VALUE OF SIMULATION

Betriebswirtschaftslehre VIIWirtschaftsinformatik

Making Simulation Work for the

Organizational Design of Communication Centers:

Challenges and Practical Experience

Michael Zapf, Katja StorchWorking Paper 6/2001

Working Papers in Information Systems

Editor: Prof. Dr. Armin Heinzl

University of Bayreuth

Department of Information Systems (BWL VII)

Universitaetsstrasse 30, D-95440 BayreuthPhone ++49 921 552807, Fax ++49 921 552216

E-Mail: wi@uni-bayreuth.deInternet: http://wi.oec.uni-bayreuth.de

Making Simulation Work for the Organizational Design ofCommunication Centers: Challenges and Practical Experience

Michael Zapf

Research and Teaching AssistantUniversity of Bayreuth/Germanymichael.zapf@uni-bayreuth.de

Keywords: Organizational design, process design, commu-nication center, case studies, process evaluation

Abstract

In this paper some important challenges for the designof communication centers are discussed. The practical valueof discrete event simulation for this domain is shown, and amethod for the systematic evaluation of organizational de-signs is presented. This approach has been tested in practiceand the results of two cases will be documented in detail.

1.CHALLENGES AND PRACTICALVALUE OF SIMULATION

Many organizations deploy communication centers inorder to establish effective and satisfactory interactions withactual and potential clients. In 1999 12.750 communicationcenters existed only in Europe and this figure is expected torise up to 28.000 until 2006 [Frost and Sullivan 2000]. Ac-cording to Cleveland, one goal of communication centermanagement is to handle the workload of incoming cus-tomer requests with the desired service level on a high-quality standard [Cleveland and Mayben 1997]. Qualifiedemployees (agents) and a well-designed organizationalstructure are prerequisites to reach this goal.

Some challenges for the design of communication cen-ters arise from the dynamic environment, especially the un-predictable behavior of customers. This often manifests invarying volumes of incoming requests, varying uses ofcommunication channels, varying processing times andvarying abort rates.

Therefore many stochastic effects have to be taken intoconsideration and stochastic discrete event simulation seemsto be an obvious tool [Law and Kelton 1991]. Simulationcan be used to evaluate the performance of specific organiz-ational designs under given environmental factors and al-lows the selection of the best design for a given purpose[Zapf and Heinzl 2000].

Nevertheless there are critical success factors for put-ting simulation into practice. We identified the followingtwo:

Katja StorchMBA-Student

University of Bayreuth/Germanykatja.storch@stud.uni-bayreuth.de

• Getting appropriate input data and• meeting the given time restrictions.

Especially for the design of new communication centersit is difficult to get “good” data because of the lack of his-torical data. Apart from that the modeling and simulationprocess have to be finished after a few days but the resultsmust build a sound base also for the future organization.

2.MAKING SIMULATION WORK

In order to handle the challenges for the organizationaldesign of communication centers we suggest two strategies:• Evaluate many environmental situations in order to an-ticipate the effects of changing environments.

• Automate the evaluation process with the help of asimulation control system to meet the time restrictions.The systematic evaluation of many environmental con-stellations can be performed in the following steps [Heinzland Zapf 2000]:

1. Identify potential organizational designs, relevant per-formance measures, relevant influence factors and ac-companying mean values and critical values.2. Simulate

a) all designs under normal conditions (mean values)and

b) worst case scenarios for all designs (critical values).3. Condense and visualize the results.

For automating this process, the simulation controlsystem SimControl has been built [Zapf 2001]. All relevantparameters and results can be stored in a simulation data-base and multiple simulation experiments can be automati-cally performed. The system has been implemented as Mi-crosoft Access application and utilizes the ARENA programas simulation kernel [Kelton et al. 1998].

The presented evaluation procedure has been applied todifferent design problems in practice [Zapf 2001, Storch2000]. In the next sections we will present the results of twocase studies in detail.

front officegeneralistsback officegeneralistsspecialistshandlestandard callif delay >= 15 secondshandleforwardhandlespecial callif delay >= 15 secondsforwardhandleasynchronousrequestFigure 1. Process Model for the Back Office Case3THE BACK OFFICE CASE3.1 Description

The first case deals with different routing strategies foran enterprise of the water supply industry. The correspond-ing communication center consists of two parts, a front of-fice and a back office. The organizational processes havebeen modeled in form of Petri-nets [Van der Aalst 2000]which are presented in Figure 1.

Most of the incoming requests are standard calls whichdeal with questions according to invoices and reminders,modifications of customer data and requests for general in-formation. Standard calls are normally handled by agentswith general knowledge (generalists) in the front office. Ifcustomers have to wait more than 15 seconds for an em-ployee in the front office, the call is automatically routed tothe back office. The same procedure is also valid for the firstrouting part of special calls which deal with technical prob-lems or difficult data modifications. The generalist whoreceives such a special call forwards it to an agent with spe-cific knowledge (specialist) in the back office. Asynchro-nous requests are posted as letters, faxes or e-mails and aredirectly routed to generalists in the back office.

In the back office case it has been examined if it ismakes sense to route exceptional standard requests withlong processing times from an agent in the front office to a

generalist in the back office. This routing strategy obviouslyhas two conflicting effects:

1. The talk times in the front office can be reduced and

therefore more agents are available to accept calls.

2. There arises an additional effort from classifying and

forwarding of exceptional standard requests.

3.2 Simulation Model and Data

The basic structure of the simulation model A withoutrouting of exceptional requests can be directly derived fromthe process model of Figure 1. Model B results from parti-tioning of standard calls into frequent and exceptional calls.Frequent calls are handled like a standard call in model A.Exceptional calls are accepted by generalists in the frontoffice and are forwarded to generalists in the back officeafter a short classification time.

Both models have been implemented with the ARENAsimulation tool-set, which provides a graphical user inter-face for the generation of SIMAN programming code [Kel-ton et al. 1998].

The initial data situation for the different request typesis presented in Table 1. The triangular distribution is usedfor modeling talk and after talk times since the exactdistribution forms are not known, but estimates for theminimum, maximum and most likely values are available[Kelton et al. 1998, 512].

Request Type

VolumeHandle TimeAfter Talkcalls perminutesTimeweekminutes

stan-frequent

3450triangulartriangulardard

(2.5, 3.6, 4.5)(1, 1.95, 3)callexcep-300triangulartriangular

tional(3.5, 4.6, 5.5)(3, 4, 5)total3750triangulartriangular

(3, 4.3, 6)(2, 3, 5)special call

20triangulartriangular(2, 3.7, 5)(13, 16, 19)asynchronous

5913triangular./.request

(7, 8.88, 10)Table 1. Simulation Data for Different Request Types in

the Back Office CaseAdditional simulation data concerning the number ofagents, the waiting tolerance of customers and the forwardtime are summarized in Table 2.

ParameterValueDimensionagentsgeneralists10# agents

front officegeneralists18# agentsback officespecialists1# agents

waitingwait time untiltriangularminutetoleranceabandonment(0.5, 1, 1.5)

call back %70%wait time untiltriangularminutecall back(1, 5, 10)

forwardexceptional20% talk timetimerequests

special requests20% talk time

Table 2. Additional Simulation Data in the Back Office

Case

3.3 Experimental Design

Both organizational designs have been evaluated underfive different environmental situations. Starting from thenormal situation, which reflects the initial data situation ofsection 3.2, different simulation parameters have beenmodified:

• The overload situation results from an increase of 50% ofthe average request volume. The volume of standard callsgoes up to 5625 calls, for example.

• The situation of staff absence results in the absence ofone front office generalist and three back office general-ists.

• A lower waiting tolerance of triangular(0.1, 0.5, 1) min-utes results in the low tolerance situation.

• In the long classification situation the classification timefor exceptional calls is doubled to 40 percentage of thetalk time.

The single experiments have been undertaken in the formof multiple terminating simulation runs [Kelton et al. 1998].The planning horizon was one week with different openinghours per day. In order to obtain expressive results, welaunched 100 runs for every experiment, every run repre-senting one week of operation.

3.4 Results

The performance of the alternative process designs isanalyzed according to two performance measures, the aver-age lost call rate for standard calls and the average through-put time for asynchronous requests. We will concentrate onthese measures, since they reflect the main results of thestudy. Additional measures which have been evaluated donot give a more detailed understanding of the designs.

Average Lost Call Rate for Standard Calls (%)302520151050normaloverloadstaff low long absencetoleranceclassificationModel AModel BFigure 2. Average Lost Call Rate for the Back Office CaseFigure 2 shows the average lost call rate for standardcalls in different environmental situations. In most situa-tions the forwarding strategy for exceptional calls (Model B)improves the performance of the design and leads to fewerlost calls. Only in the overload situation both designs havesimilar results.

The average throughput times for asynchronous re-quests are presented in Table 3. The relief of the front officein Model B leads to worse performance values of the backoffice. Especially in the overload and staff absence situationthere is a large difference between the two process designs.

front officegeneralists aftersalesback officegeneralists aftersalesspecialistshandlestandard callaftersalesif delay >= 15 secondshandleforwardhandlespecial callaftersalesif delay >= 15 secondsforwardhandlestandard asyn.request aftersalesforwardspecial asyn.request aftersalesFigure 3. Process Model for Aftersales Requests in the Networking CasehandleEnvironmentalModel AModel BDifferenceSituation(A-B)normal11,9512,38-0,43overload821,95849,82-27,87staff absence30,7581,23-50,48low tolerance11,7312,69-0,96long classification11,8912,44-0,55Table 3. Average Throughput Time for Asynchronous

Requests in Minutes for the Back Office CaseIn summary it may be said that the routing of excep-tional requests from the front office to the back office leadsto lower lost call rates in all environmental situations. Butthis advantage has to be paid with a higher load for the backoffice which is combined with longer throughput times forasynchronous requests, especially in overload situations andduring the absence of generalists in the back office.

4 THE NETWORKING CASE4.1Description

The second example is about networking communica-tion centers of an enterprise in the energy industry. Twocommunication centers A and B divide the incoming cus-

tomer requests into presales and aftersales requests, whichcan be posted as calls, letters, faxes or e-mails. Both com-munication centers have the same internal process structure,which is shown in Figure 3 for aftersales requests. Standardand special calls are accepted either in the front office or -after a short delay of 15 seconds - in the back office. Whilestandard calls are handled by the same agent who has ac-cepted the call, special calls are forwarded to a specialistwith the required know-how. A similar procedure applies toasynchronous requests which are handled completely in theback office. This process is also valid for presales requestswith one exception: A standard presales call can be routedto an aftersales generalist in the front office if no back officegeneralists is available. Please note that there are differentgroups for presales and aftersales requests but the same spe-cialist group for both request types.

The question in the networking case is, whether it ismore efficient to establish an overflow strategy which allowsthe routing of requests between location A and B in over-load situations, or to separate both communication centers.In the tested overflow strategy every call is routed from frontoffice A to front office B after a delay of 15 seconds and viceversa. If no free agent is found, the call is routed to backoffice A/B after an additional delay of 5 seconds.

4.2 Simulation Model and Data

For analyzing the networking of two communicationcenters, two simulation models have been built: The firstmodel represents the separation strategy with no call routingbetween location A and B. The second model implementsan overflow strategy between both places as described insection 4.1. Both models have been implemented withARENA the same way as in the back office case [Kelton etal. 1998]. The initial data situation for the different requesttypes is presented in Table 4.

Request Type

VolumeHandle TimeAfter Talkcalls perminutesTimeweekminutes

callsstandard

3015triangulartriangularpresales(4, 4.9, 6)(1, 2, 3)standard5481triangulartriangularaftersales(2, 2.88, 4)(1, 1.9, 3)special62triangulartriangularpresales(3, 4, 5)(2, 3, 4)special288triangulartriangularaftersales(3, 4, 5)(2, 3, 4)

asyn.standard

2098triangular./.re-presales(6, 7, 8)queststandard3933triangular./.aftersales(6, 7, 8)special110triangular./.presales(9, 10, 11)special205triangular./.aftersales(9, 10, 11)

Table 4. Simulation Data for Different Request Types in

the Networking CaseAdditional simulation data are summarized in Table 5.If a standard presales call is handled by an aftersales gener-alist the handle time is multiplied by 110%. The additionalsettling-in period for requests of a different location is mod-eled by a handle time multiplier of 105%.4.2 Experimental Design

The networking strategies have been evaluated underfour different environmental situations. Starting from thenormal situation, which reflects the initial data situation ofsection 4.2, different simulation parameters have beenmodified:

• Overload: The overload situation results from an increaseof 50% of the average request volume. The volume ofstandard presales calls goes up to 4523 calls, for exam-ple.

• Staff absence: The situation of staff absence results in theabsence of one agent per team.

ParameterValueDimensionagentspresales genera-3#agents

lists front officeaftersales gene-8#agents

ralists front officepresales genera-5#agents

lists back office

aftersales genera-9#agentslists back officespecialists2# agents

waitingwait time untiltriangularminutetoleranceabandonment(2, 4.2, 6)

call back %85%wait time until calltriangularminuteback(1, 5, 10)

forwardspecial calls20% talk timetimespecial asyn. re-10% handle time

quests

Table 5. Additional Simulation Data in the Networking

Case• Long queueing: In the long queueing situation the timebefore routing to the next agent group is raised up to 1

minute for every queue.

The single experiments have been undertaken in the form

of multiple terminating simulation runs [Kelton et al. 1998].The planning horizon was one week with different opening

hours per day. In order to obtain expressive results, we

launched 100 runs for every experiment, every run repre-senting one week of operation.

Average Lost Call Rate for Standard Calls (%)6050403020100normaloverloadstaff long queueingabsencePresales SeparatedAftersales SeparatedPresales NetworkedAftersales NetworkedFigure 4. Average Lost Call Rate for the Networking Case

4.3 Results

The process performance is analyzed as in the backoffice case according to the average lost call rate and theaverage throughput time for asynchronous requests.

In the normal situation the lost call rates for presalesand aftersales calls are equal (see Figure 4). However in theoverload and staff absence situation, the lost call rate foraftersales requests is much higher than for presales requestswhile in the long queueing situation there are more lost pre-sales calls than lost aftersales calls. This effects show theimportance of analyzing more than one environmentalsituation in order to identify the strengths and weaknesses ofdifferent designs.

Furthermore it can be clearly seen that for aftersa-les calls the networking leads to less lost calls than the sepa-ration strategy. Compared with this for presales calls theseparation of communication centers produces the best re-sults. Therefore a trade-off between aftersales and presalescalls exists.

Environmentalnormalover-stafflongSituationloadabsencequeueingpresales8,6819,3011,808,76separatedpresales12,0126,7418,2610,98networkeddifference-3,33-7,44-6,46-2,22presalesaftersales8,4226,6914,307,81separatedaftersales7,3821,198,287,32networkeddifference1,045,506,020,49aftersales

Table 6. Average Throughput Time for Asynchronous

Requests in Minutes for the Networking CaseThe advantage of the separation strategy for presalesrequests can also be identified for asynchronous requests(see Table 6). In all analyzed situations the averagethroughput time for presales requests is lower when thecommunication centers are completely separated. The oppo-site observation can be made for aftersales requests. Herethe networking strategy leads to shorter throughput times.

5 CONCLUSION

In this paper we presented some ideas to make simula-tion work for the design of communication centers. Specificchallenges of this domain have been explained and thepractical experience of two case studies has been docu-mented in detail.

ACKNOWLEDGEMENTS

We acknowledge the extensive support of MarcusGrasemann from Prisma GmbH, Neu-Isenburg/Germany,during the realization of the presented case studies.

REFERENCES

[Frost and Sullivan 2000] Frost and Sullivan. 2000. “TheEuropean Web Based Call Center Market.” Report 3856.Frost and Sullivan, Frankfurt.

[Cleveland and Mayben 1997] Cleveland, B.; J. Mayben.1997. Call Center Management On Fast Forward. Anna-polis, Maryland.

[Heinzl and Zapf 2000] Heinzl, A.; Zapf, M. 2000. “ZurReagibilität von Call Center-Prozessen.” in: Das reagibleUnternehmen, W. Dangelmaier; W. Felser, eds., Pader-born, 233-252.

[Kelton et al. 1998] Kelton, W.D.; R.P. Sadowski; D.A.Sadowski. 1998. Simulation with ARENA. Boston et al.[Law and Kelton 1991] Law, A.M.; W.D. Kelton. 1991.Simulation Modeling & Analysis. New York et al.[Storch 2000] Storch, K. 2000. “Simulationsgestützte

Evaluation von Organisationsszenarien im Call Center.”Diploma thesis, University of Bayreuth.

[Van der Aalst 1998] Van der Aalst, W. 1998. “The Appli-cation of Petri Nets to Workflow Management.” Journalof Circuits, Systems and Computers 8, No. 1, 21-66.[Zapf 2001] Zapf, M. 2001. “Gestaltung flexibler Kun-deninteraktionsprozesse im Communication Center:Theoretische Grundlagen und experimentelle Analyse.”PhD dissertation, University of Bayreuth.

[Zapf and Heinzl 2000] Zapf, M.; Heinzl, A. 2000.

“Evaluation of Generic Process Design Patterns: An Ex-perimental Study.” in: Business Process Management,W. van der Aalst; J. Desel; A. Oberweis, eds., LNCS1806, Springer, Berlin, 83-98.

BIOGRAPHY

Michael Zapf is research and teaching assistant at theUniversity of Bayreuth. He works in the field of businessprocess design and has conducted many simulation studiesin cooperation with different German companies. Commu-nication centers build one focus of his work as researcherand consultant.

Katja Storch is MBA-student and has collected exten-sive practical experience in the communication center fieldas employee of the Hewlett Packard Call Center in Amster-dam. She wrote her diploma thesis about the evaluation ofcommunication centers in cooperation with Prisma GmbH,Neu-Isenburg, a German consulting company for the designof communication centers.