Eurographics Conference on Visualization (EuroVis) 2016K-L. Ma, G. Santucci, and J. J. van Wijk(Guest Editors)

Volume 35 (2016), Number 3

Interactive Treemap Visualisation of Call Centre Data

R. Roberts1,2, C. Tong, 2 and R.S. Laramee2

1Computer Science, Swansea University, Swansea, Wales

AbstractThere are over one million employees working at over 5,000 call centres in the United Kingdom alone. In thisunique application paper we present methods for visualising the vast amount of data generated and collectedby call centres. We design the application to address the challenges of exploration, analysis and visualisation ofcomplex, time-dependant call centre data, and aim to maximise the utility the software contributes to the businessstakeholders in the call centre industry. The utility is implemented, by the creation of overview visualisations andby encouraging user interaction with the data by means of focus and context visualisation as well as interactivefiltering in a hierarchical treemap view. This application reveals new insight into the data that traditional formsof analysis would not as quickly or as easily. We also report the reaction of domain experts to our visualisations.

1. Introduction

Over the past decade, a rapid expansion has taken place inthe call centre industry. With an estimated 5000 call cen-tres, one million employees [Cal], and a revenue of £2.3 bil-lion [rev] in the UK alone, the call centre industry makesa notable contribution to the UK GDP [GDP]. Some esti-mates claim that almost 4 million people are employed ac-cross 35,000 call centres in Europe [eur]. †

Due to the growth of this large industry, it is argued thatcall centres are becoming the modern world equivalent of thefactory, generating a valuable service as opposed to a prod-uct [Hud12]. The generalised function of a call centre (oftencalled contact centre) is the transfer of product informationand services to a customer base. These services compriseof product support, product information, or product sales.As these roles are fundamentally a service, the value of theindustry hinges critically on the efficiency of informationtransfer.

Focus is often placed on the quality of the service pro-vided by the call centres. It has been noted that there areintangible aspects of customer service that cannot be mea-sured easily or the current measures in place do not accountfor [Gil01]. For example, agents working in a call centre aregiven a temporal target in which they should aim to deal witheach customer. These limiting factors can create a negative

† That is roughly the population of Georgia the country. [Geo]

impact on the customer experience and therefore need to bequickly discovered and reversed.

The drive for improvement relies on the correct measuresof efficiency and then the ability to present the call centreactivity in an accessible format so that the correct course ofaction can be taken [CM97]. The collection of the data ismostly solved and is handled by our parter company QPCLTD, however the analysis and presentation of that data isvirtually unresearched. Here we present methods for visu-alising the large amounts of data from a contact centre inan application, with the intended end user being a decisionmaking manager within the contact centre.

This paper contributes methods by which we can viewlarge amounts of call centre data, identifying temporal trendsand highlighting issues in customer wait times. In addition tothis the application allows for a visual exploration of the datathrough an interactive focus + context hierarchical treemapthat ranges from a full day overview to a single call detaillevel.

Our contributions are:

• An introduction of call centre visualisation to the visuali-sation community.

• An enhanced focus and context treemap in a feature-richapplication with customised event filtering options.

• Novel implementation of an event based treemap visuali-sation.

• Domain expert feedback of visualisation.

submitted to Eurographics Conference on Visualization (EuroVis) (2016)

2 R. Roberts, C. Tong & R.S. Laramee / Interactive Treemap Visualisation of Call Centre Data

The rest of the paper is structured as follows:

• Related Work section• Description of the data• Overview of the software and its features• Feedback from our domain expert industry parters.• Conclusions and further work

2. Related Work

Related work falls into general categories: Research relatedto call centres and treemap visualisations.

2.1. Call Centre Analysis

As an academic area of research, the call centre industryis relatively unexplored. From a technical perspective, theoperation of efficient data routing in a call centre has beenresearched in the network domain by Adetunji [ASLM07,AL08a, AL08b]. A general architecture of an ip-based callcentre was made by Smith in 2007, outlining the generaltechnical functionality of a call centre, highlighting the im-portance of the automatic call distribution (ACD) machine[Smi07]. These papers focus on the technical design and op-erations of the centre, not the efficiency of their staff, or thecustomer experience.

A call centre’s customer service quality is highly impor-tant to the business it represents because the centre existsfor the benefit of the customer. In many cases it will be theonly point of contact the customer has with the business andthey therefore need to make a positive impact on the cus-tomer. Raja et al. performed research in this area to discoverwhat effects the employees emotional intelligence, job sat-isfaction, and work load had on the quality of the customerservice they provided [RME∗10]. The cross culture impacton customer service was also researched between Filipinocall centres and Australian customers [Owe08].

Customer feedback is highly valued within the call cen-tre industry but is costly to obtain. Research into sentimentanalysis aims to allow call centres to calculate the customersexperience without having to collect feedback from the cus-tomer [PGP∗15, GKK∗13].

2.2. Treemap Visualisation

The original treemap was developed as a space filling vi-sualisation of hierarchical data in 1991 [JS91, Shn92]. Thetreemap was improved upon by Bruls et al. by creating an al-gorithm that generates more square nodes [BHVW00]. Thisenables the user to view better compare the nodes againsteach other. The layout of these nodes was originally op-timised for position, but Shneiderman developed an algo-rthm to create a treemap that maintained an order to thenodes [SW01]. This opened up the visualisation to be usedon hierarchical, temporally-ordered data.

Treemap visualisation aesthetics have been developed sothat the structure of the tree can be shown through its shading[VWVdW99]. This new layer of data can be used on top ofnode colouring to denote another dimension in the data.

Focus + context visualisation has been applied to thetreemap that preserves the treemap properties, but focuseson selected nodes by growing them whilst maintaining order[TS08]. More evaluations of treemap browsing have beenmade, analysing the focus + context fish eye and full zoomviews [SZG∗96,SIL05]. Changes in hierarchical data can bevisualised using a treemap as long as it uses a consistent vi-sual pattern layout [TS07].

Huang et al. combined the treemap with a bar chart tocreate a hybrid visualisation that can display hierarchicaldata in a histogram view [HHZ09]. These treemap mutationsare popular for improving the aesthetic of the visualisation.Zhao and Lu created a variation of the treemap that sacrificesthe traditional space filling aspect and replaces it with circu-lar nodes. The benefits of which encourage user interactionthrough focus + context features [ZL15].

As a further development of the treemap, Vliegen etal. presented business data using various adaptations of atreemap [VvWVdL06]. Some radial methods were usedand some treemap nodes were forces into triangles or leafshapes. These visualisations are aesthetically appealing butoften sacrifice readability or consistency.

Our work is the first to use visualisation of call centre datausing treemaps. We enhance the traditional treemap visual-isation with customised filtering options especially for thedetection of patterns and outliers in call centre events.

3. Data

Our partner company QPC Ltd has provided us with 24hours worth of call centre activity data. Each call record con-tains timestamps for each sub-event, flags to denote the clas-sification of call and a range of additional information aboutthe call and caller. The data has been anonymised to avoidthe release of sensitive data.The Call Components Calls can be broken down into threeevent components.

• IVR Event• Queue Event• Agent Event

The IVR (Interactive Voice Response) is the call distribu-tion system that filters calls through to their indented desti-nations. The caller listens to pre-recorded messages instruct-ing them to respond with key-presses corresponding with thedifferent departments of a call centre. Numerous IVR eventsoccur per call. Queue events are instances of the caller wait-ing in a queue. Agent events are instances of the caller beingput through to an agent. There can be multiple queue eventsand agent events per phone call. All events are given a time

submitted to Eurographics Conference on Visualization (EuroVis) (2016)

R. Roberts, C. Tong & R.S. Laramee / Interactive Treemap Visualisation of Call Centre Data 3

stamp and multiple values that represents the nature of eachevent.

Contained within the dataset are other pieces of informa-tion about the call; an abandoned flag shows whether the callwas hung up before the caller got to speak to an agent, thetype of call is recorded (i.e. internal or external call), anddetails about the different call events.

The Data Problem The biggest challenge when attempt-ing to present data of this nature is one of scale. Accountingfor each data point collected by the call centre very quicklygenerates vast amounts of data, and attempts at visualisationoften result in obfuscation. To overcome this problem pro-grammatically, an object is created to represent a single callinstance, clustering together the events of a call into a singleobject. Within this call object all important characteristicsare stored, from timestamps to event details, omitting unnec-essary details. Once the call objects have been constructed,they can be directly accessed to create the visualisations.

4. Treemap Solution

We created a visual analytics system to explore the call cen-tre data and to identify previously unseen trends using atreemap visualisation. This software utilises the temporal na-ture of the data we are visualising. Each event contains atime stamp that makes up the building blocks of a call. Theseevents are used to build a better understanding of what hap-pened during the call, as well as placing the call on a timelinethroughout the day.

Twenty-Four Hours in a Call Centre A novel feature ofthis software is the use of temporal structuring in a treemap.Twenty four individual treemaps, each representing an hourin the call centre show a full days worth of collected data.This is done by taking advantage of the hierarchical nature oftime (i.e. Day -> Hour -> Minute) and splitting the treemapaccordingly (See Figure 1).

The twenty four, hour-treemaps are places in a 4 x 6 gridand vary in size depending on the call volume during thathour (See figure 2).

Features

Pan & Zoom The twenty four hour grid initiallyshows four hierarchical layers (Day->Hour->TenMinutes->Minute), where the colour of each minute node representsthe percentage of callers who hang up before they end upspeaking to an agent. A user option enables the fifth hier-archical layer, the individual call layer. Users can click anddrag to pan over the data as well as zoom in to take a closerlook at the individual data points. This allows the user to ex-plore the data as an overview and seamlessly transition intoa detail view.

Figure 1: This shows a full hour of calls in the treemap vi-sualisation. The hour is split into six, ten-minute nodes, andthen each ten-minute node is split again into minute nodes.Within each minute node contains the leaf node calls that en-tered the call centre within that minute. The size and colourrepresents the length of each call. Each layer of the hierar-chy also maintains order. Calls appear in the minute node inthe same order they took place in the call centre.

Figure 2: This figure shows the 24 hour overview whereall centre activity is displayed. A larger call volume withinan hour node, results in an increased size. Minute nodecolours are mapped according to the percentage of peoplewho abandoned the call before talking to an agent duringthat hour.

Detail Level Hover When the fifth hierarchical layer of de-tail is active, the user can view individual calls (See figure 1).From this view, only the total call length can be established.A user option allows the user to hover the mouse over a callinstance for an information box to appear containing furtherdetails about the call 3.

Focus & Context Filtering A main feature of the softwareis to enable filtering of the data. The overview allows the userto see all records of calls, but some patterns in the data mightnot be immediately visible. Filtering the data by parameterssuch as event lengths, or call type flags.

Figure 4 shows a comparison of an hour node at 13:00

submitted to Eurographics Conference on Visualization (EuroVis) (2016)

4 R. Roberts, C. Tong & R.S. Laramee / Interactive Treemap Visualisation of Call Centre Data

Figure 4: These images shows the versatile focus + context filtering feature. a) shows a filtered view containing only inboundcalls into the call centre. b) shows all calls who waited longer than 5 minutes to talk to an agent. c) shows all calls where a thecaller was talking to an agent for more than 50 minutes. d) shows all consult (outbound) calls.

Figure 3: This figure shows the detail level hover view thatallows the user to see specific details about the call. Detailsinclude the start time of the call, a broken down list of timespent at each stage of the call, and a treemap showing theproportion of time spend at each stage of the call.

with different filtering options enabled. Sub-figure a) filtersout all outbound calls leaving only inbound calls. You cansee no pattern in the call behaviour. Sub-figure b) only showscalls that spent longer than five minutes queuing. You cansee that there are significantly more calls waiting longer thanfive minutes between 13:00 and 13:15. This suggests that thecentre might need to stagger scheduled lunch breaks for thestaff. Sub-figure c) displays all calls where the caller spoke

to an agent for longer than 50 minutes. The lack of long callsduring the busy first fifteen minutes suggest that the qualityof service decreases when the call centre is under heavy load.Agents are more likely to end calls early to prevent queuesbuilding further.

5. Domain Expert Feedback

6. Conclusions & Further Work

In this paper, we present a novel way of presenting tempo-ral event-based data using the hierarchical treemap visual-isation. The visualisation has identified previously unseentrends in the data by means of a focus + context filteringsystem for visual exploration and analysis. Domain expertshave provided feedback stating the originality of this systemas well as the valuable insights it produces.

This is a work-in-progress project. We are continuing theresearch by investigating new ways of visualising the callcentre environment, and producing visual analytic systemswith an intended end user of the call centre industry.

submitted to Eurographics Conference on Visualization (EuroVis) (2016)

R. Roberts, C. Tong & R.S. Laramee / Interactive Treemap Visualisation of Call Centre Data 5

References[AL08a] ADETUNJI A., LARIJANI H.: Enhanced bandwidth-

delay based routing algorithm for a packet-switched virtual callcentre environment. In Simulation Conference, 2008. WSC 2008.Winter (2008), IEEE, pp. 2891–2900. 2

[AL08b] ADETUNJI A., LARIJANI H.: Routing with a bandwidthbased algorithm in virtual call centres. In Network Operationsand Management Symposium, 2008. NOMS 2008. IEEE (2008),IEEE, pp. 710–713. 2

[ASLM07] ADETUNJI A., SHAHRABI A., LARIJANI H., MAN-NION M.: Performance comparison of call routing algorithmsover virtual call centres. In Personal, Indoor and Mobile RadioCommunications, 2007. PIMRC 2007. IEEE 18th InternationalSymposium on (2007), IEEE, pp. 1–5. 2

[BHVW00] BRULS M., HUIZING K., VAN WIJK J. J.: Squari-fied treemaps. Springer, 2000. 2

[Cal] Call Centres. https://www.unison.org.uk/at-work/energy/key-issues/call-centres/. Ac-cessed: 08-12-2015. 1

[CM97] CLEVELAND B., MAYBEN J.: Call Center Managementon Fast Forward: Succeeding in Today’s Dynamic Inbound Envi-ronment. ICMI Inc., 1997. 1

[eur] Contact Centres Employ 3.7 Million People in Europe.http://www.callcentrehelper.com/contact-centres-employ-3-7-million-people-in-europe-65271.htm. Accessed: 08-12-2015. 1

[GDP] Gross Domestic Product Preliminary Estimate, Quarter2 (Apr to June) 2015. http://www.ons.gov.uk/ons/dcp171778_412372.pdf. Accessed: 08-12-2015. 1

[Geo] National Statistics Office of Georgia. http://www.geostat.ge/index.php?action=page&p_id=152&lang=eng. Accessed: 08-12-2015. 1

[Gil01] GILMORE A.: Call centre management: Is service qualitya priority? Managing Service Quality: An International Journal11, 3 (2001), 153–159. 1

[GKK∗13] GALANIS D., KARABETSOS S., KOUTSOMBOGERAM., PAPAGEORGIOU H., ESPOSITO A., RIVIELLO M.-T.: Clas-sification of emotional speech units in call centre interactions.In Cognitive Infocommunications (CogInfoCom), 2013 IEEE 4thInternational Conference on (2013), IEEE, pp. 403–406. 2

[HHZ09] HUANG M. L., HUANG T.-H., ZHANG J.: Treemap-bar: Visualizing additional dimensions of data in bar chart. InInformation Visualisation, 2009 13th International Conference(2009), IEEE, pp. 98–103. 2

[Hud12] HUDSON A.: Are call centres the factories of the 21stCentury? http://www.bbc.co.uk/news/magazine-12691704, August 2012. Online; posted 10-March-2011. 1

[JS91] JOHNSON B., SHNEIDERMAN B.: Tree-maps: A space-filling approach to the visualization of hierarchical informationstructures. In Visualization, 1991. Visualization’91, Proceed-ings., IEEE Conference on (1991), IEEE, pp. 284–291. 2

[Owe08] OWENS A.: A case study of cross cultural communi-cation issues for filipino call centre staff and their australian cus-tomers. In Professional Communication Conference, 2008. IPCC2008. IEEE International (2008), IEEE, pp. 1–10. 2

[PGP∗15] PRIYADARSHANA Y., GUNATHUNGA K., PERERAN. N., RANATHUNGA L., KARUNARATNE P., THANTHRI-WATTA T., ET AL.: Sentiment analysis: Measuring sentimentstrength of call centre conversations. In Electrical, Computerand Communication Technologies (ICECCT), 2015 IEEE Inter-national Conference on (2015), IEEE, pp. 1–9. 2

[rev] Revenue of call centers in the United Kingdom from 2008to 2020 kernel description. http://www.statista.com/forecasts/397597/united-kingdom-call-centers-revenue-forecast-nace-n8220. Accessed:08-12-2015. 1

[RME∗10] RAJA A., MOHSIN W., EHSAN N., MIRZA E., SAUDM.: Impact of emotional intelligence and work attitude on qualityof service in the call centre industry of pakistan. In Managementof Innovation and Technology (ICMIT), 2010 IEEE InternationalConference on (2010), IEEE, pp. 402–407. 2

[Shn92] SHNEIDERMAN B.: Tree visualization with tree-maps: 2-d space-filling approach. ACM Transactions on graphics (TOG)11, 1 (1992), 92–99. 2

[SIL05] SHI K., IRANI P., LI B.: An evaluation of content brows-ing techniques for hierarchical space-filling visualizations. In In-formation Visualization, 2005. INFOVIS 2005. IEEE Symposiumon (2005), IEEE, pp. 81–88. 2

[Smi07] SMITH E.: Anatomy of an ip call centre. Communica-tions Engineer 5, 4 (2007), 20–24. 2

[SW01] SHNEIDERMAN B., WATTENBERG M.: Orderedtreemap layouts. In infovis (2001), IEEE, p. 73. 2

[SZG∗96] SCHAFFER D., ZUO Z., GREENBERG S., BARTRAML., DILL J., DUBS S., ROSEMAN M.: Navigating hierarchi-cally clustered networks through fisheye and full-zoom methods.ACM Transactions on Computer-Human Interaction (TOCHI) 3,2 (1996), 162–188. 2

[TS07] TU Y., SHEN H.-W.: Visualizing changes of hierarchi-cal data using treemaps. Visualization and Computer Graphics,IEEE Transactions on 13, 6 (2007), 1286–1293. 2

[TS08] TU Y., SHEN H.-W.: Balloon focus: a seamless multi-focus+ context method for treemaps. Visualization and ComputerGraphics, IEEE Transactions on 14, 6 (2008), 1157–1164. 2

[VvWVdL06] VLIEGEN R., VAN WIJK J. J., VAN DER LINDENE.-J.: Visualizing business data with generalized treemaps. Vi-sualization and Computer Graphics, IEEE Transactions on 12, 5(2006), 789–796. 2

[VWVdW99] VAN WIJK J. J., VAN DE WETERING H.: Cushiontreemaps: Visualization of hierarchical information. In Informa-tion Visualization, 1999.(Info Vis’ 99) Proceedings. 1999 IEEESymposium on (1999), IEEE, pp. 73–78. 2

[ZL15] ZHAO H., LU L.: Variational circular treemaps for inter-active visualization of hierarchical data. In Visualization Sym-posium (PacificVis), 2015 IEEE Pacific (2015), IEEE, pp. 81–85.2

submitted to Eurographics Conference on Visualization (EuroVis) (2016)