See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/237534542

Agent-based simulation of a multi-queue emergency services call centre to

evaluate resource allocation

Article · January 2007

CITATIONS READS

3 3,616

4 authors, including:

Ric D. Herbert Peter Summons

The University of Newcastle, Australia The University of Newcastle, Australia
54 PUBLICATIONS   197 CITATIONS    58 PUBLICATIONS   252 CITATIONS   

SEE PROFILE SEE PROFILE

William J. Chivers
The University of Newcastle, Australia
43 PUBLICATIONS   296 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Simulated Pharmacy Patient View project

Modelling medical reasoning View project

All content following this page was uploaded by William J. Chivers on 14 January 2015.

The user has requested enhancement of the downloaded file.

 Agent-based simulation of a multi-queue emergency services
call centre to evaluate resource allocation
Bruce G. Lewis1,2 , Ric D. Herbert1 , Peter F. Summons1 , and William J. Chivers1

1
Faculty of Science and Information Technology, The University of Newcastle,
Ourimbah, NSW 2259, Australia.
2
NSW Police Assistance Line, PO Box 3427, Tuggerah NSW 2259, Australia.
E-Mail: [email protected],au

Keywords: Emergency services, call centre, Police Assistance Line, agent-based modelling

ABSTRACT metrics at reduced costs through more efficient

resource allocation. The agent based model provides
Emergency services call centres operate to stringent the flexibility to model the resource allocation without
parameters based on providing rapid support to having to commit considerable computing resources
their communities. It is this that differentiates the and undertaking a rigorous mathematical approach.
emergency service call centre from other call centre The model uses the extensive data base of the
categories. call centre to provide information for modelling the
agents. In this paper we describe the model at the
The focus of this paper is to model the resource systems level and present the agent-based model we
allocation within the NSW Police Assistance Line will be using.
(NSWPAL), which is a multi-site, multi-queue, multi-
server call centre that takes urgent and non-urgent
calls from the community of New South Wales,
Australia. The major resource for NSWPAL is
its Customer Service Representatives (CSRs) who
answer the calls and, although multi-skilled, may not
be equally skilled across the queues.

The business case for the model is simple. Staff

costs are a major consideration for the call centre with
unscheduled incoming calls on the different queues
leading to additional staff costs. Although rostering
of staff for the queues is based on historical forecasts,
one of the challenges in running a multi-queue centre
is to staff the queues to meet the unexpected demands,
especially during emergencies. This leads to the
shuffling of staff to meet the load and invariably
affects the service level of one of the queues.

One such event occurred in July 2007 during a

normally busy period. The inbound call rate was such
that, although all calls were answered, service level
fluctuated and could not be maintained at the target
level for a period of time. It is these situations that
we aim to model to examine staffing policies and real-
time management regimes.

The literature suggests that traditional modelling

techniques will not provide the flexibility required to
model the call centre. In this paper we propose an
agent-based model of the call centre, with the main
agents being callers, call distribution manager, Team
Leader, CSRs and the queues. The objective of the
model is to improve the business through generating
practices that lead to better customer service level

11
1 INTRODUCTION in determining the staffing levels at a New Zealand
police communications centre. As we will discuss
later, this occurs at the NSWPAL and we will describe
In this paper we propose a model to examine the a particular incident. Much of the research around call
allocation of limited server resources in a multi- centres is on finding the optimum way to staff them
queue Emergency Services call centre. Staff costs Erdem & Gedikoglu (2006).
are a major consideration for the call centre with
unexpected incoming calls on the different queues Bennington et al. (2000) explain that the advantages
leading to additional staff costs. This leads to the to customers who deal with a call centre include a
shuffling of staff between queues to meet the load reduced cost to make the call, greater convenience due
and this invariably affects the service level of one of to reduced travel, quick service once connected due to
the queues. Although rostering of Customer Service newer technologies, and the ability to communicate in
Representatives (CSRs) for the queues is based on the language of their choice. This is the case for both
historical forecasts, one of the challenges in running urgent and non-urgent calls at NSWPAL. Customers
a multi-queue centre is to staff the queues to meet dial NSWPAL on a toll-free number available from
the sudden unexpected demands, especially during fixed telephone services, are connected quickly due
emergencies. The NSW Police Assistance Line to accurate staffing and technical systems and can
(NSWPAL), which is an Emergency Services call request an interpreter if so required. They state that
centre, will be used as a source of rich data and as a the advantage to the organisation that has a call centre
case study for the agent-based model. PAL is a multi- include the ability to service many customers at the
site, multi-queue, multi-server call centre that takes one time, improved call streaming with customers
urgent and non-urgent calls from the community of making selections and less management costs due
New South Wales, Australia. to reduced office space for customers. To this we
add the use of common technical systems, common
In the following sections we examine previous technical infrastructure and in-house support services
call centre studies, looking at the need for cost located with the call centre; this is the case at
minimisation while trying to meet service level. We NSWPAL. The disadvantages relate to lack of face-
also explain how an emergency call centre can be to-face contact and unknown customer perceptions
differentiated from a commercial centre. We describe (Bennington, Cummane & Conn 2000).
the operation of NSWPAL and an situation the
affected the performance of the call centre, we identify One of the major drivers for call centre efficiency
a modelling technique that gives us the freedom is cost. There are different types of costs including
to test a number of management regimes to deal staff and telecommunications. We divide these into
with unexpected increases in call volume and finally, tangible and intangible costs. The tangible include
describe the model we will be building. staff and telecommunications and the intangible
include political fallout, reputation and media
2 PREVIOUS STUDIES reporting. Hughes (1995) notes that since “incoming
call centers are often viewed as cost rather than profit
centers accurate staffing is a primary concern” and
The literature mainly defines a call centre in that staffing to meet service levels with tight cost
commercial terms (Mehrotra & Fama 2003, Omari constraints is a challenge. Consequently, much of the
& Al-Zubaidy 2005, Gilmore 2001). The things driving force associated with call centre management
that the definitions have in common are that they and research is centred around improving efficiencies
are based on telephone systems, they are staffed by due to the high cost of labour (Bennington et al.
dedicated people known as agents, there may be one 2000). Chassioti & Worthington (2004) identify this
or more queues and that there is a high variability and as being in the order of 70% of the a call centre’s total
uncertainty in the call arrival rate for inbound centres. running cost and there are significant management
challenges in the areas of human resource being
Broadly speaking, the Emergency Services call centre recruitment, absenteeism, emotional support, burnout.
fits this mould but is differentiated by two important Duder & Rosenwein (2001) identify that the cost
aspects; timeliness and caller state. The callers may to provide trained operators exceeds half of the
be in a state of high anxiety due to the situation operational costs of a call centre. In addition,
in which they are involved and, irrespective of this, increased delays in servicing customers translate to
the agent must extract the information required to increased system and telecommunications costs due
facilitate the timely dispatch of resources. Another to customers abandoning and retrying or not retrying.
differentiator is the call arrival rate. In the context We identify another operational cost that is associated
of an inbound emergency services call centre, the with an emergency service; the cost to call in staff on
organisation has no control over the arrival rate which overtime or to shorten meal breaks within the business
is contingent on natural and human phenomena. Chen rules to meet the service level for urgent calls.
& Henderson (2001) identify this an important factor

12
Call centre performance is judged by a variety of centre managers can have only a small influence on
metrics. Of these, the most common presented short-term performance”.
in the literature are the Average Speed of Answer
(ASA), Service Level or Telephone Service factor 3 THE NSW POLICE ASSISTANCE LINE
(TSF) being the percentage of calls answered within (NSWPAL)
a specified time and, percentage of abandoned calls
(Duder & Rosenwein 2001, Mehrotra & Fama 2003),
Gans et al. (2003 cited Robbins, Medeiros & NSWPAL is an emergency services multi-queue,
Dunn (2006)). Pichitlamken et al.(2003) note that multi-server, inbound call centre. It has been
since call centres aggregate their measurement into in existence as a NSW statewide operation since
intervals of, say, 30 minutes, lack of call by call December 1999. It is a 24-hour inbound telephone
information complicates analysis of the data since call centre available to police and the community.
standard parameter estimation techniques do not NSWPAL operates as a single virtual call centre
usually apply. At NSWPAL, the granularity is 15 by virtue of its technology base. It is a niche
minutes and this is used for forecasting the call rate facility in the call centre industry, being used for the
and scheduling the CSRs. reporting of both urgent and non-urgent crime and
incidents, for providing police-related information
One of the parameters that underpins the performance to the community and for providing an intelligence
of an inbound call centre is the call arrival rate source for the NSW Police.
and it is this that is driving our research. Betts,
Meadows & Walley (2000) note that call centres have NSWPAL currently runs five significant business
seasonal demand patterns and many experience short streams and caters for planned and unplanned
term spikes in demand, for example due to weather emergency and special operations. The main business
conditions, with capacity management being a trade- streams are Triple Zero (000), Crime Stoppers,
off between operator boredom and high service levels. 131-444, Customer Assistance Unit and the police
Consequently, the nature of peaks and troughs in telephone switchboard. NSWPAL also provides
demand can have a significant effect on operation and call centre services for special operations and other
so need to be understood. In a previous study (Lewis, NSW State Government agencies. Depending on the
Herbert & Bell 2003) of the NSWPAL non-urgent number of calls in a queue CSRs change queues to
inbound queue, it was shown that there were strong service the priorities and maintain service level under
daily and weekly patterns together with unexpected the management of their Team Leaders.
spikes in the incoming call rate. This then is the
dilemma for the Team Leaders who monitor the call NSWPAL uses modern information technology and
centre operations and performance; (how) can the systems throughout all of its businesses. A Computer-
service level be maintained during an unexpected Telephony Interface (CTI) system monitors the
increase in incoming calls? incoming calls and ensures that all calls are routed
to the suitably trained agent. The same system
Mehrotra & Fama (2003) note a number of reasons maintains a comprehensive database that is used
as to why call centres are interesting as being the to show the centre’s performance in real time and
worldwide size of the industry, the operational and to prepare a variety of management reports. A
mathematical complexity, multi-queue nature, the Workforce Management System is used to forecast
inbound calls arrival times randomness, the random the various incoming call queues and to schedule staff
call durations, routing technologies complexity and in accordance with their and PAL’s business needs to
the agent skill sets differences to handle a variety meet the forecast demand.
of call types. For these reasons the authors state
that decision makers have difficulty understanding In Australia, Triple Zero (000) is the number
the system dynamics without effective modelling and dialled when people require urgent assistance for
explain the challenges faced by call centre managers an emergency service organisation. When Triple
including balancing the main areas of service quality, Zero (000) is dialled, the call goes to the
cost and employee satisfaction. These, they say, lead telecommunications carrier (Telstra) Triple Zero (000)
to important questions for which decision support operator who asks whether the caller wants police,
models are valuable; number of agents; their skill fire or ambulance. If police is requested, the Telstra
sets; how agents should be scheduled including shifts, operator transfers the call to a police operator and
breaks, training, meetings and other activities; call the Telstra operator stays on the line until the call
types, quantities and when they will arrive; inbound is taken by a police agent. At the time of writing,
calls’ response time; agent multiple queue training, about 75% of police Triple Zero (000) calls in NSW
“Call centre performance against forecast, schedule are taken by the Police Assistance Line. A PAL
and routing design” and the performance of the centre. operator takes the call and immediately starts entering
In their research model, the authors find that “call the details into the police despatch system. The
job is received by the radio operator who dispatches

13
sworn officers. Underpinning this is a series of Non−urgent calls
100
systems and redundancies to ensure the call is serviced
expeditiously. 90

Number of calls & % Service Level

80

PAL’s performance is measured in terms of its service 70

level, being the percentage of call answered in a given 60

time. This metric is different for different queues. 50
Daily performance for the 131-444 non-urgent queue
40
are published on the home page of the NSW
Forecast calls
Police website at http://www.police.nsw.gov.au/. This 30
Calls received
overall performance is directly related to the 20 Service level
Target service level
performance on the staff including the CSRs and the 10
Team Leaders who continually manage the queues. 0
18 19 20 21 22 23 00 01 02 03 04 05
Time of day in quarter hour intervals from 6pm on 6/7/07 to 6am on 7/7/07

4 PROBLEM DEFINITION
Figure 1. Non-urgent calls for the period 6pm on 6
July 2007 to 6am on 7 July 2007
While cost is an important factors in effectively
managing an emergency call centre, it is the risk to
the community that needs to be minimised. At PAL,
the CSR four-weekly rosters are prepared six-weeks in 100
Triple Zero (000) calls

advance. These are based on the forecast call volume, 90

Forecast calls
Calls received
service level and the CSR preferences. Each of the Service level
80 Target service level
queues has a different service level. For the urgent Number of calls & % Service Level

queue the target is to answer 90% of call in 10 seconds 70

while the non-urgent is 80% in 27 seconds. 60

50
Friday nights and Saturday mornings are usually a 40
busy period for the urgent queue. In July 2007
30
unexpected events caused a drop in performance of
the urgent queue. Both urgent and non-urgent queues 20

were functional. In Figure 1, can be seen the forecast 10

calls and the received calls and the fact that the 0
18 19 20 21 22 23 00 01 02 03 04 05
forecast was very good. However, the urgent queue Time of day in quarter hour intervals from 6pm on 6/7/07 to 6am on 7/7/07

tells a different story. With reference to Figure 2, for

the period 10 pm to 6 am the number of calls received Figure 2. Urgent calls for the period 6pm on 6 July
far exceeded the forecast volume and in particular, 2007 to 6am on 7 July 2007
the period 11 pm to 1 am service level was such
that it was not possible to answer the calls within
10 seconds. Since for each queue, the number of
CSRs is related to the expected call rate, the capability of the 1990s and the count-data regressions of the
to meet the extended peak was not available. The most recent decade”. Furthermore, the increase of
problem for the Team Leader on duty was how best computational power has enabled a shift towards
to optimally manage the situation with the available simulation and maximum likelihood data approach
resources. The options available to the team leaders compared to the previously used estimation and
are limited and include moving CSRs between queues, optimisation approaches. The authors identify the
changes in CSR meal breaks, calling in additional staff different ways computers can be used to examine
and extending the shifts of CSRs. Changing CSRs social structures and highlight agent-based modelling
between queues can only occur if the other queues are as an important change in social science since the
able to maintain their performance. The the latter two implications of human actions in social structures can
options are at a financial cost to the NSWPAL and so be examined. They also note that large amounts
requires management approval. of data based on social structure evolution can be
rendered visually to allow patterns to be easily
detected.
5 MODELLING AS A SOLUTION
Gilbert & Terna (2000) state that although statistical
Over the past 20 years, social science modelling has and mathematical models provide a formal base,
undergone a significant change. Gilbert & Abbott the number of equations required to “represent
(2005) explain that it has evolved from “durational real social phenomena”, particularly where non-
models of the 1980s to the logistical regressions linear relationships are present, and the often

14
implausible simplifying assumptions can lead to read and prevents problems. Gilbert & Terna (2000)
misleading theories. They say that with computer state that object-oriented techniques avoid memory
simulation, there is no difficulty in representing non- management problems and high level tools such as
linear relationships although these can produce some Swarm avoid both memory management and time
methodological problems. They note that “major synchronisation problems.
growth” of computer simulation use in the social
sciences is due to the development of agent-based Macal & North (2006) explain that ABM is different
models (ABM). from object-oriented simulation and that object-
oriented simulation is used as a basis for ABM.
Gilbert (1999) explains that simulation allows They also note that ABM toolkits are generally
emergence and development of macro-level responses object-oriented. Bonabeau (2001 cited Macal &
from individual action. Hitchins (1992) notes that North (2006)) states that ABM is grounded in the
emergence occurs where “whole entities exhibit modelling of “human social behaviour and individual
properties which are meaningful only when attributed decision making.” Macal & North (2006) state that
to the whole, not its parts” and that “every system ABM is becoming popular because systems and
exhibits emergent properties which derive from its interdependencies have been and are becoming more
component activities and structure but cannot be complex and, data is being stored in databases at finer
reduced to them.” granularity.

In the case of NSWPAL, the main parameter of In our model, we want to simulate the interaction
interest in the model is the service level. This between the rate of incoming calls, the CSR’s ability
is an emergent property derived from the speed of to service this and the management of the queues
answering the calls which in turn is dependent on by the Team Leaders under different management
the number of calls waiting, the number of agents regimes. In this section we describe briefly describe
available and the individual agents’ performance the way computers are used and conclude that because
parameters. of the non-linear parallel processes involving human
interaction, agent-based modelling (ABM) based on
There are a number of techniques and computer- Object Oriented programming techniques is the most
based programs that have been used to model call suitable approach for our model
centres. Abdulmalek (2006) notes that operational
research modelling can be used to staff call 6 THE MODEL
centres efficiently. Duder & Rosenwein (2001)
use an intuitive modelling approach to influence
management strategy for ”a large scale call centre Under the management of the Team Leaders, the
operation that provides a service function”. Omari CSRs work autonomously in processing the calls for
& Al-Zubaidy (2005) use a modelling tool called the queues into which they are logged. The CTI
OPNET to simulate their call centre under study. system works independently of human intervention
They evaluate quality of service, the average wait to ensure the calls for the various queues are routed
time, the agent’s utilisation and cost based on correctly. Associated with each queue is a complex
salaries. Gilbert & Terna (2000) identify a number decision process as to how to allocate CSRs in
of suitable tools that implement object-oriented dynamic call environment. For this, the team leaders
programming systems including Lisp-Stat, Swarm, rely solely on the presented CTI statistics. One of the
MAML (Multi-Agent Modelling Language), SDML goals of our model will be to allow team leaders to
(Strictly Declarative Modelling Language), LSD assess the impact of sudden or unexpected changes in
(Library for Simulation Development), SIM AGENT, the call rate and assess the impact on performance of
SimBioSys and StarLogo. They explain their use different queue management scenarios.
and the situations in which they are suitable. Buist
& L’Ecuyer (2005) describe a Java library called The daily operation of the NSWPAL involves parallel
ContactCentres that is used to write contact centre processes, as noted in Gilbert (1999), being a number
simulators. The authors note that specialised software of CSRs on concurrent calls and calls arriving
such as ccProphet and the Arena Contact Center concurrently into the queues. When answering and
Edition from Rockwell significantly ease modelling. dealing with the calls, the CSRs act autonomously
and make decisions (Macal & North 2006) within the
Object Oriented Design (OOD)is used for computer- bounds of their training and the NSWPAL business
based modelling. In his book, Roff (2003) discusses rules. This is similar for the Team Leaders who
the attributes and advantages of OOD. He says that monitor the business queues and allocate staff the
one advantage of OOD in comparison to procedural queues based on the queue and CSR performance.
programming and says that by hiding functionality For these reasons, and since we wish to simulate the
with the use of encapsulation makes software easier to complex processes underpinning the operation of the
call centre, we have chosen to use ABM to model

15
and simulate the NSWPAL call centre facility. We urgent and non-urgent queues and a Team Leader who
also see the appeal of the ABM approach as allowing manages the queues in real time. The environment
the performance parameters of individual CSRs to be in which the model will operate consists of a clock
included in the model to allow better tuning of the to time the events, the call centre statistic files, and
queue management process. the reports. We assume that the focus is on the
operation of the call centre after a call enters the call
The model we propose is a simplified version of the management system, that there are more incoming
NSWPAL and captures the essence of the operation lines than callers and so there is no call blocking, that
to maintain an optimum service level at minimum all customers are answered from a FIFO queue, and
cost. We examine the model on two levels. First that no calls are abandoned from the queue.
we take a systems approach and then we use this
to identify the agents together with their actions and 7 CONCLUSION
attributes as an ABM. Figure 3 shows the model as
a system with feedback to optimise the performance.
The PABX/ACD/CTI subsystem routes an incoming By examining the operation of an emergency services
call to the relevant queue. This is determined by the call centre during times of an unexpected increase in
number dialled by the caller. Calls wait in one of incoming call volume and by examining the various
the queues until answered by a CSR. Once a call is computer-based methods of solving the associated
finished, the CSR waits for the next call. Calls take a resource problem, we have determined that an agent-
finite time to complete. This is called the handle time based model will be the most suitable approach
and consists of the talk time with the caller and any to modelling the NSWPAL. We have distilled the
after call work time without the caller. In our model essence of the multi-queue operation of the NSW
we measure the handle time as a single entity. There Police Assistance Line and have translated the
are two feedback paths. In one of the feedback paths, systems approach to an agent-based model with agent
the individual performance of the CSRs emerges to classes, their attributes and their methods.
the centre’s performance. In the other path, the Team
Leaders monitor the call queue metrics and determine Our next step will be to build the model and apply
what needs to be done to optimise the service level and it to previous events while looking for emergent
ensure callers do not wait an excessive time. Particular properties and theories Further research may see the
attention is given to the urgent queue over the non- theories being applied to other emergency services
urgent. organisations.

8 ACKNOWLEDGEMENT

The authors wish to acknowledge the helpful

comments and suggestions of the referees, the
permission granted by the NSW Police Force to
undertake the research and the support of the
NSWPAL Director.

9 REFERENCES

Abdulmalek, F. (2006), Improving a call center per-

formance at a telecommunication company, in
A. Genco, A. Gentile & S. Soce, eds, ‘4th
International Industrial Simulation Conference’,
EUROSIS-ETI, Palermo, Italy, pp. 93–96.
Figure 3. The call centre model showing the main
blocks that will be modelled Bennington, L., Cummane, J. & Conn, P. (2000),
‘Customer satisfacion and call centers: an aus-
tralian study’, International Journal of Service In-
dustry Management 11(2), 162–173.
The next level is the ABM and this is best seen in the
UML class diagram hown in Figure 4. The model has Betts, A., Meadows, M. & Walley, P. (2000), ‘Call
both agents and an operating environment. centre capacity management’, International Jour-
nal of Service Industry Management 11(2), 185–
We see the agents in the model as being callers, a 196.
call management system which is the subsystem that Buist, E. & L’Ecuyer, P. (2005), ‘A java library for
routes the call to a waiting queue or a CSR, CSRs simulating contact centers’, Proceedings of the
who answer and processes the calls in each of the 2005 Winter Simulation Conference pp. 556–565.

16
 Figure 4. UML representation of the call centre model

Chassioti, E. & Worthington, D. (2004), ‘A new model Lewis, B. G., Herbert, R. D. & Bell, R. D. (2003),
for call centre queue management’, Journal of the ‘The application of fourier analysis to forecast-
Operational Research Society 55, 1352–1357. ing the inbound call time series of a call centre’,
Chen, B. P. & Henderson, S. G. (2001), ‘Two issues in MODSIM 2003, International Congress on Mod-
setting call centre staffing levels’, Annals of Oper- elling and Simulation Proceedings pp. 1281–1286.
ations Research 108, 175–192. Macal, C. M. & North, M. J. (2006), Tutorial on agent-
based modelling and simulation part 2: How to
Duder, J. C. & Rosenwein, M. B. (2001), ‘To-
model with agents, in L. F. Perrone, F. P. Wieland,
wards ”zero abandonments” in call center per-
J. Liu, B. G. Lawson, D. M. Nicol & R. M. Fuji-
formance’, European Journal of Operations Re-
moto, eds, ‘Proceedings of the 2005 Winter Simu-
search. 135, 50–56.
lation Conference’, IEEE, pp. 73–83.
Erdem, A. S. & Gedikoglu, B. (2006), ‘A dss for shift
Mehrotra, V. & Fama, J. (2003), Call center simula-
design and workforce allocation in a call center’,
tion modelling: Methods, challenges and oppor-
PICMET 2006 Proceedings pp. 1279–1289. Istan-
tunities, in S. Chick, P. J. Sanchez, D. Ferrin &
bul, Turkey.
D. J. Morrice, eds, ‘Proceedings of the 2003 Win-
Gilbert, N. (1999), ‘Simulation: A new way of doing ter Simulation Conference’, IEEE, pp. 135–143.
social science’, The American Behavioural Scien- Omari, T. & Al-Zubaidy, H. (2005), Call center
tist 42(10), 1485–1487. performance evaluation, in ‘Canadian Conference
Gilbert, N. & Abbott, A. (2005), ‘Introduction’, The on Electrical and Computer Engineering’, IEEE,
American Journal of Sociology 110(4), 859. pp. 1805 – 1808.
Gilbert, N. & Terna, P. (2000), ‘How to build and use Robbins, T. R., Medeiros, D. J. & Dunn, P. (2006),
agent-based models in social science’, Mind and Evaluating arrival rate uncertainty in call centers,
Society 1(1), 57–72. in L. F. Perrone, F. P. Wieland, J. Liu, B. G. Law-
son, D. M. Nicol & R. M. Fujimoto, eds, ‘Proceed-
Gilmore, A. (2001), ‘Call centre management: is ser-
ings of the 2006 Winter Simulation Conference’,
vice quality a priority?’, Managing Service Qual-
IEEE, pp. 2180–2186.
ity 11(3), 153–159.
Roff, J. T. (2003), UML A Beginner’s Guide,
Hitchins, D. K. (1992), Putting Systems to Work, Wi- McGraw-Hill/Osborne, New York.
ley, England.
Hughes, C. (1995), ‘Four steps for accurate call-center
staffing’, HRMagazine pp. 87–89.

17

View publication stats