Advances in Services Innovations

Dieter Spath · Klaus-Peter Fähnrich (Eds.) Advances in Services Innovations Dieter Spath · Klaus-Peter Fähnrich (Eds...

Author: Dieter Spath | Klaus-Peter Fähnrich

48 downloads 1553 Views 4MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Dieter Spath · Klaus-Peter Fähnrich (Eds.) Advances in Services Innovations

Dieter Spath · Klaus-Peter Fähnrich (Eds.)

Advances in Services Innovations With 106 figures and 11 tables

123

Professor Dr.-Ing. Dieter Spath IAT Institut für Arbeitswissenschaft und Technologiemanagement Universität Stuttgart Nobelstraße 12 70569 Stuttgart Professor Dr.-Ing. habil. Dipl.-Math. Klaus-Peter Fähnrich IfI Institut für Informatik Universität Leipzig Augustusplatz 10–11 04109 Leipzig

Library of Congress Control Number: 2006934195 ISBN-10 3-540-29858-4 Springer Berlin Heidelberg New York ISBN-13 978-3-540-29858-8 Springer Berlin Heidelberg New York This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable for prosecution under the German Copyright Law. Springer is a part of Springer Science+Business Media springer.com © Springer-Verlag Berlin Heidelberg 2007 The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Typesetting by the authors using a Springer TEX macro package Production by LE-TEX Jelonek, Schmidt & Vöckler GbR, Leipzig Cover design by deblik Berlin SPIN 11573623

7/3100YL - 5 4 3 2 1 0

Printed on acid-free paper

Preface Developing innovative services and launching them successfully in international markets – these are major challenges for enterprises and national economies which aim to benefit from the opportunities of the service sector by achieving more growth and employment. Those enterprises and sites which succeed in creating “a special service experience” by offering outstanding service solutions and by excellent performance in service delivery to the customer will be successful in competition. Nevertheless, it is exactly the current discussion about innovation, particularly in Germany, which reveals that we do not primarily suffer from a lack of good ideas but rather that the translation of new findings and ideas into new services, products and processes must be substantially improved. All in all, it is obviously very urgent to increase innovative ability and to accelerate the speed of innovation. Progress on this path can be accelerated if science and research increasingly face up to the challenges of how to improve innovative power in the service sector. In my opinion, service innovations address very different fields of innovation such as, for example, linking not only new technologies and services, performance and process innovations, but also market-related innovations, for example, for the establishment of new distribution channels. On the one hand, distinct improvements have occurred in the past few years with regard to the willingness to also invest in research and development in the domain of services in order to meet these challenges and, on the other hand, the boundaries of the research disciplines are more and more often transcended for an interdisciplinary and international cooperation and for a holistic view. Accordingly, the international workshops at the Fraunhofer Institute for Industrial Engineering (IAO) in Stuttgart were not only an attempt to devise a roadmap for future R&D activities but also to present different views of the challenges of improving innovative ability in the service sector from the point of view of various disciplines of research – ranging from service marketing and management to service engineering. This book documents many of the opinions presented by the experts during these workshops and as such will hopefully provide valuable input to the community. Stuttgart, Summer 2006 Hans-Jörg Bullinger, President of Fraunhofer - Gesellschaft, Germany

Table of Content Preface ...................................................................................................................V

I

Service Engineering

Service Engineering: State of the Art and Future Trends Klaus-Peter Fähnrich, Thomas Meiren ....................................................................3 The Palm/Erlang-A Queue, with Applications to Call Centers Avishai Mandelbaum, Sergey Zeltyn ....................................................................17 The Design and Development of Industrial Service Work Holger Luczak, Christian Gill and Bernhard Sander .............................................47 An Engineering Tool for the Conceptual Design of Service Systems Reuven Karni, Maya Kaner ...................................................................................65 Integrated Development of Software and Service Dieter Spath, Christian van Husen, Kyrill Meyer, Romy Elze ..............................85

II

Service Management

From Service Management towards Service Competence Urs Fueglistaller...................................................................................................113 Innovation and Learning in Services - The Involvement of Employees Jon Sundbo...........................................................................................................131 Managing Service Networks’ Success Heiner Evanschitzky, Dieter Ahlert .....................................................................151 Success Factors in New Service Development and Value Creation through Services Bo Edvardsson, Anders Gustafsson, Bo Enquist .................................................165

III

Service Marketing

Sustainable Advantages in Service Industries Thorsten Posselt, Pablo Berger ............................................................................187 Satisfaction Measurement within the Customer Relationship Life Cycle Bernd Stauss, Matthias Gouthier, Wolfgang Seidel.............................................205

VIII

IV

Sustainable Service Research

Strengthening the Services Sector – Needs for Action and Research Walter Ganz.........................................................................................................223 Standardisation in the Service Sector for Global Markets Inka Mörschel, Hermann Behrens, Klaus-Peter Fähnrich, Romy Elze................257 Research and Development for a Sustainable Services Sector Veronika Pahl ......................................................................................................279 Future Research and Needs for Action Klaus Peter Fähnrich, Romy Elze........................................................................289 Index of Authors ................................................................................................307 Index ...................................................................................................................311

I Service Engineering

Service Engineering: State of the Art and Future Trends Klaus-Peter Fähnrich1, Thomas Meiren2 1

Chair of Business oriented Information Systems, Institute for Computer Science, University of Leipzig, Germany 2 Fraunhofer Institute for Industrial Engineering IAO, Germany

1 Introduction.........................................................................................................4 1.1 Standing out from the competition through innovative services ...............4 1.2 Services as an R&D object ........................................................................5 2 Fundamental aspects of developing a service ...................................................7 2.1 Model for developing services ..................................................................7 2.2 Use of methods and tools ..........................................................................9 3 Organisation of service development ..............................................................11 3.1 Basic organisational alternatives .............................................................11 3.2 Dissemination in practice ........................................................................12 4 Outlook ..............................................................................................................14 References.............................................................................................................14

4

Klaus-Peter Fähnrich, Thomas Meiren

1 Introduction 1.1 Standing out from the competition through innovative services As the significance of the service sector has grown, competition in many service markets has increased markedly over the last few years. Markets that used to be sluggish have transformed themselves and new players have come onto the market: there is no question that the marketplace is becoming more dynamic. When one considers the causes for the more intense competition, the following factors apply in particular: x x x x x

Increased deregulation, The entry of new competitors, Multiplication of successful service concepts, Increasing market saturation, Strategic overcapacity.

Against this background, service companies are no longer able to raise their profile simply by emphasising cost, image or quality advantages. On the contrary, differentiation through innovative service offerings is developing into a key unique selling point for them to set themselves apart from their competitors. The central challenges compel them above all to offer continuously improved and new services in the marketplace, to always remain one step ahead of the competition and at the same time to comply exactly with customer needs and expectations. However, many companies are today facing the problem that their present structures and processes become unsuitable both for developing new services efficiently and for positioning them in the marketplace. Moreover, adequate instruments for operative planning of processes to develop services are lacking. Very often the roots of the difficulties lie in the fact that the services offered by the companies are not clearly defined, i.e. there are no clear descriptions of what the service entails, the relevant processes and the resources required (Bullinger et al. 2003). Even in research, the development of new services has been addressed intensively at a relatively late stage. Although service development is now a highpriority topic (refer to the findings of an international expert survey in Ganz and Meiren 2002), in the past it was never a prominent focus of either business or engineering research. By simply emphasising the importance of developing new services, most of the work published to date fails not only to offer concrete support but also to anchor this process in strategic and operative enterprise management. Although ‘New Service Development’ began to find its way into AngloAmerican research literature as early as the 1970s and 1980s, it was at a somewhat rudimentary level (Bowers 1985). This research focused particularly on the basic underlying conditions, success factors and obstacles preventing the development of new services but the results only very rarely delivered concrete instruments that could be applied in practice. Looking at the present situation, it can be seen that at least considerably more attention is being paid to this subject, as illustrated by an

Service Engineering: State of the Art and Future Trends

5

increasing number of publications (e.g. Cooper and Edgett 1999; Edvardsson et al. 2000; Fitzsimmons and Fitzsimmons 2000). In contrast to highly marketing-oriented New Service Development, Service Engineering entails applying the appropriately modified engineering know-how established in the field of conventional product development to the development of services. Service Engineering can thus be defined as the systematic development and design of services using suitable models, methods and tools. Despite the fact that the term ‘Service Engineering’ was first coined in the literature as early as the mid-eighties (Shostack 1982), Albrecht and Zemke commented at the time: ‘The developing art/science of service engineering is so new that it really hasn’t an agreed-upon name, much less an established body of principles and techniques’ (Albrecht and Zemke 1985). It was only in the mid-1990s that service engineering began to attract greater attention, not least thanks to research initiatives in Germany and Israel. A wealth of experience has since been accumulated with proven guidelines and instruments (refer also to Meiren and Barth 2002, Bullinger and Scheer 2003). Apart from New Service Development and Service Engineering, the term ‘Service Design’ also appears in the literature. If one looks at these concepts more closely, Service Design primarily covers the perceptible elements of a service (e.g. colours, sounds, smells) at the direct interface to the customer (Erlhoff et al. 1997). However, Anglo-American researchers, in particular, interpret this term more broadly. In these countries, ‘Design’ traditionally encompasses all aspects of the actual design of a product and consequently Service Design primarily concerns procedures and methods for developing new services. In terms of what it actually involves, this work is very similar to Service Engineering (Ramaswamy 1996). 1.2 Services as an R&D object The common factor in all scientific approaches is that they must answer the central question about how services can be developed across the board. It will be necessary to clarify which aspects of a service can actually be developed and how these aspects can be structured in a suitable way. A more pronounced product-orientated view may be helpful to develop the argument at this point and this means that services must be seen more as separate products than they have in the past. In this context, the term ‘Product’ is deliberately used as the superset for all objects – goods, services, software etc. offered by the company in the marketplace (Sabisch 2000). In particular, these are products that form the interface between the company and customers. A clear product definition remains the fundamental prerequisite for the development, manufacturing and marketing of services. The consequence of applying a modern product definition to services is that these are no longer viewed as a sort of ‘black box’ but instead are viewed as a designable part of business activities (Fähnrich et al. 1999). The starting point is the approach for designing services presented below. Here an important role is played by what are termed external factors, i.e. persons, animals, goods, rights and information, which are directly integrated in the provision

6


of services by the party making the request as opposed to those aspects that relate to the manufacture of the goods themselves. A typical service can thus be said to be characterised by three different dimensions: x A structure dimension (the structure determines the ability and willingness to deliver the service in question), x A process dimension (the service is performed on or with the external factors integrated in the processes), x An outcome dimension (the outcome of the service has certain material and immaterial impacts for the external factors). An examination of the subject in this way produces three dimensions (structure, process, outcome), which are characteristic for a service and which ought to be taken into account during development. It is now sensible to elaborate appropriate outcomes for each of these dimensions in the development process, i.e. the results of service development are resource models, process models and product models (Fähnrich et al. 1999; Bullinger and Meiren 2001). Figure 1 illustrates the relationships. Customers requesting the services involve themselves or an object in the process (external factor)

Service providers are willing and able to provide a service

The provision of a service is depicted as a process

Services have material or immaterial consequences as a result of rendering a service

(Structure dimension)

(Process dimension)

(Outcome dimension)

Resource model

Process model

Product model

Service concept

Fig. 1. Services as an R&D object

The term resource models groups together development tasks which describe the provision of services. Here the key aspect is the planning of resources, needed for the subsequent provision of services. These include, in particular, the production of concepts for human resources (primarily relating to the selection and qualification of personnel) but also the planning of material resources and the concept


7

for the information and communications technology that will be used to support the service. A further development task involves the preparation of process models for the provision of the service developed. Here processes are documented with the aim of creating transparency at an early stage and achieving the greatest possible process efficiency before the service is actually offered. The aim is to eliminate activities that do not add value and remove unnecessary interfaces and media discontinuities. Apart from process models, product models are also to be prepared when developing services. These provide a description of the characteristics of the service: in particular, a definition is required of the service content and outcomes. Quality and performance standards for the new service are to be defined both for external and internal purposes. The approach presented for putting the service, as an R&D object, on a systematic basis can be applied to the vast majority of services on account of the generic character of these services. However, this approach must be integrated in specific processes and suitable methods applied, if the service is to be developed and designed so that it is suitable for application in practice.

2 Fundamental aspects of developing a service 2.1 Model for developing services Apart from the fundamental aspects of developing services, the sequence in which certain activities are to be undertaken during the development process is of special interest. Particularly those companies which regularly develop new services are seeking ways to eliminate redundant work, prevent past mistakes from being repeated and reuse existing know-how. The first step towards achieving this objective is generally to describe the development processes concerned and then to formalise the individual R&D steps up to a certain point (Fähnrich et al. 1999; Bullinger and Meiren 2001). Development processes designed in this way can be based on so-called process models. Process models document project activities and project responsibilities in detail, enabling them to support project planning, project steering and controlling. Until now, such models have been found mainly in traditional product development and software engineering, yet the basic principles behind them mean they can also be effectively applied to the field of service development (Hofmann et al. 1998). Process models for service development x Define the activities that are necessary to develop services and the sequence in which these activities are undertaken, x Create the prerequisite for successful and efficient positioning of new services in the marketplace by systematising service development,

8


x Integrate customers into the development process at an early stage in order to satisfy the specific characteristics of services. Many different process models are familiar from product and software development. But which of these process models is most suitable for services? For various reasons, there is no all-encompassing answer to this question. Firstly, the selection of a suitable process model depends on the type of service to be developed. Secondly, to date very little practical experience has been gained in the use of different process models. However, some authors point out that it is primarily waterfall models, which are characterized by the sequential arrangement of development tasks that have been most prevalent in both scientific and practical environments (e.g. Sontow 2000). A process for the development of services is explained below. This is based on the waterfall model set out in the DIN report on Service Engineering and is divided into the phases of idea management, the requirements analysis, conceptualisation of the service, implementation and market introduction. If one links this waterfall model to the dimensions presented in the previous chapter, it produces the model shown in Fig. 2 on the development of services. Idea management

Requirements analysis

Service conceptualisation

Service implementation

Market launch

Structure dimension Process dimension Outcome dimension Market dimension

Fig. 2. Model on the development of services (Meiren 1999, Meiren and Barth 2002)

In addition to the three ‘constitutive” dimensions for the structure, process and outcome, this model also contains a further dimension – the market dimension. The consideration behind this was that modern development processes must always be guided by market requirements, if not, there is a danger that newly developed services will ignore the customers' actual needs, leading to their inevitable failure in the marketplace. What being guided by the market actually means in this context is that, around the time when the development work starts, the services need to be coordinated with the actual situation in the market place and market research needs to be carried out to assess whether the new services are likely to be successful or not. In addition to this, the marketing concepts and how these will be implemented during the market launch of new services need to be drawn up.


9

2.2 Use of methods and tools

low

Contact intensity

high

In practice, developing successful services entails more than simply carrying out development work as efficiently as possible. Operational support in the shape of suitable methods and adequate (software) tools is essential (Bullinger et al. 2003). Although a rich stock has been built up as a result of product and software development, these methods and tools cannot be transferred indiscriminately to the service sector. Particularly in the case of services which exhibit a high degree of intangibility, or where the interaction of customers and employees is crucial, illconsidered attempts to apply classic product development methods are likely to be quickly doomed to failure. It is a sensible course of action here to establish characteristic ‘types’ of services and use these as a basis for further consideration in order to carry out suitable analyses and make recommendations on what action to take. To date, scientists have developed a series of what are termed typologies for the service sector but only very rarely have these been within the context of the development of services. The typologisation approach developed by Fähnrich et al. (1999) is an exception here and it also has the advantage that it has been derived empirically from a survey of 282 companies, thus giving it a strong practical bias. A factors analysis highlighted that contact intensity and variety were the crucial typologisation categories (Fig. 3).

Customer-focused services

Knowledge-focused services

Examples: Call center Retail trade

Examples: Consulting Market research

Process-focused services

Flexibility-focused services

Examples: Automatic car wash Online banking

Examples: Life insurance IT outsourcing services

low

high Variety

Fig. 3. Service typology (based on Fähnrich et al. 1999)

Here contact intensity can be seen as a yardstick of the interaction between employees and customers, whereas variety describes the number of variants in ser-

10


vices as they pertain to the product. The four service types below are based on these two typologisation categories: x Process-focused services of which low contact intensity and low variety are a feature, making it particularly suitable for process-orientated standardization measures, x Flexibility-focused services with low contact intensity and high variety, whereby from the developer's point of view the emphasis is on the systematic variant creation aspect, x Customer-focused services which are typified by high contact intensity and low variety and essentially consist of a clearly defined standard service but which may be influenced by customers within certain limits, x Knowledge-focused services distinguished by high contact intensity and high variety, which typically necessitate a considerable amount of customizing. It would be interesting at this stage to consider which methods are preferred for developing which service types. A series of methods familiar from traditional product development is evidently used in practice for services with a relatively low contact intensity. These include Quality Function Deployment (QFD), Structured Analysis and Design Technique (SADT), Failure Mode and Effects Analysis (FMEA) and, in particular, various product and process modelling methods. The latter methods also include advances, which were specifically implemented for the services sector. These include Service Blueprinting, which primarily separates the process steps into those steps that are visible to the customer and those that are not visible (Shostack 1984). If one looks at all this methods, one possible explanation for their application might be that the performance of a small number of contactintensive services is only influenced to a very limited extent by customer-imposed variances, so that the characteristics exhibited by these services bear numerous resemblances to those of physical goods and the services concerned can consequently be developed using similar methods. Whereas engineering methods are relatively widespread as instruments for developing services with a low contact intensity, their relevance for the development of contact-intensive services is considerably less (Fähnrich et al. 1999). Business and recently also a few service-specific methods predominate here – especially when it is important to ensure that customer retention is systematically built into the service development process. Factors that also come into play in the case of knowledge-focused services are social and behavioural science methods, tailored to qualifying employees or shaping customer interaction. It therefore appears that it is specifically the criterion of contact intensity that provokes a split when it comes to the methods preferred in practice: ‘It is evident that, particularly with service types where so-called soft factors play a vital role, traditional product development methods are no longer transferable and approaches originally devised by other scientific disciplines are demanded more and more frequently’ (Bullinger and Meiren 2001). Finally, at this point it is appropriate to highlight a further difference from product and software development. Although the methods that are used to develop services are now increasingly being targeted, there is no seamless link between


11

these methods and no integrated software support (Freitag 2002). Although a few initial approaches have been made (Herrmann et al. 2003; Fähnrich and Meiren 2004), development platforms that compare with the Computer-Aided Design Tools (CAD) in product development or Computer-Aided Software Engineering Tools (CASE) for software development will not be available for a long time yet.

3 Organisation of service development 3.1 Basic organisational alternatives Experience has shown that one of the questions that first confronts companies which increase their investment in the development of new services is who in the company can actually take on this task and which organisational units in the company are to provide support. For example, the rising cost of development prompts the question of whether it is sensible to create a separate department for service development. When deciding how to organise responsibilities for developing services, the following four alternatives can be considered (Bullinger and Meiren 2001): x Establish a separate organisational unit for developing services (e.g. Service Development), x Tasks taken over by organisational units that already exist (e.g. Marketing, Sales), x Form special project teams to work together temporarily to develop services, x Outsource the development of services to an external partner. A separate organisational unit to develop services can be created, the aim being to anchor service development permanently within the company. This can, for instance, take the form of a management team, a group, a department or even a centre (Luczak et al. 2000). These organisational forms offer the advantage that development work is then distinct from day-to-day operations and expertise can be specifically developed and is available to access when required. However, the associated expense is a disadvantage. Under certain circumstances, capacity may be kept ready for deployment but it may remain under-utilized at times because of the sporadic nature of service development activities. Another option is for tasks relating to service development to be handled by existing organisational units (more or less as an ‘additional task’). This has the advantage that although specific knowledge on development is always available in a clearly defined organisational unit, the capacity of this unit's members can be controlled more effectively (assuming that these employees can be flexibly deployed both for service development as well as for their normal duties). However, service development is often not seen as a core competence in such organisational units and thus only remains one activity among many. Another possible alternative is also the development of services in the form of a special project team. This solution minimises the structural changes that compa-

12


nies have to make and the members of the project team can be selected to suit the required terms of reference. A disadvantage of this is nevertheless that the expertise built up during such projects is often lost after the end of the project, when the employees involved return to their original departments and turn their attention to their former duties again. This is one reason why the establishment of a crossdepartmental working group must be listed as an interesting special case. Here employees from various divisions of the company work together with the aim of developing new services and they continue to work together permanently even when the individual development projects have been completed. The fourth basic option is to outsource the development of services to an external partner. This is a particular advantage if there is no expertise whatsoever within the company with regard to developing services or if this is not seen by the company itself as a core competence. A disadvantage that should, however, not be underestimated is the effort required to adapt service concepts developed by outside organisations to the specific environment of the company. Over and above this, it will probably also not be easy to find suppliers capable of undertaking the work required to develop new services. 3.2 Dissemination in practice Whereas all major industrial companies generally have clear arrangements in place for developing products and have their own Research and Development Department, it is to date very rare to find comparable teams for the systematic generation and implementation of service innovations. A look at the business practice in German companies showed that only a comparatively small proportion (9 percent) already have their own organisational department to develop services (Meiren 2004, see Fig. 4). Similar studies (e.g. Spath and Zahn 2003) also show that these organisational units generally entail a group reporting to management and are only rarely any more than a team. special organizational unit to develop new services

9

existing organizational unit to develop new services

83

new service development by specific project teams external development of new services

60

8

answering companies (per cent) multiple responses has been possible

Fig. 4. Organisation of service development (Meiren 2004)

n = 184


13

In most cases (83 percent), service development tasks are undertaken by other organisational units. In the study, management (61 percent), followed by Sales (53 percent), Service (47 percent) and Marketing (38 percent) were mentioned. Whereas in small companies, in particular, new services are developed by the management, indicating that services are developed on an ad hoc basis rather than systematically, Sales, Service and Marketing are divisions in the company that are close to customers. In the companies surveyed, this appears to be an important criterion for delegating the task of service development. A further option regularly practised is the formation of project teams, which are specifically established to develop new services. When it comes to the composition of the project teams, it is again the management and organisational units that are close to customers that play the major role. External cooperation partners and advisors are comparatively often involved in such project teams. Complete development of services by outside organisations is the situation that is least common. Only 8 percent of the companies surveyed said this was a practicable alternative. This is not particularly surprising because – even in the field of conventional product development – it is rare for development tasks to be outsourced in their entirety to external partners. All that can be noted is that the situation on the ground with regard to the organisational form for service development is that there does not appear to be any clear ‘one best way”. Consequently, a number of different options are selected in companies and these often operate in parallel. Also of interest here are the results of a survey of business experts. When one looks at the factors that point to successful organisation of service development, it can nevertheless be seen that there are a series of common factors across companies. It became apparent here that the following four factors were listed by more than half of the companies surveyed in each case (Meiren and Liestmann 2002): x Involvement of operative divisions in development activities, in particular utilization of the know-how of the employees on the spot, x Involvement of people to promote the idea, particularly from the company management but also those representing interest groups within the company, x Involvement of external partners, both for their specialist knowledge and also to benefit from the moderating effect they will have on those who work within the company, x Involvement of Sales and Marketing to ensure that the services developed are suitable for the market. A final consideration of the organisational forms selected by companies and the success factors clearly illustrate that the development of services is a task that affects many areas both within and outside the company. With this in mind, it can be seen that one of the main business challenges is definitely to find suitable organisational forms – in particular with regard to working in networks and dealing with interfaces – that will allow the existing complexity to be controlled and thus to establish what is actually essential for fast and efficient development of services.

14


4 Outlook Many service providers are increasingly under pressure to differentiate themselves from their competitors by means of new, innovative services and unique selling points. As a result of this need for concrete action in practice, service research is also increasingly addressing various issues relating to the development of services. In particular, systems and processes are being examined and developed which will enable new services to be launched on markets efficiently and, ultimately, profitably and yet still meet customer and employee needs appropriately. As the development of services is a comparatively recent field of study, which will benefit in particular from bringing together multi-disciplinary approaches, progress across the board can be expected in the future. A key area here will be the development and application of integrated methods and tools. In addition to the obvious consideration given to modelling services, the aspects that are particularly relevant here are the provision of integrated IT platforms for services along the lines of computer-aided service engineering tools. Further progress can be expected with regard to the design of development processes. It can therefore be assumed that the prototyping models for services which, to date, have only existed in a rudimentary form, will be developed further and take on greater significance. Generally prototyping involves a first version (‘Prototype’) of a new service being developed. This is then used to examine the important attributes of the service at an early stage and these features are then refined. Typically, the development steps of prototyping models are not discrete and may be partially overlapping. Issues relating to the testing of services are closely linked to prototyping. Only if it is possible for services to be fully or partially tested before their actual market launch will it be possible for new concepts to be evaluated at an early stage, improved and then implemented so that they are suitable for customers and employees. Research on visualising and staging services plays a special role here. It is precisely the processes of Virtual Reality and Service Theater used here that demonstrate that a number of different scientific disciplines benefit the development of services in many way and the view at the present time is that a number of trendsetting results can be expected.

References Albrecht K, Zemke R (1985) Service America! Doing Business in the New Economy. Dow Jones-Irwin, Homewood Bowers MR (1985) An Exploration into New Service Development: Process, Structure and Organization. Texas A&M University Bullinger H-J, Meiren T (2001) Service Engineering. In: Bruhn M, Meffert H (eds) Handbuch Dienstleistungsmanagement. Von der strategischen Konzeption zur praktischen Umsetzung. Gabler, Wiesbaden, pp 149-175


15

Bullinger H-J, Scheer A-W (eds) (2003) Service Engineering. Entwicklung und Gestaltung innovativer Dienstleistungen. Springer, Berlin Heidelberg New York Bullinger H-J, Fähnrich K-P, Meiren T (2003) Service Engineering. Methodical Development of New Service Products. International Journal of Production Economics 85: 275-287 Cooper RG, Edgett SJ (1999) Product Development for the Service Sector. Perseus Books, Cambridge Deutsches Institut für Normung DIN (ed) (1998) Service Engineering. Entwicklungsbegleitende Normung für Dienstleistungen (DIN-Fachbericht 75). Beuth, Berlin Edvardsson B, Gustafsson A, Johnson MD, Sandén B (2000) New Service Development and Innovation in the New Economy. Studentlitteratur, Lund Erlhoff M, Mager B, Manzini E (eds) (1997) Dienstleistung braucht Design. Luchterhand, Neuwied Fähnrich K-P, Meiren T (eds) (2004) Computer Aided Engineering für IT-basierte Dienstleistungen. Leipziger Beiträge zur Informatik, Leipzig Fähnrich K-P, Meiren T, Barth T, Hertweck A, Baumeister M, Demuß L, Gaiser B, Zerr K (1999) Service Engineering. Ergebnisse einer empirischen Studie zum Stand der Dienstleistungsentwicklung in Deutschland. IRB, Stuttgart Fitzsimmons J, Fitzsimmons M (eds) (2000) New Service Development. Sage Publications, London Freitag M (2002) Konzeption von Dienstleistungen. In: Meiren T, Liestmann V (eds) Service Engineering in der Praxis. Kurzstudie zu Dienstleistungsentwicklung in deutschen Unternehmen. IRB, Stuttgart, pp 34-43 Ganz W, Meiren T (eds) (2002) Service research today and tomorrow. Spotlight on international activities. IRB, Stuttgart Herrmann K, Klein R, The T-S (2003) Computer-Aided Service Engineering Tool – Ein Rahmenkonzept für das IT-gestützte Service Engineering. In: Bullinger H-J, Scheer AW (eds) Service Engineering. Entwicklung und Gestaltung innovativer Dienstleistungen. Springer, Berlin Heidelberg New York, pp 647-675 Hofmann H, Klein L, Meiren T (1998) Vorgehensmodelle für das Service Engineering. IM Information Management & Consulting 13: 20-25 Luczak H, Sontow K, Kuster J, Reddemann A, Scherrer U (2000) Service Engineering. Der systematische Weg von der Idee zum Leistungsangebot. TCW, München Meiren T (2004) Service Engineering im Trend. Ergebnisse einer Studie unter technischen Dienstleistern. IRB, Stuttgart Meiren T (1999) Service Engineering. Systematic Development of New Services. In: Werther W, Takala J, Sumanth DJ (eds) Productivity & Quality Management Frontiers. MCB University Press, Bradford, pp 329-343 Meiren T, Barth T (2002) Service Engineering in Unternehmen umsetzen. Leitfaden für die Entwicklung von Dienstleistungen. IRB, Stuttgart Meiren T, Liestmann V (eds) (2002) Service Engineering in der Praxis. Kurzstudie zu Dienstleistungsentwicklung in deutschen Unternehmen. IRB, Stuttgart Ramaswamy R (1996) Design and Management of Service Processes. Addison-Wesley, Reading (Mass.) Sabisch H (2000) Produkte und Produktgestaltung. In: Kern W, Schröder H-H, Weber J (eds): Handwörterbuch der Produktionswirtschaft. Schäffer-Poeschel, Stuttgart, pp 1439-1442 Shostak L (1984) Designing Services that Deliver. Harvard Business Review 62: 73-78

16


Shostak L (1982) How to Design a Service. European Journal of Marketing 16: 49-63 Sontow K (2000) Dienstleistungsplanung in Unternehmen des Maschinen- und Anlagenbaus. Shaker, Aachen Spath D, Zahn E (eds) (2003) Kundenorientierte Dienstleistungsentwicklung in deutschen Unternehmen. Springer, Berlin Heidelberg New York

Service Engineering in Action: The Palm/Erlang-A Queue, with Applications to Call Centers Avishai Mandelbaum, Sergey Zeltyn Faculty of Industrial Engineering and Management, Technion, Haifa, Israel

1 Introduction.......................................................................................................18 2 Significance of abandonment in modelling and practice...............................20 3 Birth-and-death process representation .........................................................22 4 Operational measures of performance............................................................24 4.1 Practical measures: accounting for abandonment....................................24 4.2 Calculations: the 4CallCenters software..................................................24 4.3 A general approach for computing operational performance measures ..26 4.4 Relation between average wait and the fraction abandoning...................26 5 Parameter estimation in a call center environment .......................................27 6 Approximations.................................................................................................29 7 Applications to call centers ..............................................................................32 7.1 Erlang-A performance measures: comparison against real data..............32 7.2 Erlang-A approximations: comparison against real data .........................33 8 Some advanced features of 4CallCenters......................................................334 9 Some open research topics ...............................................................................36 9.1 Cost and revenue analysis .......................................................................36 9.2 Extending Erlang-A: human behavior .....................................................38 9.3 Extending Erlang-A: structure of modern call centers ............................40 9.4 Uncertainty in parameter values ..............................................................40 References.............................................................................................................41 Appendix: The Erlang-A queue: Useful formulae for the steady-state distribution and some performance measures ..........................................44

18

Avishai Mandelbaum, Sergey Zeltyn

1 Introduction Service Engineering is a newly emerging scientific discipline. As we perceive it, it caters to operational service-challenges that arise in our post-industrial society. To this end, researchers in the area develop scientifically-based engineering principles and tools, often culminating in software, which support the design and management of service operations. Moreover, a multi-disciplinary approach is called for in order to balance service quality and efficiency, from the likely conflicting perspectives of customers, service-providers, managers and society. In our research, and clearly biased by our scientific roots, we focus on methodologies from Operations Research (OR) and Statistics. See Pollock et al. [32] for various applications of OR to public services (including an abridged history of OR). An older, but still very relevant, is Larson's [23] excellent collection of OR applications in service industries. Readers are also referred to Fitzsimmons and Fitzsimmons [14] and Lovelock [24] for broad introductions to service-related disciplines, specifically Operations Management, Industrial Engineering, Information Technology, Marketing and Human Resource Management. In our contribution, we are concerned with Contact Centers. These are service organizations for customers who seek service via the phone, fax, e-mail, chat or other tele-communication channels. A particularly important type of contact centers are Call Centers, which predominantly serve phone calls. Due to advances in Information and Communication Technology, the number, size and scope of contact/call centers, as well as the number of people who are employed there or use them as customers, grows explosively. Indeed, in the U.S. alone, the call center industry is estimated to employ several million agents that, in fact, outnumber agriculture. In Europe, the number of call center employees in 1999-2000 was estimated, for example, by 600,000 in the UK (2.3% of the total workforce) and 200,000 in Holland (almost 3%) [5]. Bittner et al. [7] assess that, in Germany in 2001, there were between 300,000 to 400,000 (1-2%) employed in the call center industry. See Gans et al. [16] for a review of state-of-the-art research on telephone call centers. Mandelbaum [25] provides a comprehensive bibliography, namely references plus abstracts, of call-center-related academic papers. Finally, Cleveland and Mayben [12] and Helber and Stolletz [22] are textbooks (the latter in German) on the operations management and design of call centers. In a large performance-leader call center, many hundreds of agents could serve many thousands of calling customers per hour; agents' utilization levels exceed 90%, yet about 50% of the customers are answered immediately upon calling; callers who are delayed demand a response within seconds, the vast majority gets it, and few of the rest, say 1-2% of those calling, abandon during peak-congestion due to impatience. But most call centers are far from achieving such levels of operational performance. To these, scientific models are prerequisites for climbing the performance ladder.

The Palm/Erlang-A Queue, with Applications to Call Centers

lost calls retrials arrivals

retrials

19

agents

busy

1

ACD

1 2 3 queue k

abandonment

2 … n returns

lost calls Fig. 1. Schematic representation of a telephone call center

It will become clear below that call center operations can be naturally viewed as queueing systems. A particular mathematical queueing model, which we refer to as the Palm/Erlang-A model, is an appropriate starting point: it is basic enough to provide insight and be useful, yet it is complex enough to capture some central features in call center operations. The Palm/Erlang-A model is thus the main subject of our paper. Modelling a Call Center. A simplified representation of traffic flows in a call center is given in Fig. 1. Incoming calls form a single queue, waiting for service from one of n statistically identical agents. There are k + n telephone trunk-lines. These are connected to an Automatic Call Distributor (ACD) that manages the queue, connects customers to available agents, and also archives operational data. Customers arriving when all lines are occupied encounter a busy signal. Such customers might try again later (‘retrial’) or give up (‘lost call’). Customers who succeed in getting through at a time when all agents are busy (that is, when there are at least n but fewer than k + n customers within the call center) are placed in the queue. If waiting customers run out of patience before their service begins, they hang up (‘abandon’). After abandoning, customers might try calling again later while others are lost. After service, there are ‘positive’ returns of satisfied customers, or ‘negative’ returns due to complaints. Note that the model in Fig. 1 ignores multiple service types and skilled-based routing that are present in many modern call centers. However, a lot of interesting questions still remain open (see Section 9) even for models with homogeneous servers/customers. In basic models, the already simple representation in Fig. 1 is simplified even further. Specifically, in the present paper we assume that there are enough trunk-lines to avoid busy signals (k = f). This assumption prevails in today's call centers. In addition, we assume out retrials and return calls, which corresponds to absorbing them within the arrivals. (See, for example, Aguir et al. [2] for an analysis that takes retrials into account.) However, and unlike most models used in practice,

20


here we do acknowledge and accommodate abandonment. The reasons for this will become clear momentarily.

2 Significance of abandonment in modelling and practice The classical M/M/n queueing model, also called the Erlang-C model, is the one most frequently used in workforce management of call centers. Erlang-C assumes Poisson arrivals at a constant rate Ȝ, exponentially distributed service times with a rate µ, and n independent statistically-identical agents. (Time-varying arrival rates are accommodated via piecewise constant approximations.) But Erlang-C does not allow abandonment. This, as will now be argued, is a significant deficiency: customer abandonment is not a minor, let alone a negligible, aspect of call center operations. We now support this last statement, first qualitatively and then quantitatively. x Abandonment statistics constitute the only ACD measurement that is customersubjective: those who abandon declare that the service offered is not worth its wait. (Other ACD data, such as average waiting times, are ‘objective’; they also do not include the only other customer-subjective operational measures, namely retrial/return statistics.) x Some call centers focus on the average waits of only those who get served, which does not acknowledge abandoning customers. But under such circumstances, the service-order that optimizes performance is LIFO = Last-In-FirstOut [17], which clearly suggests that a distorted focus has been chosen. x Ignoring abandonment can cause either under- or over-staffing: On the one hand, if service level is measured only for those customers who reach service, the result is unjustly optimistic - the effect of an abandonment is less delay for those further back in line, as well as for future arrivals. This would lead to under-staffing. On the other hand, using workforce management tools that ignore abandonment would result in over-staffing as actually fewer agents are needed in order to meet most abandonment-ignorant service goals. The Palm/Erlang-A model: Palm [31] introduced a simple (tractable) way to model abandonment. He suggested to enrich Erlang-C (M/M/n) in the following manner. Associated with each arriving caller there is an exponentially distributed 1 patience time with mean T . An arriving customer encounters an offered waiting time, which is defined as the time that this customer would have to wait given that her or his patience is infinite. If the offered wait exceeds the customer's patience time, the call is then abandoned, otherwise the customer awaits service. The patience parameter ș will be referred to as the individual abandonment rate. (We shall omit ‘individual’, when obvious.) We denote this model by M/M/n + M, and refer to it as Palm/Erlang-A, or Erlang-A for short. Here the A stands for Abandonment, as well as for the fact that the model interpolates between Erlang-C and


21

Erlang-B. (The latter is the M/M/n/n model, in which there are n trunk lines (k = 0), hence customers that cannot be served immediately are blocked.) Delay probability

Average wait

Fig. 2. Comparison between Erlang-A and Erlang-C 18,42] 48 calls per min., 1 min. average service time, 2 min. average patience

With Erlang-A, the quantitative significance of abandonment can be demonstrated through simple numerical examples. We start with Fig. 2, which shows the fraction of delayed customers and the average wait, when calculated via Erlang-C (M/M/n), and a corresponding Erlang-A (M/M/n + M) model. In both models, the arrival rate is 48 calls per minutes, the average service time equals 1 minute, and the number of agents is varied from 35 to 70. Average patience is taken to be 2 minutes for the Erlang-A model. Clearly, the two curves convey rather different pictures of what is happening in the system they depict, especially within the range of 40 to 50 agents: in particular, and as shown below, Erlang-C is stable only with 49 or more agents, while Erlang-A is always stable. The above M/M/n and M/M/n + M models are further compared in Table 1. Note that exponential patience with an average of 2 minutes gives rise to 3.1% abandonment. Then note that the average wait and queue length are both strikingly shorter with only 3.1% abandonment taking place. Indeed, ‘The fittest survive’ and wait less – much less. (Significantly, this high-level performance is not achieved if the arrival rate to the M/M/n system is decreased by 3.1%; for example, the ‘average speed of answer’ in such a case is 8.8 seconds, compared with 3.7 seconds. The reason is that abandonment reduces workload precisely when needed, namely when congestion is high.) Finally, note that system performance in such heavy traffic is very sensitive to staffing levels. In our example, adding 3 or 4 agents (from 50 to say 54) to M/M/n would result in M/M/n + M performance, as emerging from the horizontal distance between the graphs in Fig. 2. Nonetheless, since personnel costs are the major operational costs of running call centers (prevalent estimates run at about 60-75% of the total), even a 6-8% reduction in personnel is economically significant (and much more so for large call centers that employ thousands of agents).

22


Table 1. Comparing models with/without abandonment 50 agents, 48 calls per min., 1 min. average service time, 2 min. average patience

Fraction abandoning Average waiting time Waiting time's 90th percentile Average queue length Agents' utilization

M/M/n 20.8 sec 58.1 sec 17 96%

M/M/n + M 3.1% 3.7 sec 12.5 sec 3 93%

M/M/n, Ȝ p 3.1% 6.8 sec 28.2 sec 7 93%

As a final demonstration of the significance of abandonment, we now use it to explain a phenomenon that has puzzled queueing theorists: It is the observation that, in practice, simple deterministic approaches often lead to surprisingly good results. For example, consider a call center with averages of 6000 calls per hour and service time of 4 minutes. Such a call center gets an average of (6000: 60) · 4 = 400 minutes of work per minute. The deterministic approach then prescribes 400 service agents to cope with this load (1 agent-minute per 1 work-minute), which is a questionable recommendation according to standard queueing models. For example, Erlang-C would then be unstable, and its waiting times and queue-lengths would increase indefinitely. But now assume that customers abandon, as they actually do, and assign a reasonable parameter to their average patience, say equal to the average service time. Then, under Erlang-A, about 50% of the customers would be answered immediately upon calling, the average wait would be a mere 5 seconds, agents' utilization would be 98%, and all this at the cost of 2% abandonment – a remarkable performance indeed. (See the Remark in Section 6 for a more formal explanation.)

3 Birth-and-death process representation Figure 3 provides a representation of the traffic flows in Erlang-A, and a comparison with Fig. 1 clearly reveals its limitations. (Nevertheless, and as we hope to demonstrate, Erlang-A still turns out to be very useful and insightful, both theoretically and practically.) Erlang-A is characterized by 4 parameters: O , P ,T and n. Here Ȝ is the calling rate (calls per unit of time); µ is the service rate (1/µ is the average duration of service); 1/ș is the average patience of a customer; and n is the number of servers/agents. More formally, in the Erlang-A model customers arrive to the queueing system according to a Poisson (Ȝ) process. Customers are equipped with patience times W that are exp(ș), iid across customers. And service times are iid exp(µ). Finally, the processes of arrivals, patience and service are mutually independent.


23

agents 1

queue

arrivals

2 …

O

abandonment

n P

T

Fig. 3. Schematic representation of the Erlang-A model

For a given customer, the patience time W is the time that the customer is willing to wait for service – a wait that reaches W results in an abandonment. Let V denote the offered waiting time – the time a customer, equipped with infinite patience, must wait in order to get service. The actual waiting/queueing time then equals W = min {V, W} Denote by L(t) the number-in-system at time t (includes both customers being served and waiting in the queue). Then L {L (t ), t t 0} is a Markov birth-anddeath process, with the following transition-rate diagram:

0

O

P

O 1

P

O n-1

2

nP

O n

O

n+1 n+2 nPT nPT

Fig. 4. Transition-rate diagram of the Erlang-A model

An analysis of a birth-and-death process usually starts with verifying that it reaches steady-state (it always does, in our case), and it then continues with calculation of its limiting/steady-state distribution, defined by:

S j lim P{L(t ) t of

Alternatively,

Sj

j},

j

0,1, 2,...

(3.1)

can be characterized as the fraction of time that the system

spends in state j, when in steady-state. Formulae for the steady-state distribution of Erlang-A are presented in the Appendix.

24


4 Operational measures of performance In order to understand and apply the Erlang-A model, one must first define its measures of performance, and then be able to calculate them. Moreover, since a call center can get very large (thousands of agents), the implementation of these calculations must be both fast and numerically stable. 4.1 Practical measures: accounting for abandonment

The most popular measure of operational (positive) performance is the fraction of served customers that have been waiting less than some given time, or formally P{W d T , Sr} , where W is the (random) waiting time in steady-state, {Sr} is the event ‘customer gets service’ and T is a target time that is determined by Management/Marketing. However, as explained before, performance measures must take into account those customers who abandon. Indeed, if forced into choosing a single number as a proxy for operational performance, we recommend the probability to abandon P{Ab}, the fraction of customers who explicitly declare that the service offered is not worth its wait. Some managers actually opt for a refinement that excludes those who abandon within a very short time, formally P{W > İ; Ab}, for some small İ > 0, e.g. İ = 3 seconds. The justification is that those who abandon within 3 seconds cannot be characterized as poorly served. There is also a practical rational that arises from physical limitations, specifically that such ‘immediate’ abandonment could in fact be a malfunction or an inaccuracy of the measurement devices. The single abandonment measure P{Ab} can be in fact refined to account explicitly for those customers who were or were not well-served. To this end, we propose the following four-dimensional service measure, given 2 parameters T and İ: x x x x

P{W d T; Sr} P{W > T; Sr} P{W > İ; Ab} P{W d İ; Ab} -

fraction of well-served; fraction of served, with a potential for improvement; fraction of poorly-served; fraction of those whose service-level is undetermined see the above for an elaboration.

Our proposed 4-component measure is not commonly used and most workforce management software tools are incapable of calculating it. To have it practical, we now describe how it can be implemented via the software tool 4CallCenters [15]. 4.2 Calculations: the 4CallCenters software

Black-box Erlang-A calculations, as well as many other useful features, are provided by the free-to-use software 4CallCenters [15]. (This software is being regularly upgraded.) The calculation methods are described in Appendix B of [18];


25

they were developed in the Technion's M.Sc. thesis of the first author, Ofer Garnett. Figure 5 displays a 4CallCenters output and demonstrates how to calculate the four-dimensional service measure, introduced in Subsection 4.1. The values of the four Erlang-A parameters are displayed in the middle of the upper half of the screen. Let T = 30 seconds and İ = 10 seconds. Then one should perform computations twice: with Target Time 30 and 10 seconds. (Both computations appear in Fig. 5.) We get: x P{W d T; Sr} x P{W > T; Sr} x P{W > İ; Ab} x P{W d İ; Ab} -

fraction of well-served is equal to 71.1%; fraction of served, with a potential for improvement is 16.4% (87.5% –71.1%); fraction of poorly-served is 8.6% (12.5% – 3.9%); fraction of those whose service-level is undetermined is 3.9%.

Note that the 4CallCenters output includes many more performance measures than those displayed in Fig. 5: one could scroll the screen to values of agents' occupancy, average waiting time, average queue length, etc. In Section 8 we describe several examples of the more advanced capabilities of 4CallCenters.

Fig. 5. 4CallCenters. Example of output

26


4.3 A general approach for computing operational performance measures

Some explicit expressions of Erlang-A performance measures are provided in the Appendix. (See also Riordan [33].) However, we recommend to use more general M/M/n + G formulae, as the main alternative to 4CallCenters software. Indeed, Erlang-A is a special case of the M/M/n + G queue, in which patience times are generally distributed. A comprehensive list of M/M/n + G formulae, as well as guidance for their application, appears in Mandelbaum and Zeltyn [28]. The preparation of [28] was triggered by a request from a large U.S. bank. Consequently, this bank has been routinely applying Erlang-A in the workforce management of its 10,000 telephone agents, who handle close to 150 million calls yearly. The handout [28] also explains how to adapt the M/M/n + G formulae to Erlang-A, in which patience is exponentially distributed. Specifically, see Sections 1,2 and 5 of [28]. 4.4 Relation between average wait and the fraction abandoning

A remarkable property of Erlang-A, which in fact generalizes to other models with patience that is exp(ș), is the following linear relation between the fraction abandoning P{Ab} and average wait E[W]: P{Ab} = T E[W ]

(4.1)

Proof:

The proof is based on the balance equation

T E[Q ] O P{Ab} ,

(4.2)

and on Little's formula

E[Q ] O E[W ]

(4.3)

where Q is the steady-state queue length. The balance equation (4.2) is a steady-state equality between the rate that customers abandon the queue (left-hand side) and the rate that abandoning customers (i.e. – customers who eventually abandon) enter the system. Substituting Little's formula (4.3) into (4.2) yields formula (4.1). Figure 6 illustrates the relation (4.1). It was plotted using yearly data of an Israeli bank call center [11], which is analyzed in a Service Engineering course that is taught at the Technion [34, 11]. (See also Brown et al. 10 for statistical analysis of this call center data.) First, P{Ab} and E[W] were computed for the 4158 hour


27

intervals that constitute the year 1999. The left plot of Fig. 6 presents the resulting ‘cloud’ of points, as they scatter on the plane. For the right plot, we are using an aggregation procedure that is designed to emphasize dominating patterns. Specifically, the 4158 intervals were ordered according to their average waiting times, and adjacent groups of 40 points were aggregated (further averaged): this forms the 104 points of the second plot in Fig. 6. (The last point of the aggregated plot is an average of only 38 hour intervals.)

Fig. 6. Probability to abandon vs. average waiting time [10]

We observe a convincing linear relation (line) between P{Ab} and E[W]. Based on (4.1) and Fig. 6, the slope of this line is an estimate of the average patience, which here equals 446 seconds. In Brown et al. [10] and Mandelbaum et al. [30] it is shown that some M/M/n +M assumptions do not prevail for the data [11]. Although arrivals are essentially Poisson, the service times are not exponential (in fact, they are very close to being lognormal). Patience times were shown to be non-exponential either. Yet, Erlang-A is proved useful for the performance analysis, which we demonstrate further in Section 7. It is therefore important to understand the circumstances under which one can practically use simple relations that, theoretically, apply perhaps only to models with exponential patience. A recent paper of Mandelbaum and Zeltyn [27] addresses this question for (4.1), demonstrating that the linear relation is practically rather robust. See also [43] where we demonstrate a similar linear relation on another data set.

5 Parameter estimation in a call center environment In order to apply Erlang-A, it is necessary to input values for its four parameters: O , P ,T and n. Typical applications use estimates, which are based on historical ACD data, and here we briefly outline procedures for their estimation/prediction. For a more detailed exposition, including some subtleties that occur in practice, readers are referred to [10].

28


Arrivals: Arrivals of incoming calls are typically assumed Poisson, with timevarying arrival rates. The goal is to estimate/predict these arrival rates, over short time-intervals (15, 30 minutes or one hour), chosen so that the rates are approximately constant during an interval. Then the time-homogeneous model is applied separately over each such interval. The goal can be achieved in two stages. First, time-series algorithms are used to predict daily volumes, taking into account trends and special days (eg. holidays, ‘Mondays’, special sales). Second, one uses (non)parametric regression techniques for predicting the fraction of arrivals per time-interval, out of the daily-total. This fraction, combined with the daily total, yields actual arrival rates per each timeinterval. (See Section 4 in [10] for a detailed treatment.) Services: Service durations are assumed exponential. Average service times tend to be relatively stable from day to day and from hour to hour. (However, they often change depending on the time-of-day! See [10].) In practice, service consists of several phases: mainly talk time, wrap-up time (after-call work), and what is sometimes referred to as auxiliary time. An easierto-grasp notion is thus ‘idle-time’, namely the time that an agent is immediately accessible for service. It is thus also possible to estimate the average service time during a time interval by:

Total Working Time - Total Idle Time , Number of Served Customers where Total Working Time is the product of the Number of Agents by the Interval Duration. (See Adler et al. [1] for an application of this approach, in the context of product development.) Number of agents: In performance analysis, the number of agents n is an Erlang-A input. In staffing decisions, n is typically an output. In both cases, n is in fact the needed number of FTE's (Full Time Equivalent positions), and hence it must be normalized by the rostered staff factor (RSF), or shrinkage factor, which accounts for absenteeism, unscheduled breaks, etc. (See Cleveland and Mayben [12].) For example, if 100 agents are required for answering calls, in fact more agents (105, 110,}) should be assigned to shift, depending on RSF. Patience: (Im)patience time is assumed exponential, say exp(ș). One must then estimate the individual abandonment rate ș, or equivalently, the average patience (1/ș). A difficulty arises from the fact that direct observations are censored – indeed, one can only measure the patience of customers who abandon the system before their service began. For the customers receiving service, their waiting time in queue is only a lower bound for their patience. There are statistical methods for ‘un-censoring’ data; see [10]. Another, more basic problem for estimating ș, is that most ACD data contain only averages, as opposed to call-by-call statistics that are required by the available ‘uncensoring’ methods. To this end, we suggest here two methods for estimating average patience. The first is based on the relation (4.1) between the probability to abandon and average wait. The average wait in queue, E[W], and the fraction of customers abandoning, P{Ab}, are in fact stan-


29

dard ACD data outputs, thus, providing the means for estimating ș as follows:

Tˆ

P{Ab} E[W ]

% Abandonment . Average wait

A second more general approach is to calculate some performance measure (see Section 4) and compare the result to the value derived from ACD data. (This approach is applied in [10].) The goal is to calibrate the patience parameter until these estimates closely match. One advantage of this method is the flexibility in choosing the performance measure being matched, which might depend on the given ACD data. Furthermore, this calibration represents a form of validation of the model's assumptions, and can compensate for discrepancies.

6 Approximations Although exact formulae for the Erlang-A system are available and can be incorporated in software (see Sections 3 and 4) they are too complicated for providing guidelines and insights to call center researchers and managers. Consequently, some useful and insightful approximations have been developed, which we now describe. It has been found useful to distinguish three operational regimes, as in Garnett et al. [18] and Zeltyn and Mandelbaum [43]. Each regime represents a different philosophy for operating a call center. One regime is Efficiency-Driven (ED), another is service Quality-Driven (QD), and the third one rationalizes efficiency and quality, namely it is Quality and Efficiency-Driven (QED). We are interested mainly in not-too-small call centers. Hence, we think of the service and abandonment rates, µ and ș, as fixed, and the arrival rate Ȝ is large enough (formally, it increases indefinitely). Actually, the regimes are determined by the offered load parameter R, which is defined as

R

O ; R represents the amount of work, measured in time-units of P

service, that arrives to the system per unit of time. ( R

O P1

is thus a more tell-

ing representation.) R is also the staffing level n that would be prescribed by the deterministic approach (see the end of Section 0), which ignores stochastic variability. An emphasis of efficiency (service quality) would conceivably lead to n < R (n > R); the deviation of n from R then increases with the intensity of the emphasis. We now proceed with a formal description of the three operational regimes. QED (Quality and Efficiency-Driven) regime

n| RE R ,

f E f;

(6.1)

30


ȕ is a service-grade parameter – the larger it is, the better is the service-level. The staffing regime, described by (6.1), is governed by the so-called Square Root Rule. This rule was already described by Erlang [13], as early as 1924. He reported that it has been in use at the Copenhagen Telephone Company since 1913. A formal QED analysis for the Erlang-C queue appeared only in 1981, in the seminal paper of Halfin and Whitt [19]. (The service grade ȕ must be positive in this case.) Garnett et al. [18] explored Erlang-A in the QED regime and Zeltyn and Mandelbaum [43] treated the M/M/n + G queue with a general patience distribution. (With abandonment, ȕ can be also 0 or negative.) In the QED regime, the delay probability P{W > 0} converges to a constant that is a function of the service grade ȕ and the ratio P / T (or 1/ T , which is average 1/ P patience that is measured in units of average service time). The probability to abandon and average wait vanish, as O , n n f , at rate 1 . Formulae for differn ent performance measures can be found in [18] or [28]. It is significant and useful to mention that the QED approximations are often valid over a wide range of parameters, from the very large call centers (1000's of agents) to the moderate-size ones (10's of agents).

Fig. 7. Asymptotic relations between service grade and delay probability

Figure 7 illustrates the dependence between ȕ and P{W > 0}, for varying values of the ratio P / T . In addition, we plotted the curve for the Erlang-C queue, which is meaningful for positive ȕ only. Note that for large values of P / T (very patient customers) the Erlang-A curves get close to the Erlang-C curve.


31

Remark. When ȕ = 0 in (6.1), the staffing level corresponds to the simple rule that does not take into account stochastic considerations: assign the number of agents equal to the offered load O / P . In Erlang-C, this ‘naive’ approach would lead to system instability. However, in Erlang-A (which is a much better fit to the real world of call centers than Erlang-C) one would get a reasonable-to-good performance level. For example, if the service rate µ is equal to the individual abandonment rate ș and ȕ = 0 then 50% of customers would get service immediately upon arrival. (Check it in Fig. 7. Note that for Erlang-C, 50% delay probability corresponds to ȕ = 0.5.) This suggests why some call centers that are managed using simplified deterministic models, actually perform at reasonable service levels. (One obtains the ‘right answer’ from the ‘wrong reasons’.) The QED regime enables one to combine high levels of efficiency (agents' utilization close to 100%) and service quality (agents' accessibility). The scatterplots in Fig. 8 illustrate this point. The plots display data from ACD reports of two call centers: Italian and American, collected in half-hour intervals during a single working day. The service grade ȕ is calculated via

E

nR . R

We observe moderate-to-small values of abandonment for the service grade

1 d E d 2 . Plots of average waiting time exhibit a similar behavior – see [29]. U.S. data

Italian data

Fig. 8. Service grade for call centers - correlation with abandonment

ED (Efficiency-Driven) regime

n | R (1 J ),

J ! 0.

(6.2)

In this case, virtually all customers wait, the probability to abandon converges to J and average wait is close to J / T . (See Mandelbaum and Zeltyn [28] for additional performance measures.) This regime could be used if efficiency considerations are of main significance. Indeed, it has gained importance in recent re-

32


search (see, for example, few papers of Whitt [37, 41]), following the observation that ED could yield performance that is acceptable for many call centers, for example those operating in not-for-profit environments. QD (Quality-Driven) regime

n | R (1 J ),

J ! 0.

(6.3)

The staffing regime (6.3) should be implemented if quality considerations far dominate efficiency considerations (e.g. high-valued customers or emergency phones). Major performance measures (delay probability, fraction abandoning, average wait) vanish here at an exponential rate in n. Remark. Above, we considered steady-state performance measures. Processlimit results for the number-in-system process L = {L(t), t t 0} are available for the QED and ED regimes in Garnett et al. [18] and Whitt [37], respectively.

7 Applications to call centers 7.1 Erlang-A performance measures: comparison against real data

We now validate the Erlang-A model against the hourly data for the Israeli bank call center, already used for the example in Section 4. Three performance measures are considered: probability to abandon, average waiting time and delay probability. Their values are calculated for the hourly intervals using exact Erlang-A formulae. Then the results are aggregated along the same method employed in Fig. 6. The resulting 86 points are compared against the line y = x: the better the fit the better Erlang-A describes reality. Computation of the Erlang-A parameters. Parameters Ȝ and µ are calculated for every hourly interval. We also calculate each hour's average number of agents n. Because the resulting n's need not be integral, we apply a continuous extrapolation of the Erlang-A formulae, obtained from relationships developed in [31]. Finally, for ș we use formula (4.1). The results are displayed in Fig. 9. The figure's two left-hand graphs exhibit a relatively small yet consistent overestimation with respect to empirical values, for moderately and highly loaded hours. The right-hand graph shows a very good fit everywhere, except for very lightly and very heavily loaded hours. The underestimation for small values of P{W > 0} can be probably attributed to violations of work conservation (idle agents do not always answer a call immediately). Summarizing, it seems that these Erlang-A estimates can be used as close upper bounds for the main performance characteristics of our call center.


33

Fig. 9. Erlang-A formulae vs. data averages [10]

7.2 Erlang-A approximations: comparison against real data

In Section 6 we discussed approximations of various performance measures for the Erlang-A (M/M/n + M) model. Such approximations require significantly less computational effort than exact Erlang-A formulae. Figure 10, based on the same data as Fig. 9, demonstrates a good fit between data averages and the approximations. In fact, the fits for the probability of abandonment and average waiting time are somewhat superior to those in Fig. 9 (the approximations provide somewhat larger values than the exact formulae). This phenomenon suggests two interrelated research questions of interest: explaining the overestimation in Fig. 9 and better understanding the relationship between Erlang-A formulae and their approximations.

Fig. 10. Erlang-A approximations vs. data averages [10]

The empirical fit of the simple Erlang-A model and its approximation turns out to be very (perhaps surprisingly) accurate. Thus, for the call center in considera-

34


tion – and those like it – use of Erlang-A for workforce management could and should improve operational performance.

8 Some advanced features of 4CallCenters The 4CallCenters software [10] provides a valuable tool for implementing ErlangA calculations. Its basic feature is ‘Performance Profiler’ that enables calculation of all the useful performance measures, given the four Erlang-A parameters as input. In addition, 4CallCenters allows many advanced options: staffing queries, graphs, export and import of data and more. Here we demonstrate, as an example, two advanced capabilities of 4CallCenters. Example 1: Advanced profiling: One can vary any input parameters of the Erlang-A queue and display the corresponding model output (performance measures) either in a table or graphically. For example, let the average service time equal 2 minutes and average patience 3 minutes. Let the arrival rate vary from 40 to 230 calls per hour, in steps of 10, and the number of agents from 2 to 12. Then one can immediately produce a table that contains values of different performance measures for all combinations of the two input parameters. Probability to abandon

Average wait

Fig. 11. 4CallCenters. Advanced profiling

Figure 11 shows the dependence of the probability to abandon and average wait on different number of agents. Note that the two plots look identical: the reason is relation (4.1). In addition, the red curves on both plots in Fig. 11 illustrate Economies of Scale (EOS): while offered load per server remains constant along this


35

O 2 , performance significantly improves as the number of agents innP 3 creases. For example, the probability to abandon is equal to 13.7% for n = 2, 5.1% for n = 5 and 1.5% for n = 12. Finally, note that both P{Ab} and E[W] actually vanish as n gets large. curve

Example 2: Advanced staffing queries: 4CallCenters enables staffing queries with several performance goals. For example, assume that the average service time is equal to 4 minutes, and average patience is 5 minutes. Our goal is to calculate appropriate staffing levels for arrival-rate values that vary from 100 to 1200, in steps of 50. The performance targets are:

x Probability to abandon less than 3%; x 80% of customers served within 20 seconds.

Fig. 12 presents the screen output of 4CallCenters.

Fig. 12. 4Callcenters. Advanced staffing queries

The first plot of Fig. 13 displays the minimal staffing level that adheres to both goals. The EOS phenomenon is observed here as well: 10 agents are needed for 100 calls per hour but only 83 (rather than 10 12 120 ) for 1200 calls per hour. (Despite its look, the curve in the first plot is not a straight line.) The second plot displays the values of the two target performance measures. (This plot, unlike the

36


first one, is not an immediate output graph of 4CallCenters but rather an edited version of it.) Remark. Since the number of agents must be an integer, we observe performance ‘zigzags’ in the right plot of Fig. 13.

Fig. 13. 4CallCenters. Dynamics of staffing level and performance

9 Some open research topics Here we discuss some important extensions of the models and methods presented above. We elaborate on several topics that are directly relevant to our current research, while others are only briefly reviewed. See Gans et al. [16] for an extensive review on research prospects in the area of call centers. 9.1 Cost and revenue analysis

One can search for an optimal staffing level, given the trade-off between staffing cost, cost of customers' waiting and cost of abandonment. Borst et al. [8] referred to this problem as dimensioning and solved it for the Erlang-C queue (no abandonment): if the staffing cost and the cost of waiting are comparable, the optimal staffing should take place in the QED regime, described in Section 6 (with positive ȕ in formula (6.1)). Ongoing research by Borst et al. [9] is dedicated to the same question for the Erlang-A queue. For example, let the average operational cost (per unit of time) be equal to


37

U (n, O ) c n O a P{Ab} , where c is the staffing cost, and a is the abandonment cost. Our goal is to minimize cost. (Note that this is in fact mathematically equivalent to maximizing revenues.) Define the abandonment/staffing cost ratio by r a / c , and let

s P /T .

Assume that a ! c / P . (Otherwise, the asymptotic optimal policy is n* = 0: not to provide service at all.) Then we suggest that the asymptotic optimal staffing level is equal to

n*

ª R y* ( r ; s ) R º , ¬ ¼

(9.1)

where the square brackets in (9.1) denote the nearest integer value and the func* tion y () is defined by

and h() I () /(1 ) ()) is the hazard rate of the standard normal distribution ( I () is its density function and ) () is the cumulative distribution function). As in [8], Fig. 14 compares the rule (9.1) with the exact optimal staffing values. We consider five exponential patience distributions with different means and perform comparisons by varying the value of the ratio r. A perfect fit is observed for all the special cases! Arrival rate Ȝ = 100, service rate µ = 1. Small patience means

Large patience means

Fig. 14. Cost optimization. Approximation vs. exact optimum

38


Numerical experiments for other cost optimization problems (e.g. with waiting cost, instead of abandonment cost) also demonstrate a very close correspondence between exact values and the corresponding analogs of (9.1). Hence the goal is to develop a theoretical framework, parallel to [8], that will support our experimental research. In addition, we are working on a constraint satisfaction version where one chooses the least number of agents that adheres to a given constraint on the waiting and/or the abandonment cost. This latter formulation is in fact closer to the way that managers perceive their staffing problems in practice. A more general analysis of revenues and costs should combine call-by-call data (for example, waiting and service times) and Customer Relationship Management data, where customers’ transactions are recorded (e.g. purchasing information). Hopefully, this kind of research will provide insight on the interaction of service experience and marketing behavior of customers. 9.2 Extending Erlang-A: human behavior Non-exponential patience. The Erlang-A model assumes exponential iid patience times that do not depend on the state of the system, time-of-day, etc. In practice, these assumptions are not always valid. In Fig. 15, we display estimates of the hazard rates of the customers' patience for two banks: a large U.S. bank and a small Israeli one. In the two cases we observe different, but clearly non-exponential patterns. (Recall that the hazard rate of an exponential random variable is a constant.) American customers are very impatient at the beginning of their wait, but their patience stabilizes after approximately 10 seconds. In contrast, Israeli customers have two clear peaks of abandonment: approximately at 15 and at 60 seconds. (It turns out that these two surges of abandonment take place immediately after two recorded messages to which customers are exposed: the first one when they enter the queue and the second after approximately one minute.)

American bank

Fig. 15. Bank data: hazard rates of patience times

Israeli bank


39

Therefore, at least in some applications customers' patience times are nonexponential and applicability of the Erlang-A formulae to such systems should be studied. (Recall Section 7.) Zeltyn and Mandelbaum [43] study the M/M/n+G model with generally distributed patience times, providing both a convenient framework for exact calculations and approximations for the three operational regimes defined in Section 6. Whitt [40] provides approximations for the M/G/n+G queue with general service and patience distributions. (Exact formulae are not available for this model.) Patience index. In search for a better understanding of customers' (im)patience, we have found a relative definition to be of use. Specifically, we define the patience index to be

Theoretical Patience Index

time a customer is willing to wait time a customer is required to wait average patience E[W ] . average offered wait E[V ]

While this patience index makes sense intuitively, its calculation requires the application of survival analysis techniques to call-by-call data. Such data may not be available in certain circumstances. Therefore, we wish to find an empirical index that will work as an auxiliary measure for the patience index. We found the following to be a very useful definition:

Empirical Patience Index

% served . % abandoned

The empirical index is easily calculable since both the numbers of served and of abandoned calls are very easy to obtain from call-center reports.

Fig. 16. Patience index – empirical vs. theoretical [10]

40


Figure 16 demonstrates how well the empirical patience index estimates the theoretical patience index for the Israeli bank data [10]. (Aggregated data of 68 quarter hours between 7 a.m. and midnight is used.) Under certain circumstances, one can explain the closeness of the theoretical and empirical indices [10]. However, these explanations are unsatisfactory and hence leave open this research direction. Adaptive behavior. In Zohar at al. [44] and Mandelbaum and Shimkin [26, 35], the authors analyze models for an adaptive behavior of tele-customers, which ‘tune’ their patience according to anticipated or perceived systems congestion. Data from the call center of our Israeli bank supports the applicability of these models, but more is to be done in this direction. 9.3 Extending Erlang-A: structure of modern call centers

As we mentioned in Section 1, the Erlang-A model ignores some important features of modern call centers. Below we discuss the following two central features. IVR and speech recognition. Many large call centers use Interactive Voice Response (IVR) units, which either provide full service to customers or perform its preliminary stage. Current development of speech recognition technologies increases the range of IVR-supported self-service, motivating the study of related models. See, for example, Srivasanan et al. 36] that analyze a two-station network model: IVR followed by an agent. Mandelbaum et al. [30] provides a description of IVR service-time distributions in a banking call center. However, research in this direction is still scarce and should be pursued. Skills-based routing (SBR) technology allows to distinguish between many types of calls and many skills of agents. It turns out that the asymptotic operational regimes, introduced in Section 6, can be analyzed in some SBR models. For example, Armony and Mandelbaum [3] and Armony et al. [4] explore two classical and basic SBR models in the QED regime. Bassamboo et al. [6] and Harrison and Zeevi [20] use ED-related approximations, which enable the analysis of very general models. See Gans et al. [16] for a detailed review on SBR technologies and related research. 9.4 Uncertainty in parameter values

In the real world, one hardly knows the exact values of the four Erlang-A parameters. Therefore, it is essential to study the sensitivity of the model's performance measures. In a recent paper [39], Whitt calculates elasticities in Erlang-A, which measure the percentage change of a performance measure caused by a small percentage change in a parameter. Both exact numerical algorithm and several types of approximations are used. It turns out that Erlang-A performance is quite sensitive to small changes in the arrival rate, service rate, or number of agents, but relatively insensitive to small changes in the abandonment rate.


41

In staffing planning, of which the number of agents is the output, it is reasonable to assume knowledge of the service rate. Thus, the problem of uncertainty in the arrival rate surfaces as the most significant. In Brown et al. [10] it was demonstrated that the Poisson arrival rate in an Israeli call center varies from day to day and its prediction raised statistical and practical challenges. This motivates the study of queueing models, in which the Poisson arrival rate / (the arrival-rate function) is a random variable (random process). If E( / ) o f and its standard deviation is of the order

E(/ ) , we expect

that the QED operational regime and the square-root staffing rule will play a role that is similar to the one with known (deterministic) arrival rate; the offered load R in (6.1) will be replaced by the average offered load E( / ) / P , and uncertainty will manifest itself through a different value of the service grade ȕ. However, if V (/ ) is of the order E (/ ) , the ‘cruder’ ED regime seems to be the most appropriate; see Whitt [0], and Bassamboo et al. [6].

References 1.

2.

3.

4.

5.

5. 7.

8.

Adler PS, Mandelbaum A, Nguyen V, Schwerer E (1995) From project to process management: An empirically-based framework for analyzing product development time. Management Science 41: 458–484 Aguir MS, Karaesmen F, Aksin OZ, Chauvet F (2004) The impact of retrials on call center performance. OR Spectrum, Special Issue on Call Center Management 26 (3): 353-376 Armony M, Mandelbaum A (2004) Design, staffing and control of large service systems: The case of a single customer class and multiple server types. Working paper, available at http://iew3.technion.ac.il/serveng/References/references.html Armony M, Gurvich I, Mandelbaum A (2004) Staffing and control of large-scale service systems with multiple customer classes and fully flexible servers. Working paper, available at http://iew3.technion.ac.il/serveng/References/references.html Bain P, Taylor P (2002) Consolidation, ‘Cowboys’ and the developing employment relationship in British, Dutch and US call centres. In: Holtgrewe U, Kerst C, Shire K (eds.), Re-Organising Service Work. Ashgate Publishing Limited, pp 42-62 Bassamboo A, Harrison JM, Zeevi A (2004) Design and control of a large call center: Asymptotic analysis of an LP-based method. Submitted for publication Bittner S, Schietinger M, Schroth J, Weinkopf C (2002) Call centres in Germany: Employment, training and job design. In: Holtgrewe U, Kerst C, Shire K (eds), ReOrganising Service Work. Ashgate Publishing Limited, pp 63-85 Borst S, Mandelbaum A, Reiman M (2004) Dimensioning large call centers. Operations Research 52 (1): 17-34

42 9. 10.

11. 12. 13.

14. 15. 16.

17.

18. 19. 20. 21. 22. 23.

24. 25. 26. 27.

28. 29.

Avishai Mandelbaum, Sergey Zeltyn Borst S, Mandelbaum A, Reiman M, Zeltyn S (2004) Dimensioning call centers with abandonment, in preparation Brown LD, Gans N, Mandelbaum A, Sakov A, Shen H, Zeltyn S, Zhao L (2002) Statistical analysis of a telephone call center: A queueing science perspective. Journal of the American Statistical Association (JASA) 100 (469): 36-50 Call Center Data (2002) Technion, Israel Institute of Technology, available at http://iew3.technion.ac.il/serveng/callcenterdata/index.html Cleveland B, Mayben J (1997) Call Center Management on Fast Forward. Call Center Press, Annapolis Erlang AK (1948) On the rational determination of the number of circuits. In Brockmeyer E, Halstrom HL, Jensen A (eds), The life and Works of A.K.Erlang. The Copenhagen Telephone Company, Copenhagen Fitzsimmons JA, Fitzsimmons MJ (2004) Service Management: Operations, Strategy and Information Technology. 4th edn., McGraw-Hill/Irwin 4CallCenters Software (2002), available at http://iew3.technion.ac.il/serveng/4CallCenters/Downloads.html Gans N, Koole G, Mandelbaum A (2003) Telephone call centers: a tutorial and literature review. Invited review paper, Manufacturing and Service Operations Management 5 (2): 79-141 Garnett O, Mandelbaum A (2000) An Introduction to Skills-Based Routing and its Operational Complexities. Teaching note, Technion, Israel, available at http://iew3.technion.ac.il/serveng2004/Lectures/SBR.pdf Garnett O, Mandelbaum A, Reiman M (2002) Designing a telephone call-center with impatient customers. Manufacturing and Service Operations Management 4: 208-227 Halfin S, Whitt W (1981) Heavy-traffic limits for queues with many exponential servers. Operations Research 29: 567-588 Harrison JM, Zeevi A (2004) A method for staffing large call centers using stochastic fluid models. To appear in Manufacturing and Service Operations Management Helber S, Mandelbaum A (2004) GIF Research Proposal Helber S, Stolletz R (2004) Call Center Management in der Praxis. Springer, Berlin Heidelberg New York, in German Larson RC (1988) Operations research and service industries. In: Guiles BR, Quinn JB (eds), Managing innovation: cases from the services industries. National Academy Press, Washington D.C., pp 115-143 Lovelock CG (1992) Managing Services: Marketing, Operations and Human Resources, Prentice-Hall Mandelbaum A (2003) Call Centers. Research Bibliography with Abstracts. Version 5, available at http://iew3.technion.ac.il/serveng/References/references.html Mandelbaum A, Shimkin N (2000) A model for rational abandonment from invisible queues. Queueing Systems: Theory and Applications (QUESTA) 36: 141-173 Mandelbaum A, Zeltyn S (2004) The impact of customers’ patience on delay and abandonment: Some empirically-driven experiments with the M/M/n + G queue. OR Spectrum, Special Issue on Call Center Management 26 (3): 377-411 Mandelbaum A, Zeltyn S (2004) M/M/n + G queue. Summary of performance measures, available at http://iew3.technion.ac.il/serveng/References/references.html Mandelbaum A, Zeltyn S (2004) The Palm/Erlang-A Queue, with Applications to Call Centers. Teaching note to Service Engineering course, available at http://iew3.technion.ac.il/serveng/References/references.html


43

30. Mandelbaum A, Sakov A, Zeltyn S (2001) Empirical analysis of a call center. Technical report, Technion, available at http://iew3.technion.ac.il/serveng/References/references.html. 31. Palm C (1957) Research on telephone traffic carried by full availability groups. Tele, Vol. 1, 107 pp (English translation of results first published in 1946 in Swedish in the same journal, which was then entitled Tekniska Meddelanden fran Kungl. Telegrafstyrelsen.) 32. Pollock SM, Rothkopf MH, Barnett A (eds) (1994) Operations Research and the Public Sector. Handbooks in Operations Research and Management Science, Vol.6, NorthHolland 33. Riordan J (1962) Stochastic Service Systems, Wiley 34. ‘Service Engineering’ course website, Technion, available at http://iew3.technion.ac.il/serveng 35. Shimkin N, Mandelbaum A (2004) Rational abandonment from tele-queues: non-linear waiting costs with heterogeneous preferences, Queueing Systems: Theory and Applications (QUESTA) 47: 117-146 36. Srinivasan R, Talim J, Wang J (2004) Performance analysis of a call center with interactive voice response units. Sociedad de Estadistica e Investigacion Operativa Top 12 (1): 91-110 37. Whitt W (2004) Fluid models for many-server queues with abandonments. To appear in Operations Research 38. Whitt W (2004) Staffing a call center with uncertain arrival rate and absenteeism. Submitted to Management Science 39. Whitt W (2004) Sensitivity of performance in the Erlang A model to changes in the model parameters. Submitted to Operations Research 40. Whitt W (2005) Engineering solution of a basic call-center model. Management Science 51 (2): 221-235 41. Whitt W (2005) Two fluid approximations for multi-server queues with abandonments. Operations Research Letters 33: 363-372 42. Zeltyn S (2004) Call centers with impatient customers: Exact analysis and many-server asymptotics of the M/M/n + G queue, Ph.D. thesis, Technion, available at http://iew3.technion.ac.il/serveng/References/references.html 43. Zeltyn S, Mandelbaum A (2004) Call centers with impatient customers: many-server asymptotics of the M/M/n + G queue. Submitted to QUESTA, available at http://iew3.technion.ac.il/serveng/References/references.html 44. Zohar E, Mandelbaum A, Shimkin N (2002).Adaptive behavior of impatient customers in tele-queues: Theory and empirical support. Management Science 48: 566-583

44


The Erlang-A queue: Useful formulae for the steady-state distribution and some performance measures Steady-state distribution. Palm [31] derived the following representation for the steady-state distribution, defined in (3.1):

Sj

n! S n , j ! ( O / P )n j ° ° ® (O / T ) j n , °S n j n nP k 1 T k °¯

0 d j d n, j t n 1

where

Sn

E1, n

nP O 1 ª A , 1º E1, n »¼ ¬« T T

A( x, y )

xe y yx

,

J ( x, y ),

and y

J ( x, y ) ³ t x 1et dt ,

x ! 0, y t 0

0

is the incomplete Gamma function; E1,n denotes the blocking probability in the M/M/n/n (Erlang-B) system: (O / P ) n n! j n (O / P ) ¦ j 0 j!

E1, n

Remark. A simple way for calculating E1,n is the recursion

E1,0 in which

U

0;

E1, n

U E1, n 1 , 1 U E1, n 1

is the offered load per agent, namely

U

O . nP

n t 1,


45

Performance measures. As discussed above, 4CallCenters [15] provides a convenient tool for Erlang-A calculations. Mandelbaum and Zeltyn [28] present a theoretical framework for computations in the more general M/M/n + G system and explain how to adapt it to Erlang-A. Both approaches can be used for calculation of the practical measures from Subsection 4.1. Below we present several explicit expressions. Average wait and the probability to abandon are the most widely used performance measures in practice. To these we add the delay probability, which is important in view of the fact that it characterizes the operational regime (ED, QD or QED) – recall Section 6. Delay probability. Following Palm [31],

P{W ! 0}

1 A

nP

1 E

A T , TO E1, n nP O ,

T

T

(A.1)

1, n

Probability to abandon. The probability to abandon of delayed customers is equal to

P[ Ab | W ! 0]

The

fraction

abandoning, P[ Ab | W ! 0] u P[W ! 0} .

1

1

nP O , UA

P{Ab},

is

T

T

1

U

(A.2)

.

simply

the

product

Average waiting time. Average waiting time of delayed customers is computed via (A.2) and (4.1):

ª º 1 « 1 1» E[W | W ! 0] 1 » . T « U A nP , O U « » T T ¬ ¼

(A.3)

The unconditional average wait E[W] equals the product of (A.1) with (A.3).

Architecture for Service Engineering – The Design and Development of Industrial Service Work Holger Luczak, Christian Gill and Bernhard Sander Forschungsinstitut für Rationalisierung (FIR), RWTH Aachen, Germany (Research Institute for Operations Management (FIR), Aachen University of Technology)

1 Introduction.......................................................................................................48 2 Service Engineering as a New Research Discipline for Successful Service Design and Service Development ...............................................................48 3 Service Planning for Promising Service Ideas................................................50 4 An Architecture for Service Conception.........................................................51 4.1 The Service Result Branch of the Architecture .......................................53 4.2 The Service Process Branch of the Architecture .....................................55 4.3 The Service Skills and Resources Branch of the Architecture ................57 5 Application of the Architecture – Lessons Learned.......................................59 6 Outlook ..............................................................................................................60 Acknowledgements ..............................................................................................61 References.............................................................................................................62

48

Holger Luczak, Christian Gill, Bernhard Sander

1 Introduction As indices of ‘tertiarization’ not only for Germany but also on a worldwide basis show, the service sector has become increasingly important compared to the agriculture and physical goods production sectors (European Commission 2002). Service companies contribute about 70% of the total GDP of Western Europe’s economies. Despite these figures of GDP share and the strategic intention to successfully run the service business, knowledge on how to develop service products and who should design them is marginal. This is in strong contrast to the profound knowledge and skills in the engineering of physical goods. Although some methods for designing and developing physical goods can also be used for services, recent surveys show a fundamental lack of knowledge and specialized skills in the use of procedures, methods and tools to design and develop professional services. This shortage of adequate procedures, methods and tools is mainly based on the immanent characteristics of services that strongly differ from those of products (Luczak et al. 2000). In consequence of the lack of supporting methods and procedures, service operations frequently deliver ineffective and low quality services that do not meet the customer’s needs. Furthermore, service processes, which have been inadequately planned and developed, are hard to reproduce with constant quality, requiring strong improvisation efforts by the service personnel. Combined with an insufficiently designed and developed working environment, profitability is negatively affected since only motivated personnel can deliver high quality services that satisfy the customers in the long term (Heskett et al. 1997; Liestmann and Meiren 2002). Based on this balance of arguments this paper presents an architecture that comprises steps to be taken to successfully design and develop professional services.

2 Service Engineering as a New Research Discipline for Successful Service Design and Service Development A new research discipline moving from scientific discourse into business is called ‘Service Engineering’. Service Engineering mainly deals with the following aspects (Luczak et al. 2002): 1. Improving the procedures for designing and developing services more professionally 2. Establishing service design and development as a corporate function 3. Adapting a service specific human resource management The term ‘Service Engineering’ implies a basis of engineering knowledge and originates from the assumption that services can be designed and redeveloped in a

The Design and Development of Industrial Service Work

49

similar way to physical products (Bitran and Pedrosa 1998; Meyer and DeTore 1999; Luczak et al. 2000). Accordingly, engineering procedures, methods and tools form the core of this approach (VDI 1980, 1993). Service engineering encompasses what Ramaswamy (1996) established as ‘Service Design’, stretching its focus with regard to the extent of the innovation process and the addressed aspects as listed above. This approach differs to some extent from the proximate research field of New Service Development (NSD), which also deals with the issues of how to develop new service products (Easingwood 1986; Bowers 1989; Scheuing and Johnson 1989; Edvardsson and Olsson 1996; Sundbo 1997; Edvardsson et al. 2000). NSD has its roots in service quality, because quality is said to strongly contribute to the understanding of the service logic and the drivers of customer satisfaction (Parasuraman et al. 1988). In contrast to Service Engineering, NSD mainly addresses consumer services rather than business-to-business (B2B) services and approaches the issues of service innovation from a marketing perspective (Johne and Storey 1998). Service Engineering Organization

Develop. & Design Results

Design & Development Develop. & Design Processes

Develop. & Design Methods / Tools

Service Planning Capability Analysis

Market Analysis

Idea Finding

Idea Idea Description Assessment

Service Conception Functional Concept Human Resource Management

Human Resource Concept

Marketing Concept

Sales Concept

Implementation Planning Detailed ER Plan

Resource Setup

Pilot Implementation

Fig. 1. Aspects of Service Engineering and phases of the design and development of services

Within Service Engineering the process of service design and development consists of three major phases, namely the service planning, service conception and service implementation, as shown in Fig. 1 (Luczak et al. 2003). The first phase, service planning, is centered on idea generation, forming and evaluation. The subsequent phase, service conception, leads to a more precise description of

50


these ideas in terms of their content, eventually resulting in a service that is ready to be launched. The launch itself takes place in the implementation phase. This paper will focus on service planning and service conception and thus contributes to the first aspect within the framework of the discipline of Service Engineering as mentioned above.

3 Service Planning for Promising Service Ideas The phase of service planning starts with a systematic idea generation. The use of the contradictory expressions ‘systematic generation’ and ‘idea’ might cause confusion. Obviously, the search for ideas can only be supported by systematic approaches, however it cannot be systematized in the sense of an automatic generation, as generating ideas will always remain a creative process. Depending on the size and the strategic goals of a company, the idea generation might have different focuses. One can distinguish the resource-oriented and the market-oriented idea generation (Luczak et al. 2000). The starting point of a resource-oriented idea generation is a firm’s set of capabilities and new possibilities to put them into effect. During the planning process this input focus moves to the needs of the market. In the B2B sector these needs are mainly determined by the customer problems in the sense of services being a solution to an existing problem. The market-oriented approach starts with the analysis of market opportunities and customer problems as source for ideas and subsequently takes necessary capabilities and resources into account (Luczak et al. 2000, 2003). An adequate method, which brings together both mentioned aspects, is the policy deployment (Akao 1991). In any case, both the company’s core capabilities and customer problems serve as valuable resources for idea generation. The main criterion for identifying and selecting core capabilities is their potential to redound to the company’s sustainable competitive advantage. According to Barney (1991) this is the case if the resources are rare, valuable and can neither be imitated nor substituted. A Value Chain Analysis helps to gather this input data (Sontow 2000). Result of this analysis with respect to the Barney criteria is a catalogue of sustainable and superior core capabilities and underlying resources. For gathering customer problems internal and external information sources can serve as input. Departments with a high degree of interaction with the customer, i.e. sales or after-sales services usually have a huge amount of information about customer problems and customer needs. Even if this knowledge is rarely documented, it can be processed by means of workshops. Of course, one can also directly ask key customers (e.g. in workshops) what their problems are and how they think one can solve them. This is especially useful for professional services, where business relations are close and intimate. The Purpose of this analysis of customer problems is to gain a better understanding of what the problems are, what effects they have and how services can be applied as a problem solution. By bringing together sustainable competitive capabilities and problems with the help of Interdependence Analysis Method, the creation of ideas takes place sys-


51

tematically. Thus, combinations of sustainable competitive capabilities that highly contribute to the solution of a severe customer problem build an attractive basis for a service idea. The actual creation of an idea, again, is a matter of creativity suitably assisted by creativity techniques like BrainBlooming, Mind-Mapping or the 6-3-5 Method. The description of the addressed customer problems, the necessary capabilities / resources and a rough description of the solution process combine to form the service idea and conclude the phase of service planning.

4 An Architecture for Service Conception Based on the formulated service ideas, the main focus of the service conception phase is to bring more substance and depth to the ideas. The architecture as shown in Fig. 2 consists of five essential components for designing and developing professional services: The Service Development Process Model (SDPM) comprises development steps that are necessary to determine requirements and to form the functions and processes that fulfill these requirements. This model also contains steps to identify the skills and resources that are essential to perform these processes professionally. The steps included in the SDPM will be described in detail in the following sections. The architecture component Service Development Methods (SDMe) comprises methods that enable a systematic approach to the development targets. The methods best suited to support the design and development will also be shown in depth in subsequent sections. The architecture component Service Development Tools (SDTo) only contains tools that directly support distinct methods. In the understanding of this architecture, the tools of the SDTo operationalize the methods of the SDMe. The Service Development Result Description Model (SDRDM) documents the respective outcomes of the design and development steps, as well as of the service work itself. Thereby, this model also builds a common understanding among the design and development team members. The SDRDM combines functional and graphical aspects of the representation of development results. The Service Development Management Model (SDMM) integrates the four other components. The SDMM connects the development steps of the SDPM with the methods and tools of the SDMe and SDTo respectively in order to archive the development result represented in the SDRDM. Furthermore, the first practical experiences have shown that the SDMM can be used as a guideline for developing and designing professional services as well.

52


Service Development Result Description Model (SDRDM)

Service Development Process Model (SDPM)

Service Development Methods (SDMe)

Service Development Tools (SDTo)

Development Step

Methods Tools

Dev. Step Result

Service Development Management Model (SDMM)

Service Results

Service Processes

Service Resources/Skills

Total System

Subsystems … Components

Fig. 2. Essential components of the architecture

To minimize the complexity of a development project, one should not attempt to outline every detail of the service from the very beginning. Instead, the development can be commenced in such a way, that the identified requirements for the service system are implemented first in a general concept. Afterwards, the general concept can be divided into components. The determined characteristics of the general concept result in requirements for those components. Each component can then be considered independently. This procedure of specifying concepts into partial concepts and their subsequent configuration can be continued at all levels of detail in the same way. An appropriate method to detail a service system is the Function Tree Analysis under consideration of Sun’s axiomatic design (Akiyama 1991; Suh 1990). Among other things, Suh states that one can only detail a function tree with the embodying concept in mind. Based on the essential characteristics of professional services, the architecture itself is divided into three partial models with regard to the constituent elements of services: results, processes and resources. The partial models are closely connected in the sense of means-end relationships. Since results are generated by a set of processes, which still have to be specified, a certain service result for its part implies requirements for the service processes. Hence, service processes are means, which generate predetermined results. The processes in turn necessitate resources for their implementation. For this reason processes and resources represent a means-end relationship. Therefore, a complete service concept always contains a result concept, a process concept and a resources/skills concept.


53

4.1 The Service Result Branch of the Architecture This partial model of the architecture comprises activities to incorporate the external requirements of customers, as well as the internal requirements on behalf of the company, to check their plausibility, to prioritize and to substantiate them. The first step on this level is the investigation of customer and company requirements (Fig. 3). One recommended method for this is the Advanced Sequential Incident Method (Parasuraman et al. 1988; Kamiske 1997). Advanced Sequential Incident Technique (SIT)

Adv. SIT-Structure / Information Sources List

Customer Requirements

Gathering Internal Requirements

Advanced Sequential Incident Technique (SIT)

Adv. SIT-Structure / Information Sources List

Internal Requirements

Plausibility Check of Requirements

Qualitative Interdependence Analysis

L-Matrix

Service Requirements

Prioritization of the Requirements

Pairwise Comparison

Comparisons Scheme

Prioritized Service Requirements

Concretion of Requirements

Progressive Abstraction

Abstraction Scheme

Precise Service Requirements

Plausibility Check of Precise Requirements

Qualitative Interdependence Analysis

V-Matrix

Consistent Service Requirements

Benchmarking of Requirements

Advanced Competitive Product Analysis

Assessment Scheme / Information Sources List

Assessed Service Requirements

Attributes:

! Customer reception

Customer decision

Easy to find Permanent opening hours

Working

Advances in Services Innovations

Advances in Services Innovations

Advances in Location-Based Services

Innovations and Advances in Cognitive Behaviour Therapy

Innovations in Financing Public Services: Country Case Studies

Advances in Accounting Education: Teaching and Curriculum Innovations, Volume 8 (Advances in Accounting Education Teaching and Curriculum Innovations)

Advances in Accounting Education: Teaching and Curriculum Innovations (Advances in Accounting Education Teaching and Curriculum Innovations, 11)

Advances in Accounting Education: Teaching and Curriculum Innovations, Volume 7 (Advances in Accounting Education Teaching and Curriculum Innovations)

Advances in Accounting Education Teaching and Curriculum Innovations, Volume 6 (Advances in Accounting Education Teaching and Curriculum Innovations)

Innovations in Food Packaging

Innovations in Macroeconomics

Innovations in Macroeconomics

Innovations in Chemical Biology

Innovations in Hospice Architecture

Innovations in GIS 6: Innovations in GIS 6 (Innovations in Gis Series)

Innovations in Instructional Technology

Innovations in Chemical Biology

Advances in Accounting Education: Teaching and Curriculum Innovations, Volume 9

Innovations in Chemical Biology

Innovations in swarm intelligence

Innovations in Macroeconomics

Innovations In GIS

Advances in Accounting Education: Teaching and Curriculum Innovations, Vol. 10

Advances in Accounting Education Teaching and Curriculum Innovations

Innovations and Advances in Computer Sciences and Engineering

Biotechnological innovations in chemical synthesis

Teaching Innovations in Lipid Science

Biotechnological Innovations in Chemical Synthesis

Innovations in intelligent image analysis

Innovations in Macroeconomics, 3rd Edition

Energy Psychology (Innovations in Psychology)

Advances in Services Innovations