doubling of daily refuse tonnage transported by barge to the world's largest landfill on Staten Island (Fresh Kills Landfill). The strategic planning rule then "in good currency" was to "size" the fleet of barges in direct proportion to daily tonnage carried. If tonnage doubles, barge fleet size should also double, according to this tradition-based rule-of- thumb. If the fleet size were to double the city would have had to purchase an additional 40 barges, estimated then at $1 million per barge, representing a potential commitment in capital expense of $40 million. Not willing to trust the "linear rule-of-thumb" for such an important decision, DOS commissioned an operations research study to determine the required fleet size to handle the projected new loads. The study resulted in the creation and implementation of a simulation model Barge Operation

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES 125 System Simulator (BOSS). DOS personnel, while performing numerous pro- duction runs with BOSS, identified a savings through 1990 of at least 10 barges and perhaps 20 barges without impeding service levels. The savings were not unexpected by the operations researchers who saw the barge and tug refuse transportation system as a closed "multiserver queuing system;" such systems almost always display economies of scale, in which workloads may be increased and servers (i.e., barges) need not be increased in direct proportion, while still maintaining similar (acceptable) performance char- acteristics. At the time of contracting for new barges (1983) the ship-building industry was severely depressed, resulting in new barge purchase cost of only $600,000 per barge; hence, as a result of using BOSS for fleet sizing, the city has saved $6 million and may save an additional $6 million (if an additional 10 barges are not ordered in 19901. BOSS cost New York City $100,000, yielding an immediate return on investment of 60:1. (Larson et al.,19881. New York City's Department of Environmental Protection (DEP) com- missioned an operations research analysis in September 1985 to provide a computer-based tool to help DEP planners design a new logistics system to transport sewage sludge to a new ocean dumping site. Sewage sludge is the final product of primary and secondary sewage treatment; it is 97 percent water, 3 percent solid and has a specific gravity of 1. For decades New York City had been dumping its sewage sludge a few miles outside New York's harbor entrance. In 1983 the federal Environmental Protection Agency (EPA) placed New York City under court order to commence a scheduled process whereby eventually all of the city's sludge would be transported to a new EPA-designated site approximately 106 miles south southeast of the harbor entrance. It is at this "106-mile site" that sludge is to be dumped in the future. After requesting bids to transport the sludge from "private haulers," DEP decided that the new sludge transport system should be primarily under DEP's (not a private hauler's) control. The commissioned operations research model was to be able to depict alternative ways of operating the new sludge transport system, including computation of costs and performance characteristics of alternative fleet sizes and fleet mixes, use of transshipment points, impact of dredging and other capital improvements, and increased land-based sludge storage capacity at one or more sites. The model that was ultimately developed, Strategic Logistical Unified Design GEnerator (SLUDGE), accomplished all the desired tasks; it operated on an IBM PC AT desktop computer and required only approximately 2 to 3 seconds for each production run. DEP planners have executed the model in production runs well over 1,000 times in determining the appropriate type and size of vessel to assign to the oceangoing link and to the inner harbor. SLUDGE has also been used to determine the best locations for a primary

126 RICHARD C. LARSON and a backup transshipment site. It is currently being used to "fine tune" the design of the inner harbor system. Although it is difficult to identify a precise "savings" accrued by New York City as a result of using the model, the dollar magnitude of the decisions being made is significant. For instance, the result of the first set of production runs was to design the oceangoing fleet as four 15,000-ton oceangoing barges (towed by special tugs under contract to DEP); the size of the barge con- struction contract was $21 million. Other decisions of almost equivalent dollar consequence are currently being made. The cost to New York City for the model (and accompanying analyses) was $330,000 (Larson, 1988~. CASES IN PRODUCTION-RELATED SERVICES The efficient production of things often requires careful coordination and timing of flows of materials, subassemblies, and the like through a complex sequential (and at times, parallel) process. Due to the logistical nature of many production processes, one is not surprised to learn that operations research has played a significant role in the design and operation of production processes. For years the "production paradigm" has been successfully used to model certain services industries [see, for example, Heskett (1986)]. A well-known Harvard Business School case cites the design and operation of Benihana Restaurants as following a production process, where in effect the customer is the item being processed ("Benihana of Tokyo," HBS Case No. 96730571. ORSA awarded its 1985 award in the practice of operations research to Burger King restaurants who used production processing ideas to develop mathe- matical models of alternative Burger King operations; among the innovations adopted was the decision in many high-volume restaurants to have customers "make their own drinks," a very labor-intensive activity. Production of Hydroelectric Power But operations research has played major roles in more standard aspects of production. Consider, for instance, the production of hydroelectric power by Pacific Gas and Electric Company (PG&E), the world's largest privately held utility. Supplying gas and electricity to more than 10 million people over 94,000 square miles in northern California, PG&E had operating rev- enues in 1984 of $7.S billion. PG&E generates electricity using a mix of hydropower, fossil fuels, nuclear energy, more wind energy than any other utility, solar power, the world's largest geothermal steam engine, solid waste, and biomass. The Sierra Nevada hydrosystem provides more than 20 percent of the electricity the company sells, representing 16 billion kilowatt-hours, the

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES 127 demand of 2.5 million homes. If PG&E had to burn fossil fuel instead of using hydropower from the Sierra Nevada system, an additional 20 million barrels of oil would be required in an average year. The efficient management of the system is necessary to provide energy at low cost. The magnitude of the management problem becomes clear when one considers that the system includes 86 hydropower plants in 23 river basins, covering an extensive region with marked variability in water supply due to seasonality, randomness of weather, and other uses of the water (e.g., drinking and irrigation). Other complications include the cumulative effect of flow in the river basin and the fact that energy produced depends on the pressure of the water flowing through the turbines, producing a nonlinear relationship with total rainfall. Over a 2-year period PG&E developed an operations research software package called HYSS, whose purpose is to compute optimal water release policies for each of the 86 plants in the basin. The objective of the mathe- matical model is to maximize the total megawatt hours of electricity generated over a 1-year planning horizon. HYSS has been used to compute release schedules for PG&E for more than 3 years. According to company estimates, the value of the increased energy attributable to the use of HYSS schedules is $10 to $45 million per year. In addition to improved operations through release scheduling, HYSS has contributed to better resource planning, par- ticularly with regard to scheduling of construction projects. As an illustration, HYSS was used to show that PG&E could generate $13 million worth of additional hydropower if construction of the Kerckoff 2 plant could be com- pleted 2 months early, due mainly to California receiving record quantities of rain and snow that year (LLura et al., 1986) A somewhat different but equally complex energy planning model was adopted by 18 utilities in Brazil to determine optimal allocation of hydro and thermal power-generating resources in the system. The model stochastic dynamic programming (SDP) was extensively validated by the 18 utilities before adoption. Comparisons with a previously adopted model projected savings of $87 million in 5 years, a 28 percent reduction in generating costs. Actual (measured) savings from 1979 to 1984 are reported to be approxi- mately $260 million. The OR model required 540 days to develop and implement, at a cost of $50,000. Like the PG&E case, the model originally developed for near- and mid-term operational planning is now the focal point for longer term generation expansion-planning activities (Terry et al., 1986; Terry, private communication, January 19881. For those further interested in the use of operations research models for hydroelectric power management and, more generally, in water resources management, see Goeller et al. (1985) who report on a 125-person-year effort in the Netherlands for nationwide water resource planning. This path-breaking effort, initiated in 1977, has already resulted in reported savings in excess of $50 million per year.

128 RICHARD C. LARSON Production Scheduling More traditional use of operations research methods in production is in the coordination of a wide range of manufacturing-related activities. Pro- duction planning can be seen as a hierarchy of managerial decision-making activities. The hierarchy ranges from strategic planning through tactical plan- ning to operations control. Hierarchical integration of production planning, scheduling, and inventory control is required to coordinate organizational levels responsible for developing and executing plans. Owens-Corning Fiberglas (OCF) produces in a large facility in Anderson, S.C., a variety of mat products, sold in rolls in various widths and weights, treated with one of three process binders, and perhaps trimmed on one or both edges. There are two parallel production lines with distinct character- istics, producing 200 distinct mat items, with 28 of these products (the "standards") representing more than 80 percent of total demand and the remaining low-volume products called special orders. The scheduling system developed for the mat line addresses the interaction between aggregate planning (relevant costs for work force, overtime, and inventory), lot size determination (production line quantities, line assign- ments, and inventory levels for each products, and ultimate job sequencing (for standard and special order items). The methods, all computer-imple- mented, range from simple "back-of-the-envelope" approximations to math- ematical optimization modules involving thousands of variables. The key optimization module uses derived aggregate monthly inventory levels and individual standard product demands to generate inventory levels, lot sizes, and line assignments for each of the standard items. The resulting scheduling covers a 3- to 12-month planning horizon. OCF used the module to schedule over 20 million pounds of mat production during 1981-1983. As a result the average number of monthly production changeovers decreased from 70 in 1981 to less than 40 in 1982/1983, resulting in an estimated savings of $100,000 or more. Operating efficiencies improved dramatically during the same period of time, although net effects are difficult to quantify (Oliff and Burch, 1985~. The OR/MS development cycle lasted about 120 days, costing OCF approximately $30,000 (Oliff, private communication, December 19871. In 1983 Monsanto opened a second production plant (this one in Pensacola, Florida) to produce maleic anhydride, a chemical used in polyester resins, oil additives, agricultural chemicals, and fumaric acid. The sudden additional production capacity mandated close coordination between and within the plants to minimize costs. Three operations research models were developed, one for each plant and a global model combining both plants. The effort was similar in spirit to that of Owens-Corning, but with one more level of hi- erarchy due to multiple plants. The total time spent on all three models was approximately 1,000 person-hours, with a direct cost not exceeding $50,000.

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES 129 The direct cost savings were estimated to fall between $1 and $3 million per year. Additional cost savings were derived from using the model in the following areas: · the modeling system is used to evaluate many operating policies that would cost thousands of dollars per policy to evaluate in engineering time the system was used to determine whether or not certain compressors should be repaired, the decision yielding a net savings of more than $250,000 the possibility of using vent steam in summer months to run the com pressors was analyzed, resulting in a savings of more than $100,000 (Boykin, 1985) For another example in production planning see Liberatore and Miller (1985) who describe a system implemented at American Olean Tile Com- pany. The system reportedly required less than $100,000 to develop and is saving $400,000 to $750,000 in distribution costs annually, representing approximately 5 percent of the variable costs of production and distribution (Liberatore, private communication, December 19871. It has also resulted in improved coordination and communication between manufacturing and mar- keting and the development of new sales forecasting procedures. Programming Sophisticated Machines Weyerhaeuser is one of the world's largest forest products companies. In 1984 company revenues were more than $5 billion, predominantly through domestic and foreign sales of logs and timber, lumber, plywood, and paper products. Forest products is primarily a commodity industry, which means there is little control over the prices realized from the sale of products. Together with a very competitive environment (in 1984 profits averaged 2.5% of sales for large firms), it is imperative to use raw materials efficiently. Achieving improved use of raw materials meant seeking the best use of each individual tree (i.e., how a tree is cut at the mill). Depending on the decisions that the cutting machine operator makes, the value of the tree can vary by 50 percent or more. Weyerhaeuser cuts approximately 15 million trees an- nually, approximately 100 trees per minute, and there are hundreds of cutting machine operators at dispersed locations. In response to such a challenging problem, M. Lembersky while working at Weyerhaeuser developed over several years a sophisticated "product" called VISION. The first component of VISION is a dynamic programming optimization procedure that determines the best economic use for any tree. The second component is a video-game-like computer system that allows mill personnel as well as company managers to grasp easily the best use of

130 RICHARD C. LARSON any given tree. Operated together, the result is better decisions in practice. VISION allows its user to make his or her own decisions, see what revenues they yield, and compare them with mathematically derived optimal decisions in a very user friendly environment. Implementation of VISION was not without its difficulties, particularly convincing top management to devote sufficient resources to its development and to train and motivate field workers in its use. VISION has been used at Weyerhaeuser since 1977, yielding dramatic results. The value added through additional profits through 1985 is estimated to exceed $100 million. In addition, VISION has had important ancillary impacts up and down the line. At the operational level it has helped workers to adjust quickly to changing operating conditions in the field. At the top corporate level, VISION has been used by George Weyerhaeuser (chief executive officers and others who have committed millions of capital dollars to in-field real-time implementation of VISION and related (subsequent) operations research products in such different areas as facility design and truck routing (Lembersky and Chi, 19861. The mere use of operations research ideas does not guarantee success. The model may be inappropriate, the data too costly to collect and update, or aspects of organizational structure may inhibit proper implementation. One of my favorite examples from the last category involves a Boston-area computer manufacturer. One particularly complex large printed circuit board required more than 10,000 holes to be drilled in it at precise but irregular locations. The time to drill each board using a state-of-the-art programmable electric drill ex- ceeded 1 hour. This total "service time" was roughly 75 percent drill time and 25 percent drill movement time, the latter required to position the drill for the next hole. Anyone who watched the machine operate could see that the drill was routed "all over the place," crisscrossing previous paths, thus significantly increasing total service time per board. Yet when one spoke with technical staff associated with programming of the drill, they insisted that a well-known operations research "heuristic" solution to the "traveling salesman problem" was used and that the machine should be operating near optimally. This heuristic is nothing fancy, being called the "nearest neighbor heuristic," meaning that the drill, when finished with one hole, would be routed next to the nearest undrilled hole location. A drill properly pro- grammed with the nearest neighbor rule would not meander all over the board several times as could be observed in practice. The solution to this puzzle was found when it was discovered that the four different device types to be placed were the responsibilities of four different engineering groups within the firm. Each group, with the help of an in-house optimization group, had programmed the drill for its "own holes," oblivious to the need to coordinate their activities with the other groups. In particular,

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES 131 the in-house optimization group never realized that the same drill bit was being used for all four hole drilling processes. As a result, the drill imple- mented each group's instructions sequentially: the machine would first drill "device type A's" holes, then "device type B's," then "C's," and finally "D's," resulting in the observed meandering (A. Port, private communi- cation, 19871. As a consequence of such a "suboptimal" programming of the machine, one could reasonably expect the drill traveling distance to be approximately 70 percent greater than necessary (Larson and Odoni, 1981, p. 4081. The percentage reduction in drill positioning time was attenuated somewhat by the "fixed" effects of acceleration and deceleration in the positioning process. Still, attainable improvements in the total processing time per board ranged from a few percent to more than 10 percent, not insignificant considering the cost ($150,000) per drilling machine. The failure of operations research here was in fact a failure for like groups to communicate and coordinate their activities at a focal point in the man- ufacturing process where their interests merged. Although this example is small in scale, it illustrates a general condition that when projected over the entire organization implies an urgent need for new forms of internal com- munication in order to stay competitive. At Metelco S. A., a medium-sized manufacturer of printed circuit boards near Athens, Greece, the drilling machines prior to March 1983 were often sequenced in the order of coordinates of holes specified by the company's customers. There, too, drilling of a single board often consumed more than 1 hour. Magirou (1986) reports that he developed a "nearest neighbor heu- ristic" and implemented it on Metelco's programmable drills, thereby in- creasing average throughput by more than 10 percent. Metelco has reportedly saved at least $10,000 per year because of fewer machine operators' hirings than planned. A. J. Nicolitas, managing director of Metelco, reports addi- tional side benefits, especially by saving "money by avoiding common hu- man errors, i.e., multiple Grillings of the same hole," and in making "our management aware of the potential benefits of the interface between man- agement science and electronic hardware" (Magirou, 1986~. This "soft- ware" technology has recently been "hard coded" on a microchip and the chips are being sold throughout Europe as device controllers for programm- able drilling machines (A. R. Odoni, private communication, 1988~. This example illustrates as well as any the interchangeability between software and hardware, and between services and goods. ILLUSTRATIVE CASES IN WORK FORCE PLANNING One of the most important areas of application of operations research to the services industries has been in deploying and scheduling of personnel. Various work force planning "packages" have been developed in such widely

132 RICHARD C. LARSON divergent services as retail sales, air transport, urban services, telemarketing and telephone sales, clerical services, restaurants, and banking. Like logistics and production-related services, work force planning relates to the scheduling (and sometimes movement or placement) of "allocable resources," in this case people. The potential returns in this area are often sizable due to the labor intensiveness of many services; for instance, work force scheduling procedures have been developed for services whose costs are 95 percent attributable to salaries and related fringe benefits. Scheduling In 1984 United Airlines recorded earnings of $259 million on a revenue base of $6.2 billion. The profitability resulted in part from an ambitious expansion plan implemented in the previous year which brought about a 6 percent growth in revenue with only a 2 percent growth in costs. A major factor in cost containment is the airline's newly created computer- based work force planning system for scheduling shift work at its reservation offices and airports. United's 11 reservation offices employ more than 4,000 reservation sales representatives (RSRs) and support personnel, with require- ments for work force determined by a forecasting of call volumes based on historical trends and a queuing model to determine (at a given demand level) the number of employees to provide the desired level of service. Also covered in the developed Station Manpower Planning System (SMPS) are the 1,000 customer service agents (CSAs) at its 10 largest airports, the CSAs divided between counter and gate employees. Although work force requirements vary widely by time of day and day of week, work rules require employees to have the same starting time every day and to work the same shift length every day. SMPS uses developed requirements for 30-minute intervals over a 7-day period to produce monthly shift schedules. The system uses state- of-the-art mathematical programming techniques and encompasses the entire scheduling process from forecasting of requirements to printing emolovee schedules. ~, Because of a company-perceived urgent need for an effective scheduling tool, not enough time was initially devoted to involving employees in de- veloping the new procedures. Although economically "optimal" in some (infeasible) sense, the model's exclusion of factors important to employees delayed implementation until 1983. However, since 1983 SMPS has been used to develop work schedules for 4,000 employees on a regular basis and is eventually expected to schedule 10,000 employees or 20 percent of United's total work force. The system has produced savings of more than $6 million annually while earning strongly positive reviews from United's upper man- agement, operating managers, and affected employees. Hard-to-quantify cap- ital benefits include the following:

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES · additional revenue generated by improved service levels · benefits from the use of SMPS in contract negotiations 133 · savings from reduced support staff requirements · savings from reduced manual scheduling efforts · reduced training requirements (lIolloran and Byrn, 19861. According to Holloran, the development cost of SMPS was approximately $500,000, allocated over a 260-day development period (Holloran, private communication, November 1987~. Services associated with "income tax" time are highly seasonal, requiring careful scheduling of personnel during short peak work load periods. For the Financial Services Group (FSG) of Canada Systems Group, Inc., the "season for giving" to one's tax-deductible retirement fund apparently lasts approx- imately 6 weeks (late January to early March). In 1984 the incremental cost to FSG for managing this intense period was approximately $500,000. Prior to the following season, the FSG developed a linear programming work force planning tool that, based on projected work loads, developed hiring needs and shift assignments for the 6-week period (well in advance of that period). The incremental cost of managing the 1985 season, with the new tool, was reduced to $170,000, a 64 percent reduction, despite somewhat higher wages and a 25 percent increase in volume. As has been shown to be common with other operations research installations, other intangible benefits were also reported, particularly an enhanced reputation for reliability of service that has resulted in successful acquisition of new clients and all but one major client renewing contracts (Haehling von Lanzenauer et al., 19871. Spatial Deployment Work force planning may also relate to the allocation of personnel over service territories. In fact the subfields of "optimal location" and "optimal districting" are two of the most active fields in operations research. In the 1970s the literature of both operations research and marketing began to offer detailed consideration to the use of mathematical programming mod- els to assist in sales territory design decisions. Models were developed to allocate work load among a fixed number of salespersons, to calculate the best number of salespersons, and to determine territory boundaries. Later refinements dealt with constraints on time limitations of the salespersons, supervision, salesperson experience and competition, as well as adjustments to boundaries taking into account natural obstacles. The Houston-based Variable Annuity Life Insurance Company (VALIC) markets annuity contracts to not-for-profit organizations and governments. In 1982 VALIC decided that it needed quantitative guidance in the design of its service territories and in the structure of its field offices. At that time there were 336 salespersons nationwide, allocated over 16 regions, each with

134 RICHARD C. LARSON a manager and an office. Among the management issues were the number and design of sales territories and regions, while considering equity of "mar- ket potential" in each and morale problems associated with redesigns. An operations research analysis commenced whose purposes were threefold: (1) to determine the cost associated with the current 16-region configuration; (2) to determine the lowest cost solution in both number of regions and their geographic configuration; and (3) to estimate expected cost savings if the change in configuration were to be adopted. The analysis presented an in- teresting trade-off between fixed and variable costs, the fixed costs associated with regional offices and the variable costs associated with intraregion travel times. The first use of the resulting program was focused on the then present regional configuration, showing a model-derived cost of $1S,826,000. Also, it was determined that by closing one regional office and moving a few regional boundaries, VALIC could reduce total costs by 4 percent. More interesting was the cost difference when the number of regions was allowed to vary. The total cost of the solution resulting from 25 regions was $9,933,000, a savings of $8,833,000. Not surprisingly there were obstacles along the way, such as initial results violating constraints on disproportional market potential among regions and the apparent uncaring attitude of the company toward changing the locations of current regional offices. As of the time the case was reported (1984), VALIC had launched a 5-year phase in of the resulting "fine-tuned" recommendations. Management appeared confident in the projected cost savings but had decided to "go slow" in the sensitive area of personnel relations (Gelb and Khumawala, 19841. The deployment of ambulances throughout a city represents a totally dif- ferent type of spatial deployment problem. The reader is referred to Brandeau and Larson (1985) and Eaton et al. (19851. In redeploying ambulances in Austin, Texas, Eaton reports that his $30,000 study saved the city $10.8 million over the following 7 years; ambulance first response times decreased 7 percent in the face of a 52 percent increase in demand. This example, coupled with the earlier "urban services" cases, demonstrates that significant returns to investment in operations research are available in the public as well as private sectors. CONCLUSIONS What have we learned from our tour of OR/MS applications in services industries and in production-related services? I would like to offer the fol- lowing: . An OR/MS product can assume many forms, from a computer program implemented in color graphics to a consultant's report, to card files, to

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES . 135 a "smart" machine tool, to an educational "video game." The nu- merous embodiments contribute to the field's fuzzy image. In implementation, often unanticipated "side benefits" of an OR/MS effort will dominate the benefits accrued in the original target area of the work. Such side benefits can be limited to additional unanticipated cost savings (or service enhancements), or they can extend more fun- damentally into managerial structure and flow of information. In the latter case the potential magnitude of the effect of OR/MS is often accompanied by a comparably large organizational resistance to change. · The costs of implementing OR/MS can vary enormously, from $50 per implementation for "meals on wheels" to millions of dollars for large- scale, multiple-site decision support systems. · The reported benefits of OR/MS work are often one or two orders of magnitude greater than the costs. · In-the-field knowledge of even rudimentary properties of operations research models (and thus of operating systems) is often lacking. · Managers do not like to state explicitly target service levels that im- plicitly admit to failure a certain fraction of the time (i.e., "probabil- istically stated objectives". · Markedly successful ORIMS efforts seemed to be accompanied by (1) top level corporate enthusiasm and long-range commitment and (2) involvement of operations personnel during implementation. · Due in part to the amorphous nature of its products and the highly technical nature of its process, as a profession OR/MS runs the risk of being absorbed by related and more easily identifiable fields such as computer science. Although there are numerous OR/MS "success stories, " several of them reported here, the overall market penetration of OR/MS in services remains shallow. The field's limited impact to date may be due to excessive academicism in the field, fear of technical approaches by operating managers, need until recently to use mainframe computers, and exclusion by many operations researchers of broader nonmathe- matical aspects of the problem. · Operations research offers the potential for great productivity improve- ment in languishing services sectors, improvements often greater in percentage terms than those typically associated with the manufacturing . sector. VALUE ADDED FROM OPERATIONS RESEARCH Is it possible to estimate a priori the "value" of any proposed OR/MS effort? Investment in capital equipment is a familiar activity of U.S. cor- porations. By now standard techniques exist to estimate costs and benefits

136 RICHARD C. LARSON of many "hardware" investment alternatives, including discounting cost/ benefit streams into the future, estimating time until the investment is re- couped, and computing total (discounted) returns on investment. But con- siderably less is known about investing in various kinds of "software" technologies. Occasionally, as in "desktop publishing," the savings and productivity improvements are so demonstrable that the investment is clearly a good one. More problematic are investments in software and services that are aimed to improve "decision making," either at high corporate levels or at operational levels. Examples would include software/services for invest- ment planning, assembly line balancing, airline scheduling, allocation of marketing dollars, analyzing potential new markets, deploying work forces, analyzing customer service satisfaction levels, and designing a new logistics system. Many of these types of decisions are based on intuition and methods derived from "years of experience." When I contacted them in relation to this paper, OR/MS practitioners and researchers were doubtful that any formal mechanism could be devised. According to Y. Sheffi fa well-known logistics specialist and coauthor of a case reported for Marshalls, Inc. (Carlisle et al., 197, Estimation of cost/benefit: the burden is on me and a project champion in the or- ganization.... Mostly, no formal analysis is undertaken as decision maker in the organization gets finally convinced by hand waving. No specific time is used to recoup costs. The potential has to be enormous-otherwise the project is not done. In other words, the benefit/cost ratio has to be (an implied value of) 20-200 for people to feel comfortable (Y. Sheffi, private communication, October 19871. The emphasis on large benefit/cost ratios was repeated by others. Ac- cording to Amedeo Odoni (a recognized expert in OR/MS as applied to airport planning), In my experience I have not really encountered any formal mechanisms for evaluating the costs and benefits of an OR study. The reason may be that the benefits are usually of a different order of magnitude than the costs (e.g., in a $100K study of an airport's layout, one may "save" $25+ million, a real example). Airport benefits are also often difficult to quantify in dollar terms (Amedeo Odoni, private communication, November 1987~. According to John Bartholdi (coauthor of the "meals-on-wheels" project), In my consulting experience costs/benefits must be clear and large before a client undertakes action. tHe or she] expects immediate payback (or at least within 1 year) and wants insignificant risk. Future is not discounted, since action not taken unless improvement will be enormous (John Bartholdi, private communication, November 1987~. There appear to be some settings in which "scientific" a posterior) eval- uation of an OR/MS product is possible. For instance, in logistics, if the concern is solely transportation cost reduction, one can analyze the decisions

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES 137 of truck dispatchers with and without the OR/MS technology to estimate savings. This was apparently the approach taken in the celebrated Air Prod- ucts case. According to the principal author of the work, Marshall Fisher, with regard to evaluating the benefits, Air Products was quite thorough and methodical in this regard and developed a model of how distribution costs related to various parameters of the distribution operation, such as location and volumes of customer demands during a particular time period. This model was used to predict what costs would have been in the future if the vehicle schedule system had not been introduced, and therefore to provide a bench- mark against which to assess improvements (Marshall Fisher, private communication, December 1987~. But after-the-fact evaluation still does not answer a manager's question regarding investment in OR/MS technology prior to demonstrated beneficial results. I particularly like what Warren Powell, an expert in logistics, has to say about this problem area: For some companies it (the decision to invest in OR MS) is a pure cost/benefit decision, using very conservative estimates of cost savings as benefits. For other companies the decision is driven by one or two individuals "with a vision" that a model is critical to success. Operations people generally fall in the first group, marketing/finance types in the second. The person with a vision is critical to imple- mentation. In reality, success usually depends on making someone's life easier. Rigorously documenting savings is rare (e.g., numbers prepared for the Edelman Award are generally not reliable). Service and profit benefits are virtually impossible to quantify, because side-by-side analyses with and without a model are never available. It is most tempting to evaluate the value of a model in terms of how much money it saves each year. To be sure, it is a useful and often important exercise to at least try to estimate the economic impacts of a model.... The current emphasis on cost numbers is having the result that (a) only implementations at big companies which may yield substantial cost savings are important; (b) traditionally conservative people in operations, who often will acknowledge only savings they can rigorously verify, are to be avoided; and (c) traditional applications to operations, which yield direct cost savings are preferred over richer applications to improve pricing, marketing, customer service, or financial planning with notoriously intangible benefits (Warren Powell, private communication, November 1987~. It may be that no satisfactory formal mechanism will ever be devised for deciding before-the-fact whether or not to invest in OR/MS. For certain narrowly defined applications areas, one can simulate proposed new pro- cedures and compare them with status quo procedures to assess potential benefits. But the cases in this paper illustrate that the greatest potential benefits of an information/knowledge technology such as OR/MS are orga- nizational, affecting fundamentally the ways firms manage and operate. Sometimes the OR/MS model serves as the catalyst for managers from dis

138 RICHARD C. LARSON parate departments within a firm to communicate; perhaps they should have communicated before creation of the model, but the at once bald and sci- entifically neutral assumptions of the model can focus a group's discussions on difficult decisions. OR/MS products implemented on a day-to-day basis, by affecting information flows and providing immediate evaluative results of decisions, can markedly change managerial behavior. In increasingly competitive environments one can argue that the effective processing of data to develop decision consequential information remains for many services industries a viable mechanism for achieving competitive ad- vantage. For the impacts of OR/MS and related information technologies to grow, many managers may need a new point of view regarding investment. According to George Kozmetsky, "Managers need to understand that in- formation, science, and technology are not free economic goods but are assets to be used, planned, earned on, and replenished" (Kozmetsky, 1984, p.4~. ACKNOWLEDGMENTS I would like to thank the MIT School of Engineering for providing support for developing OR/MS course material (of which this paper is a part) in a new engineering schoolwide undergraduate elective on operations research in engineering. I would also like to thank the National Academy of Engi- neering for supporting my very productive research assistant, Luiz F. M. Vieira, who is a doctoral candidate in operations research at MIT. Finally, particular thanks are due to Bruce Guile at NAE who carefully read earlier versions of this paper and greatly contributed to its final form. APPENDIX A DETAILED CASE: SCHEDULING 911 OPERATORS In 1968 the mayor of New York, John Lindsay, opened the first-in-the- nation big-city "911 system" for responding to calls for emergency service (police, fire, ambulance) from the public. To call the police one formerly had to memorize seven emergency telephone numbers, one for each of seven dispatching zones throughout the city. (And, not insignificantly, one had to know from which dispatching zone one was calling.) The new system allowed a caller simply to dial "911" from anywhere in the city. Approximately 15,000 calls per day were processed by "911 operators" located at a central dispatch and communications room of the New York City Police Department (NYPD) in lower Manhattan. Within weeks after opening the new facility, complaints started pouring in (by telephone, letters, radio talk shows, and letters to the editor) that the new multimillion dollar system that was supposed to speed processing of calls was plagued with

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES 139 delays. One letter to the editor of the New York Times complained of calling 911 on a Saturday evening and waiting more than 25 minutes for someone to answer the phone; eventually the caller thought he might have called the "wrong number," so he hung up and tried again, this time getting an answer after "only" approximately 20 minutes of waiting. These types of complaints forced the police commissioner to put together a study team to analyze the problem. I was one of three members of the team, the other two being police lieutenants trained in police planning and operations. Quickly we jointly discovered that the hourly average volume of 911 calls varied predictably by a factor of eight or more on a daily basis (and more when measured over a week), while the hourly number of 91 1 telephone operators varied only by a factor of two (with maximum deployment averaging 25 operators during all hours of the day except the early morning period, 3:00 to 7:00 A.M., when the number of operators dropped to 12 or 131. In other words, hourly deployment of operators virtually ignored predictable changes in call vol- umes, except during the very quiet early morning hours. We desired to develop (in 1 month) an easy-to-use scheduling procedure that took advantage of economies of scale. Although the 911 system incor- porated several complications not found in more standard telephone an- swering systems, we found it acceptable as an approximation to apply Erlang's original formulas (circa 1915) describing the operating properties of multis- erver queues to schedule the operators. Two interesting encounters during the implementation process, both with a senior managing police officer, deserve mention. First, when in a formal briefing I displayed graphically the data showing the true (deplorable) state of affairs with regard to queue delays (with 40 percent of callers on Saturday evenings experiencing delays greater than 30 seconds), the senior officer declared that the data I was using were inaccurate; after all, his officer in charge of the Communications Division had informed him that there were few problems and that the loud public outcries were not representative of the service levels being provided. Luckily, the two lieutenants and I had worked together side-by-side within the Communications Division for 1 month; when questioned by the senior officer, the lieutenants verified the accuracy of the data. Second, after the presentation of the data, I requested from the senior officer his department's "performance objectives" with regard to 911 op- erator scheduling. In particular, I requested from him two numbers, T and P. such that operators would be scheduled so that during no hour would more than P percent of the callers incur delays greater than T seconds. For instance, if T and P were set at 15 seconds and 5 percent, respectively, I would use Erlang's formulas to schedule 911 operators each hour so that no more than 5 percent of the callers would experience delays exceeding 15

140 RICHARD C. LARSON seconds. At first the senior officer refused to give any such numbers. Then, when pressed, he relented, announcing his department's values: T = P = 0.00 (I. Queues with uncertainty can virtually never achieve such perfect operation with a finite amount of resources. Ultimately, we back- tracked and rescheduled the department's existing number of operators on a weekly basis, achieving major reductions in delays and "time equity" in level of service (i.e., with all hours of the week having nearly the same delay characteristics) [see Larson (197211. Unlike many other operations research studies, this one was implemented in its entirety within 1 month after completion. Prior to the study, 17 percent of all 911 calls had been delayed 15 seconds or more (when averaged over an entire month), with terrible congestion at predictable times (e.g., 40 percent of calls delayed more than 30 seconds from 8:00 P.M. to midnight, Saturday evenings). After implementation of the recommendations of the study, no hour of the week experienced more than 5 percent of the calls having 15-second delays. The "product" of the study consisted of seven charts or tables, each containing for a particular day of the week the recommended number of 911 operators to assign each hour of the day. The "cost" to the NYPD was 3 person-months of profes- sional effort. No additional 911 operators were hired; rather the hours of working operators were simply reassigned. If additional 911 operators had been hired under the "old" scheduling scheme to obtain the same new performance levels at all hours of the week, the operator pool would have increased by approximately 50 percent. Before leaving this case, two other reflections are in order. We found early on in the study that nearly all the data we needed had been recorded meticulously by a full-time officer whose only job was to place the operating statistics into a large loose-leaf book. By 5:00 P.M. each day, he had com- pleted the previous day's entries, inserted the final completed sheet of num- bers, and went home. Not one decision had been influenced by the entries in the book! As far as we could tell, virtually no one other than this recording officer had ever looked at the numbers. This book was so complete (and accurate) that for our work we needed only approximately 10 percent of its entries. To ensure implementation, the two lieutenants decided to modify the in- centive and reward system for the captains on duty in the 911 center. The "data recorder's" job was modified so that the first thing he or she did each morning was to draw a large graph displaying in an hour-by-hour fashion the previous day's performance. The name of the captain on duty for each 8-hour tour of duty was prominently displayed, as well. Each time during a tour that one or more callers experienced a call-answering delay exceeding 30 seconds, bells would sound and lights would go on and revolve, not unlike that in modern discotheques. Each such event was labeled a "bell." On the

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES 141 large graph, next to each captain's name, the data recorder displayed in large font the number of "bells" incurred on that captain's 8-hour tour of duty. This display was one of the first things seen by the many New York citizens, school children, and tourists who were given tours of the 911 center each day. Needless to say, the number of bells was kept to a reasonable minimum REFERENCES AND BIBLIOGRAPHY Bartholdi, John J. III, and Loren K. Platzman, R. Lee Collins, and William H. Warden III. 1983. A minimum technology routing system for meals on wheels. Interfaces 13(June): 1 8. Bell, W. J., L. M. Dalberto, M. L. Fisher, A. J. Greenfield, R. Jaikumar, P. Kedia, R. G. Mack, and P. J. Prutzman. 1983. Improving the distribution of industrial gases with an on- line computerized routing and scheduling optimizer. Interfaces 13(June):4-23. Bellman, R. E. 1957. Dynamic Programming. Princeton, N.J.: Princeton University Press. Bodin, L., B. Golden, A. Assad, and M. Ball. 1983. Routing and scheduling of vehicles and crews: the state of the art. Computers and Operations Research 10(1):63-211. Boykin, Raymond F. 1985. Optimizing chemical production at Monsanto. Interfaces l5(January/ February):88-95. Brandeau, M., and R. C. Larson. 1985. Extending and applying the hypercube queueing model to deploy ambulances in Boston. In Urban Service Systems, A. Swersey and E. Ignall, eds. New York: North Holland, pp. 121-153. Browne, Jim. 1984. Management and Analysis of Service Operations. New York: North- Holland. Carlisle, David P., Kenneth S. Nickerson, Stephen B. Probst, Denise Rudolph, Yosef Sheffi, and Warren Powell. 1987. A turnkey micro-computer based logistics planning system. Interfaces 17(July/August): 16-26. Charnes, A., and W. W. Cooper. 1984. Creative and Innovative Management. Essays in Honor of George Kozmetsky. Cambridge, Mass.: Ballinger. Cochard, Douglas D., and Kirk A. Yost. 1985. Improving utilization of Air Force cargo aircraft. Interfaces l5(January/February):53-68. Dantzig, George B. 1963. Linear Programming and Extensions. Princeton, N.J.: Princeton University Press. Davis, Samuel G., George B. Kleindorfer, Gary A. Kochenberger, Edward T. Reutzel, and Emmit W. Brown. 1986. Strategic planning for bank operations with multiple check-pro- cessing locations. Interfaces 16(November/December): 1- 12. Eaton, David J., Mark S. Daskin, Dennis Simmons, Bill Bulloch, and Glen Jansma. 1985. Determining emergency medical service vehicle deployment in Austin, Texas. Interfaces l5(January/February):96- 108. Edwards, Jerry R., Harvey M. Wagner, and William P. Wood. 1985. Blue bell trims its inventory. Interfaces l5(January/February):34-52. Finke, Gary F. 1984. Determining target inventories of wood chips using risk analysis. In- terfaces 14(September/October):53-58. Fitzsimmons, James A., and Robert S. Sullivan. 1982. Service Operations Management. New York: McGraw-Hill. Ford, L. R., Jr. and D. R. Fulkerson. 1962. Flows in Networks. Princeton, N.J.: Princeton University Press. Gardner, Everette S., Jr. 1987. A top down approach to model-in" U.S. Navy inventories. Interfaces 17(July/August):l-7.

142 RICHARD C. LARSON Gelb, Betsy, D., and Basheer M. Khumawala. 1984. Reconfiguration of an insurance com- pany's sales regions. Interfaces 14(November/December):87-94. Goeller, Bruce F., and the PAWN Team. 1985. Planning the Netherland's water resources. Interfaces l5(January/February):3-33. Guile, Bruce R., and Harvey Brooks, eds. 1987. Technology and Global Economy. Wash- ington, D.C.: National Academy Press. Haehling von Lanzenauer, Christoph, Erwin Harbauer, Brian Johnson, and David H. Shuttle- worth. 1987. RRSP Flood: LP to the Rescue. Interfaces 17(July/August):27-40. Harrison, H. 1986. Management science and productivity improvement in Irish Milk Coop- eratives. Interfaces 16(July/August): 31-40. Hax, Arnoldo C., ed. 1987. Planning Strategies that Work. New York: Oxford University Press. Hehnen, M. T., S. C. Chou, H. L. Scheurman, G. J. Robinson, T. P. Luken, and D. W. Baker. 1984. An integrated decision support and manufacturing control system. Interfaces 14(September/October): 44-52. Heskett, James L. 1986. Managing in the Service Economy. Boston, Mass.: Harvard Business School Press. Holloran, Thomas J., and Judson E. Byrn. 1986. United Airlines station manpower planning system. Interfaces 16(January/February):39-50. Huysmans, Jan H. B. M. 1970. The Implementation of Operations Research. New York: John Wiley & Sons. Lkura, Yoshiro I., George Gross, and Gene Sand Hall. 1986. PGandE's state-of-the-art sched- uling tool for hydro systems. Interfaces 16(January/February):65-82. Kantorovich, R. E. 1939. Mathematical methods in the organization and planning of production. Publication House of the Leningrade State University, 68 pp. Translated in Management Science, 1960. 6:366-422. Kleutghen, P. P., and J. C. McGee. 1985. Development and implementation of an integrated inventory management program at Pfizer Pharmaceuticals. Interfaces l5(January/Febru- ary):69-87. Kozmetsky, George. 1984. Creative and innovative management, a new academic frontier. Pp. 3-25 in Creative and Innovative Management. Essays in Honor of George Kozmetsky, A. Charnes and W. W. Cooper, eds. Cambridge, Mass.: Ballinger. Larson, Richard C. 1972. Improving the effectiveness of New York City's 911. Pp. 151-180 in Analysis of Public Systems, A. W. Drake, R. L. Keeney, and P. M. Morse, eds. Cambridge, Mass.: MIT Press. Larson, Richard C. 1988. Transporting sludge to the 106 mile site: an inventory/routing model for fleet sizing and logistics system design. Transportation Science. Larson, Richard C. 1972. Urban Police Patrol Analysis. Cambridge, Mass.: MIT Press. Larson, Richard C., Alan Minkoff, and Paul Gregory. 1988. Fleet sizing and dispatching for the marine division of the New York City Department of Sanitation. In Vehicle Routing: Methods and Studies, Bruce Golden and Arjang Assad, eds. New York: John Wiley & Sons. Larson, Richard C. and Amedeo R. Odoni. 1981. Urban Operations Research. Englewood Cliffs, N.J.: Prentice-Hall. Lembersky, Mark R., and Uli H. Chi. 1986. Weyerhaeuser decision simulator improved timber prof~ts. Interfaces 16(January/February):6- 15. Liberatore, Mathew J., and Tan Miller. 1985. A hierarchical production planning system. Interfaces 15 (July/August): 1 - 13. Magirou, Vangelis F. 1986. The eff~cient drilling of printed circuit boards. Interfaces 16(July/ August): 13-23. Maltz, Michael D. 1984. Recidivism. Orlando: Academic Press.

OPERATIONS RESEARCH AND THE SERVICES INDUSTRIES 143 Martin, Clarence H., and Sanford L. Lubin. 1985. Optimization modeling for lousiness planning at Trumbull Asphalt. Interfaces l5(November/December):66-72. Martland, Carl, Henry S. Marcus, and George B. Raymond, Jr. 1986. Boston and Maine achieves control over railroad performance. Interfaces 16(September/October): 1- 16. McCloskey, Joseph F. 1987. The beginnings of operations research: 1934-1941. Operations Research 35(1):143-152. Morse, P. M. 1977. In at the Beginnings. Cambridge, Mass.: MIT Press. Oliff, Michael D., and E. Earl Burch. 1985. Multiproduct production scheduling at Owens- Corning Fiberglass. Interfaces l 5(September/October) :25-34. Quinn, James Brian. 1987a. Managing strategic change. Pp. 1-18 in Planning Strategies that Work, Arnoldo C. Hax, ed. New York: Oxford University Press. Quinn, James Brian. 1987b. The impacts of technology in the services sector. Pp. 119-159 in Technology and Global Economy, Bruce Guile and Harvey Brooks, eds. Washington, D.C.: National Academy Press. Parsons, Gregory L. 1987. Information technology, a new competitive weapon. Pp. 196-211 in Planning Strategies that Work, Arnoldo C. Hax, ed. New York: Oxford University Press. Riccio, Lucius J., Joseph Miller, and Ann Lithe. 1986. Polishing the big apple: How man- agement science has helped make New York cleaner. Interfaces 16(January/February):83- 88. Shapiro, Roy D., and James L. Heskett. 1985. Logistics Strategy, Cases and Concepts. St. Paul, Minn.: West Publishing Co. Stankard, Martin F. 1981. Successful Management of Large Clerical Operations. New York: McGraw-Hill. Sullivan, Robert S., and Stephen C. Secrest. 1985. A simple optimization DSS for production planning at Dairyman's Cooperative Creamery Association. Interfaces l5(September/Oc- tober):46-53. Swersey, Arthur J., and Edward J. Ignall. 1986. Delivery of Urban Services. Amsterdam: North-Holland. Terry, L. A., M. V. F. Pereira, T. A. Araripe Neto, L. F. C. A. Silva, and P. R. H. Sales. 1986. Coordinating the energy generation of the Brazilian National Hydrothermal Electrical Generating System. Interfaces 16(January/February): 16- 38. Walker, W. E., J. M. Chaiken, and E. J. Ignall, eds. 1979. Fire Department Deployment Analysis. Amsterdam: North-Holland. Welch, Norma, and James Gussow. 1986. Expansion of Canadian National Railway's line capacity. Interfaces 16(January/February):51-64. Whitmore, William F. 1953. Edison and operations research. Operations Research 1:83-85.

Next: Cellular Mobile Telephone Services »
Managing Innovation: Cases from the Services Industries Get This Book
×
Buy Paperback | $45.00 Buy Hardback | $45.00
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

This book of case histories is devoted solely to service industries and the technologies that drive them, as told by those who have developed segments of these industries. The chapters cover innovations such as Federal Express's advanced system for package tracking, Citicorp's development of the Automated Teller Machine, AT&T's experience with mobile telephones, Bell & Howell's introduction of an automated automotive parts catalog, and the New York Stock Exchange's development of electronic trading. Some broader analyses discuss the interfaces between services technologies and manufacturing, operations research in services, and technology in professional services.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!