28
Editorial Manager(tm) for Computer Supported Cooperative Work Manuscript Draft Manuscript Number: COSU50R2 Title: How Can I Help You? Call Centres, Classification Work and Coordination Article Type: Research Article Section/Category: Keywords: Call centres; classification; ethnography; ethnomethodology; generalization; ontology Corresponding Author: Dr Dave Martin, Corresponding Author's Institution: First Author: Dave Martin Order of Authors: Dave Martin; Jacki O'Neill, PhD; Dave Randall, PhD; Mark Rouncefield, PhD Manuscript Region of Origin: Abstract: As a comparatively novel but increasingly pervasive organizational arrangement, call centres have been a focus for much recent research. This paper identifies lessons for organizational and technological design through an examination of call centres and 'classification work' - explicating what Star (1992) terms the 'open black box'. Classification is a central means by which organizations standardize procedure, assess productivity, develop services and re-organize their business. Nevertheless, as Bowker and Star (1999) have pointed out, we know relatively little about the work that goes into making classification schema what they are. We will suggest that a focus on classification 'work' in this context is a useful exemplar of the need for some kind of 'meta-analysis' in ethnographic work also. If standardization is a major ambition for organizations under late capitalism, then comparison might be seen as a related but as-yet unrealized one for ethnographers. In this paper, we attempt an initial cut at a comparative approach, focusing on classification because it seemed to be the primary issue that emerged when we compared studies. Moreover, if technology is the principle means through which procedure and practice is implemented and if, as we believe, classifications are becoming ever more explicitly embedded within it (for instance with the

How Can I Help You? Call Centres, Classification Work and Coordination

Embed Size (px)

Citation preview

Editorial Manager(tm) for Computer Supported Cooperative Work

Manuscript Draft

Manuscript Number: COSU50R2

Title: How Can I Help You? Call Centres, Classification Work and Coordination

Article Type: Research Article

Section/Category:

Keywords: Call centres; classification; ethnography; ethnomethodology; generalization; ontology

Corresponding Author: Dr Dave Martin,

Corresponding Author's Institution:

First Author: Dave Martin

Order of Authors: Dave Martin; Jacki O'Neill, PhD; Dave Randall, PhD; Mark Rouncefield, PhD

Manuscript Region of Origin:

Abstract: As a comparatively novel but increasingly pervasive organizational arrangement, call centres have

been a focus for much recent research. This paper identifies lessons for organizational and technological

design through an examination of call centres and 'classification work' - explicating what Star (1992) terms

the 'open black box'. Classification is a central means by which organizations standardize procedure, assess

productivity, develop services and re-organize their business. Nevertheless, as Bowker and Star (1999)

have pointed out, we know relatively little about the work that goes into making classification schema what

they are. We will suggest that a focus on classification 'work' in this context is a useful exemplar of the need

for some kind of 'meta-analysis' in ethnographic work also. If standardization is a major ambition for

organizations under late capitalism, then comparison might be seen as a related but as-yet unrealized one

for ethnographers. In this paper, we attempt an initial cut at a comparative approach, focusing on

classification because it seemed to be the primary issue that emerged when we compared studies.

Moreover, if technology is the principle means through which procedure and practice is implemented and if,

as we believe, classifications are becoming ever more explicitly embedded within it (for instance with the

development of so-called 'semantic web' and associated approaches to ontology-based design), then there

is clearly a case for identifying some themes which might underpin classification work in a given domain.

Response to Reviewers: All of the corrections have been made.

How Can I Help You? Call Centres, Classification Work and Coordination

DAVID MARTIN, JACKI O'NEILL1, DAVE RANDALL2 & MARKROUNCEFIELD3

1Xerox Research Centre Europe, Grenoble, France ; 2Manchester Metropolitan University, Manchester, UK ; 3Lancaster University, Lancaster UK

Abstract. As a comparatively novel but increasingly pervasive organizational arrangement, call centres have been a focus for much recent research. This paper identifies lessons for organizational and technological design through an examination of call centres and ‘classification work’ - explicating what Star (1992) terms the ‘open black box’. Classification is a central means by which organizations standardize procedure, assess productivity, develop services and re-organize their business. Nevertheless, as Bowker and Star (1999) have pointed out, we know relatively little about the work that goes into making classification schema what they are. We will suggest that a focus on classification ‘work’ in this context is a useful exemplar of the need for some kind of ‘meta-analysis’ in ethnographic work also. If standardization is a major ambition for organizations under late capitalism, then comparison might be seen as a related but as-yet unrealized one for ethnographers. In this paper, we attempt an initial cut at a comparative approach, focusing on classification because it seemed to be the primary issue that emerged when we compared studies. Moreover, if technology is the principle means through which procedure and practice is implemented and if, as we believe, classifications are becoming ever more explicitly embedded within it (for instance with the development of so-called ‘semantic web’ and associated approaches to ontology-based design), then there is clearly a case for identifying some themes which might underpin classification work in a given domain.

Key words: Call centres, categorization, classification, ethnography, ethnomethodology, ontology

1. IntroductionTwo complaints (see Plowman et al. 1994) are commonly levied at the corpus of ethnographic and particularly ethnomethodologically-informed ethnographic studies of work and technology (see Luff et al. 2000 for an introductory review). The first is that concrete design implications of a given study can be elusive, and the second is that findings are so situation specific that re-use or generalization across situations is problematic. Arguably, the first complaint has been addressed, and continues to be addressed, in work discussing methods, practices and mechanisms for communicating fieldwork results (e.g. Hughes et al. 1994; 1997; Twidale et al. 1993; Pycock et al. 1998; Viller and Sommerville 1999; Beyer and Holzblatt, 1998). Other work has specifically, and in detail, shown how requirements and interaction models can be generated from ethnographic studies and in situ evaluations of systems and applications (e.g. Martin et al. 1998, Buscher et al. 2001; Crabtree et al. 2002). The second issue, concerning the re-use and generalization of findings, has attracted rather less attention (while remaining a bone of contention). This is perhaps because ethnographers in the field of CSCW have (mis)understood ethnomethodology as insisting on the specificities of a setting to the exclusion of any form of generalization (see Sharrock and Randall, 2004). Nevertheless, the major ‘selling point’ of ethnographic work has undoubtedly been to demonstrate how the socio-technical configuration peculiar to a setting is often crucial for the accomplishment of activities within that setting. The ‘rich’ or ‘thick’ descriptions they offer can provide valuable detail about the application domain in an accessible form for systems developers and can complement other processes - and information centred systems descriptions, for

ManuscriptClick here to download Manuscript: martin et al. class_2-04-07_final.doc

example - by drawing attention to the importance of the activities surrounding information display, distribution and pick up. This attention to detail and the desire to explicate social practices and interactions as they are played out has typically been contrasted with the application of theoretical (and thus generalizable) constructs. Ethnographic studies have, to put it simply, emphasized differences rather than similarities. We continue to believe that the ‘rich’ pictures of work they generate are an important and valuable resource. We will suggest here, however, that problems of specificity and generalizability need not be considered a function of ethnography at all, but a function of (1) the ordinary and practical difficulties present in any design project concerning the relevance of any body of findings, and (2) similar difficulties in deciding when it is safe to produce more general findings. In some ways, this perspective is echoed in a sociological/anthropological literature on ‘meta-ethnography’ (Noblit and Hare, 1988) and ‘multi-sited’ ethnography (Marcus, 1998) which attempts to answer standard criticisms of the ‘case study’ approach. Some of the purposes seen for this meta-ethnography include systematic comparison of case studies to draw cross-case conclusions, providing a way of talking about work conducted by different researchers, and providing a synthesis of ethnographic studies (1988:13). Noblit and Hare suggest (1988:38),

“… we conceive of meta-ethnographic syntheses as translations (one case is like another, except that … [our emphasis]) When ethnographies are roughly about similar things, the synthesis takes the form of reciprocal translations of each case into each of the other cases. That is to say, in an iterative fashion, each study is translated into the terms (metaphors) of the others and vice versa.”While we do not wish to buy all of the implications of this work, it nevertheless

has implications for casual assumptions about the role of the case study. A significant part of what we attempt to do here, then, is to develop an argument about appropriate forms of generalization in and through a comparative analysis of call centre work. More specifically, we do so through the analysis of classification work in call centres. The process - and we are fully aware that it is not fully mapped out - is, as Noblit and Hare suggest, one of translation. In other words, the move from a set of discrete studies towards some general conclusions will be a process of re-thinking and refining focus, re-interpreting the meaning of our data and which data seems important, and re-formulating the terminology we use in order to understand it. It is impossible to spell out the steps taken in order to arrive at more general conclusions in any mechanical fashion, because procedures will vary in accordance with the original focus of each study and the problems and issues which cropped up at that time and the kinds of solution actively under consideration, and for that matter with the purposes of the generalizing move. In very rough accordance with Glaser and Strauss (1967), however, the generation of categories and their associated properties (see also Strauss and Corbin, 1990; Glaser, 1998) would be the main purpose of such a task.

Some sort of reuse or generalization is evidently desirable. It is intuitively obvious that findings in one setting may be of relevance to design questions in another and it follows that identifying a means to draw together studies across several settings in a particular domain may deliver some kind of reusable results. Informally, it is common to see that findings in a particular ethnographic study are related to other work as a way of deepening and broadening analysis. However, less work has been reported that specifically sets out to perform this kind of deeper analysis across a set of studies by providing a uniform set of categories. Recently, the authors have been involved in a project, which had as its goal the discovery and presentation of ‘Patterns of Cooperative Interaction’, which serve as a pragmatic resource for comparing and

contrasting various ethnographic findings, demonstrating the analytic sensibilities of the approach and which can serve as a re-usable design resource (see Martin et al. 2001; 2002). A similar example of work which specifically performs a deeper analysis to bring various studies in a particular domain together is that reported by Pettersson and colleagues (2002). They present details of a study of emergency services control work and compare this with other similar studies. They argue that although phenomena such as mutual awareness, peripheral monitoring and cooperation are important in understanding the accomplishment of the work, a stronger (and more encompassing) analytic leverage can be gained by considering what occasions suchphenomena. They argue that much of this activity is provoked by the need to resolve various ambiguities in the most economic manner:

“…..their (the various other studies) several different analytic interests can be seen jointly as emphasizing the way in which awareness and other phenomena can act as techniques for resolving various kinds of ambiguity in various ways. In these studies, as with our own, ambiguity resolution is a core problem for operators. Their mutual awareness is only sometimes deployed to resolve problems. We have something to learn about those occasions upon which awareness is relevant and those where it is not.” (Pettersson et al. 2002). The success of such a re-specification of the ‘problems’ of such work is that it

redirects new design away from more “conservative” and less focused considerations;for example, away from only looking to support more or better awareness when this may not be the best solution for the specific ambiguity. Instead the focus is shifted to the core problem of identifying which forms of ambiguity routinely arise and considering how they may be resolved. These resolutions may be various and both technical and social and may or may not involve design to promote better mutual awareness.

This paper seeks to follow the form of such previous work by explicating a secondary, deeper and broader analysis of a number of studies of the different domain of call centre work. The analysis we seek to provide in this paper is broader in that it relates and brings together the findings of a number of call centre studies (most of which were conducted by the authors) and is deeper in that it synthesizes their findings under a number of topical headings only some of which have been alluded to in the original studies but which have not previously been applied to the group as a whole. The main novel analytic work of the paper is to re-specify the various design issues and problems that are highlighted as being related to the use of various classification schemes embedded in various systems of work. Not only does this allow for an over-arching view across various situations in a particular domain so suggesting it may be re-applicable in new situations; we also would argue that it helps frame the design questions in a more tractable manner. Furthermore, it suggests a vocabulary - of classification and by extension system ontology - through which ethnographers can better communicate with system designers and developers in moving from ethnography to design and in turn to system evolution through use.

A comparable type of analysis, of the relations of several case studies to classification, was presented to the CSCW audience by Simone and Sarini (2001) as a “reflection on practice” (Schon and Bennet, 1996). In their paper Simone and Sarini aim to:

“…. make sense of existing contributions, coming from various perspectives, to the theme of CS (classification schemes) in cooperation. Obviously the outcomes are neither systematic nor complete. They are an artifact typically used in design: a scaffold to sustain the work oriented to build something else. As a scaffold they are

not designed to last for a long period: their main value is to help the work to progress.”Like Simone and Sarini we acknowledge that classification schemes, as

idealizations of working practice, are often and initially produced from outside the working practice and the domain in which it is practically situated. We also agree, and our field studies attest to the fact, that neither these schemes, nor ‘the work to makethem work’ are normally treated as a fundamental object in the analysis of cooperative work. Our work is similar in that it brings together, and attempts to learn from, several empirical studies - though, in our case, drawn from a single domain. We deal specifically with the domain of the call centre and additionally only with ethnomethodologically-informed field studies. By restricting our material this way, we can confidently assume that our materials and analyses have a shared orientation. Analytic uncertainty is also lessened by the fact that most of the studies were undertaken by the authors. Finally, and as a result of this shared orientation, our focus is on the classification work, as work in action, done in customer-facing organizational encounters. 2. Classification

“Work classification systems are central to the management of a wide range of enterprises… They are important in organizing work, and they are often used explicitly as vehicles for organizational transformation, via accounting and legitimization processes. They appear as parts of accounting schemes, in technologies of organizational change such as business process re-engineering and total quality management; in addition to record keeping and accounts, they also classify people and their importance in organizations.” (Bowker and Star 1999: 229)

In recent years the ‘turn to the social’ in system design (as heralded by Grudin in his 1990 essay) has produced an emphasis on understanding the detail of social settings as a means of informing design. Understanding end-use becomes a much more prominent feature of system development to ensure that technology is appropriate both for the application domain and potential users. This paper continues this approach, tackling the twin issues of the movement from ethnography to design and generalization/re-use, identifying some lessons for organizational and technological design through an examination of ‘classification work’ and explicating what Star (1992) terms the ‘open black box’ in the perspicuous setting of the modern call centre. Call Centres - the ‘new factories’ (Bowers and Martin 2000) - are a comparatively novel but increasingly pervasive organizational arrangement. They are frequently a product of organizational change management programmes such as BPR, and the primary point of customer contact for many organizations. Call centres are prime examples of the ways in which categorization is part and parcel of modern organizational life. As part of organizational re-alignment tasks and activities are differentiated, compared, arranged and grouped together. Customers and clients, their contacts, and transactions, and contracts with the organization, are all categorized in some organizationally relevant manner whether as a note or letter or details in a file, or as recorded data on a computer system.

In order to further develop our interest in classification work, we extend and build upon the work of Bowker and Star (and Simone and Sarini, 2001), looking at the design relevance of classification through examining its actual everyday, operational nature as it features in call centre work. We do so because it became increasingly evident to us, when looking over the material we were examining, that the

standardization processes we were looking at were almost always classification processes - operators in the customer-facing work we were interested in were expected to see their encounters with customers not only in terms of a standardized procedure but also as ‘types’ of encounter, generated by schema embedded in machinery. Bowker and Star provide a wide-ranging archaeology of classification which highlights the extensiveness and general implications of categorization in organizational life. Importantly, they also provide some insight into the role of classification in systems design:

“… at the level of encoding, tools need to be sensitive to the working conditions of those encoding the data." (p159) and "Imposed standards will produce workarounds. Because imposed standards cannot account for every local contingency, users will tailor standardized forms, information systems, schedules and so forth to fit their needs…. When designing tools for distributed, organizational decision and policymaking, a detailed catalogue and analysis of such responses could become part of the designers’ tool kit; incorporated in the system, it could point out styles of workarounds at the level of coding.” (1999: 159).Bowker and Star explicitly focus on the gap between formal and informal

classification, pointing to the way in which most analyses (and we can include design ontologies here) tend to treat classification schema as natural and uncontested, ignoring the work (and contestation) that has gone into their construction. Our emphasis is slightly different. We are less concerned with the before and after, or the relationship between process and structure in classification work. We focus more on classification in action provided through studying organizational interactions with customers. Following this, the analysis of classification in action throws certain ‘challenges for classification’, as Bowker and Star term them, into sharp relief. They have, we think, a clear relationship to design problems in relation to classification. The first of these challenges is that of comparability. Put simply, comparability across sites ensures a ‘regularity in semantics and objects’ (1999: 231) and is thus almost by definition vital to any reuse exercise. That is, it seemed to us fairly evident that the degree to which terms were, and could be, stabilized, would be of considerable interest. The second is visibility. This refers to the problem of classification as it relates to invisible work. ‘Invisible’ areas of work are ‘by definition unclassifiable except as the residual category: “other”’ (1999: 231). The problem of the residual, as we shall see, is critical to the construction of design ontologies. The third is control. Bowker and Star argue that, “… any prescription contains some amount of control to be exercised by the user, be it as small as in the most Taylorist factory or prison or as large as the most privileged artists’ retreat ….. the managerial trick is to measure the degree of control required to get the job done well, for most people, most of the time.”As they go on to say, and this is central to our concerns, “From the point of view of design, the creation of a perfect classification scheme ideally preserves common-sense control, enhances comparability in the right places, and makes visible what is wrongly invisible, leaving justly invisible discretionary judgement.” (1999: 232). That is, control might be seen as the outcome of decisions concerning visibility and comparability. Taken together, these factors suggest that there are good grounds for examining classification practices because they will ramify in the development of systems which rely on classificatory schema.

Call centres are a perspicuous setting for these concerns insofar as technology already plays a fundamental role in these classificatory interactions. Customer details are fed into the corporate database, service requests are classified according to the

system set up, selling opportunities are suggested according to customer categorization, utterances are designed around customer ‘types’, details are input to provide ‘expert’ diagnosis and cure, call types are recorded for management information purposes, and so on. All of these activities are related to and involve categorization in some form. Classification is a central means by which organizations standardize procedure and practice, assess efficiency and productivity, develop and differentiate services and re-organize the business. Technology’s role in all this is as a means through which procedure and practice are implemented, while also working as a tool for recording, sorting and highlighting data - then generating statistics and reports for management purposes. These, in turn, can be used as a resource for further work and technology refinement, reconfiguration and design. Therefore there is an intimate tie between the ‘classification life-cycle’ and the systems life-cycle.

Below (section 3), we discuss four ethnomethodologically-informed ethnographies of call centre work (we discuss our reasons for limiting our analysis to these studies in the conclusion). Under a number of topical headings we examine how in different situations with different technologies, operators use classification systems in customer facing work. For example, reporting how the systems affect the structure and process of interactions and reciprocally how such interactions affect categorical selection. We also look at the sometimes multifarious entities and activities that refer to single categorizations and examine the various requirements and problems that arise by having categorizations with various users, in various situations of use.

3. Classification and Call Centres“Writings on call centre work have been overwhelmingly about rather than of the work.” (Whalen et al. 2002: 240)

Classification schemes, as embedded in systems and interwoven into work practices and intra-organizational and customer-organization interactions are an interesting topic for CSCW research. What is particularly interesting is the manner in which they serve as the coordinative mechanism for different groups of users and do so both synchronously and asynchronously. For instance, if the organization can successfully utilize a single classification system to coordinate an operator in various service interactions with different customers, and at the same or different times to coordinate a customer in various service interactions with different operators it should have a model system for distributed continuity of service. In order to do so, however, we have suggested that issues of visibility, comparability and control need to be addressed, and in what follows we will show how findings can be organized to show their relevance.

We focus on four different studies to explicate how different forms and schemes of classification are embedded and made operational in call centre work. In the first study (Martin 2000) - a software help desk located in a bank - we look at the use of a readily recognisable categorization system for the purposes of recording the different types of call being dealt with, implemented on a system called ‘Capricorn’. In the second study - of a telephone banking call centre (Bowers and Martin 2000; Martin and Rouncefield 2003) - we look at various categorizations in operation. We describe the work involved in classifying customer requests, how customers are categorized for selling products to (using a system called ‘Emerald’), how these features reciprocally impinge on interaction, and how interaction affects categorization. In the third study (also reported in Martin and Rouncefield 2003) - of insurance advice provided through videoconferencing - we look at how categorization is used to activate types of scripted interaction with customers and furthermore, how increased visibility of

categorization activity affects interaction. In the final study - of a document machine service centre (Whalen and Vinkhuyzen 1999) - we focus on how the concealed structure of categorization in an expert system (CasePoint) leads to problems in using the software for diagnosing and solving problems.

We have already pointed out that our intention in this paper is to provide a deeper analysis of the studies we refer to, rather than to provide detailed examples of data. Nevertheless, we use the findings of those studies to develop arguments about classification in action and how it relates to the twin organizational concerns of work arrangement and technology design. The reader will realize that many of the matters are intrinsically inter-related, as such is the reality of organizational life. Much of what we are discussing can be thought of as being to do with such questions as, firstly, how explicit and complete should the categories in any classification scheme be? For Bowker and Star this is a trade-off between comparability and visibility, insofar as comparability allows for use across a variety of settings, but risks an increasing degree of inappropriateness for each local setting where it is used (hence the appeal of tailorable systems). They make the point that the more comparable embedded classification schemes are (i.e. aimed at use across a number of settings), the less visible the work that goes into maintaining them will be and the less they will fit ‘local’ arrangements. They use the example of ‘indirect care’ in nursing work (p. 244 passim) to illustrate this theme. The second, but evidently related question has to do with how explicit, complete and organized a classification system might be. The importance of this lies in the degree of control afforded to the system as against the operator, or for that matter the client. We should perhaps note here that the importance of this trade-off is more than simply moral. It relates as well to the efficiency with which systems can be deployed. We begin with a discussion of visibility.

3.1 Visibility and invisibilityWe have alluded to the proposition that the issue of visibility relates to that of completeness in classification systems. Classification systems frequently rely on some use of residual categories in order to render themselves complete, and it is in this instance that the ‘invisible’ work of informally classifying is most likely to be encountered. Large, vague residual categories may render classification schema more or less useless (we enter into Borges territory with a large entity describing ‘all those other cases which don’t fit’) without extensive additional work, but at the same time a huge number of alternative categories might make certain kinds of real-time work difficult, not to say impossible since ‘deciding between’ then becomes an issue. These problems clearly impact on the degree of simplicity as against complexity required of any classification scheme and have clear implications for issues of usability and usefulness. In this way, this topic goes right to the heart of CSCW concerns. The prospect of a coherent and complete classification system, with minimal recourse to the residual, may however turn on the degree to which it can be adapted to a variety of purposes. Systems that have to serve various situations and purposes may turn out to be serving one user group more successfully than another (as has been suggested about a number of medical systems), or may support asynchronous interaction but be problematic in synchronous interaction. A contrast, for instance, might be drawn between the coordinated work required between an operator and a customer to achieve a given call (where synchronous coordination is necessary) and between a customer and a number of different operators over a series of related service encounters. In the second case maintaining a continuity of service through a persistent

record of previous contacts and service episodes (which might be considered as asynchronous coordination support) becomes an issue. But, in reality we see that a given call may well be only adequately supported by both the required historic information and systems that support this interaction, now. At the same time, an entirely different user group of classification systems are managers who use them precisely for the purpose of arranging coordination in different ways. We will argue that the degree of complexity or simplicity required of a classification scheme in these different contexts can be highlighted through a comparative study of actual use patterns. The concern for visibility, then, should draw attention to problems involving classification systems that are used both intra-organizationally and inter-organizationally (e.g. between customers and organizational representatives), and both synchronously and asynchronously.

The ‘informal’ or ‘invisible’ work that goes on around the machinery has a specific relationship to the usability of classification schemes. We try to show this in and through a discussion of how working with customers remains an invisible but consequential feature of classification. We are interested in what forms this classification at work might take; how classifications relate to activity, when classifications are problematic, and so on. The first and most obvious aspect of classification work in customer-facing work is that 'translating and mediating' work is frequently invisible. In the studies detailing the work of telephone banking and insurance work (Bowers and Martin 2000; Martin and Rouncefield 2003) a detailed account is provided of the role of the operators in working out how customer requests fit with the organizational procedures and terminology. Indeed, this is one of the key elements of the operator’s role. Below, we give some examples.

3.1.2 Managing requestsOperators utilise their knowledge of both banking and more everyday or mundane financial reasoning to mediate between customer conceptions and categorizations of what they wish to achieve and how such requests need to be formed for the organization. This is done, for example, through working up vague customer requests into organizationally relevant processes and actions and through reconciling customer perspectives and bank process through explanations of organizational categories. Thus, in the following example, the customer’s reliance on biographical detail (the order in which accounts were opened) is not matched by the bank’s categorization of accounts by ‘sort code’.

Example 1 C: …..could I transfer fifty (1.5) in:to the number one accountO: Which account number haha is that the one ends nineeightohtwo (.) yesC: Uumm nineeightohtwo is number one accountO: =You see it doesn’t actually say on screen number one account [ numbe] r two account so this is what’s confusing for[ us]C: [I know] [yes]C: =Number one is my oldest one (0.3)[ I’v]e had for a long O: [yeah]C: =time[and number t]wo is my new oneO: [I thi- yeah ]O: =I think um the way round it perhaps is uh in future if you say the one with the Swindon sort code is your number one account (0.4) [ a]nd the one with the C: [mm]

O: =central sort code is your number two accountC: Right okay then

What we see here, then, is the invisible work of making categorizations compatible by a gentle process of informing the customer how the bank’s systems work and suggesting that the customer use categories that are consistent with those of the bank. An interesting adjunct to this is observed in the videconferencing insurance advice study (Martin and Rouncefield 2003) where the operators use organizationally constructed scripts to make the talk and categorization of banking ‘at home in the life-world’ of the customer. This is done, for example, by explaining ‘standard building design and construction’ or explaining owner liability through stories and jokes that make them familiar and understandable in an everyday context.

Operators, in these studies, are knowledgeable and skilled at using the systems and their categorization schemes and of making such translations. From a CSCW perspective, here we focus particularly on the synchronous use of classification systems in the collaboration between a customer and an operator. We can see that the interaction is more accomplished when the operator is skilled and/or supported in making the categorizations understandable and sensible for the customer. Such findings are important, of course, for training regimes but also of import when we consider replacement technologies. The point is that invisible classification work reflects the fact that whatever classification scheme is used, and no matter how organizationally relevant it might be- that is, no matter how complete and consistent it might be for the organization’s purposes - it may have little or no relevance from the perspective of the customer. This translates directly into an allocation of function problem. Call centre work is typically, to a greater or lesser degree, scripted and scripts are inevitably designed to fit the organization’s classification schema (indeed they may be embedded in the technology which serves it through workflow applications). The usability of such scripts and the degree to which they can proceduralise previously discretionary work will be a function of the mediation and translation work we have described above. Bowers and Martin (2000) discuss the need to account for the mediating role when designing a replacement Internet banking application, and in the CasePoint study detailed below problems occur when the organization attempts to switch the expertise from the operator to the system.

3.1.3 Demeanour workDemeanour work or ‘emotional labour’ (Hochschild 1983) cannot, in and of itself, be considered to be a form of classification work (although it may well carry with it some implicit categories of ‘customer’). Nevertheless, the analysis of demeanour work reveals an important and often neglected aspect of the use of classification systems. That is that the act of classifying takes place within a specific context, and the context of customer-facing work requires that it be done promptly, visibly, speedily and satisfactorily. This work is done to appear competent and maintain the confidence of the customer (see Harper et al, 2000). Here we wish to draw attention to the fact that such work is often necessitated by the technology and the embedded classification system. Operators often need to maintain customer interactions while engaging with the system, for example, by making their interactions with the system visible. Signals of interaction with local technology include explanations given by operators, markers signalling ongoing interaction with system (such as ‘bear with me’) or upcoming interaction with system (e.g. ‘let me just have a look for you’) and

audible key strokes. The example below demonstrates the need to deviate from pre-existing scripts in order to preserve customer ‘confidence’:

Example 2O: Thank you mister ** the reason I ask we have a security systemC: =RightO: =Whereby we ask our customers to have a pass number C: (0.2) Right cos I take it you know I’m one of your new cus[tomers yes]O: [That’s right] yes you are so-o could I ask you now to choo:se a four digit number for the futureC: W-wo[uld]O: [And] then each time you ring would what happens is you choose a four digit number now that you know you won’t forget C: Yeah (clears throat)O: =And then when you ring in the future .hh the computer prompts it will say please can I have the first digit and the fourth digitC: RightO: Cos the operators then don’t have access to it

With CasePoint the system design treated operator-customer interaction as relatively unproblematic, easily scripted and straightforward with no attention being paid to the local, synchronous collaboration achieved through talk. Operators were meant to ask the questions exactly as they appeared on the system. However, if customers displayed frustration or reluctance, operators exited CasePoint or operators rephrased scripts so as not to appear to question the customer's competency. Similar observations are made by Whalen and Vinckhuyzen in their study of the system called CaseBase (see p118-120)

Operators similarly omitted questions or answered them without asking the customer. This is because questions are not just requests for information. They have other interactional import with respect to competencies and expertise on both sides i.e. they are accountable social actions.

A corollary to this is that in telephone help desk situations cases- where the technology is somewhat concealed (i.e. only made visible through talk) - an accomplished operator can engage in demeanour work to conceal or render sensible problematic categorizations. In the case of the videoconferencing situation such features really were visible, and thus could affect customer perceptions of organizational ‘integrity’ – a feature that was acknowledged in the subsequent abandonment of the project. It is one thing to acknowledge that organizational categorization schemes are pragmatic ‘forcing devices’ that must be used by operators (but may be improved). It is another to allow these shortcomings to become public. The design relevance of such observations lies in the issue of non-functional requirements, which dictate not only that machinery must perform certain functions but must perform them in timely and appropriate ways.

3.2 ComparabilityAs suggested, comparability refers to regularity in semantic use, and might be both temporal and geographic. That is, it seemed to us in our re-examination that some part of the work we were looking at had to do with the degree to which terms were, and could be, stabilized. Telephone call centres are “specialist technology-intensive offices that are established by organizations in order to deliver services to customers

over the telephone, replacing or complementing face-to-face interaction with the public” (Richardson 1994). The calls consist of customer and organizational representative (operator) interaction, with the operator concurrently interacting with information technology. The information system can be considered as a third party in the interaction between customer and organization (see Bowers and Martin 2000), and this fact can influence the progress of that interaction. Although different call centres deal with different content, call centres for different industries often have much in common. We might think of comparability, then, as being provided- on the face of it-in three different ways:

There is a trend towards standardization of customer-operator interaction to increase operator efficiency and eliminate operator error. In the financial services sector, one aim of standardization is to ‘configure the customer’ (Woolgar 1991: Martin and Rouncefield 2003). In banking, for example, this might be done by getting customers to phrase requests in a manner that fits with organizational processes and by having the required artefacts (chequebook, bills etc.) and details (account number, sort code etc.) to hand. The aim of this is to make customer interactions more predictable and thereby reduce call time.

As a feature of standardization many call centres aim to achieve a dispersed service through the information system, allowing any operator anywhere to take any call. In these cases continuity of service, i.e. a relationship between the organization and customer across successive ‘contacts’, is maintained through the system rather than through a relationship between the customer and an individual organizational representative.

As call centres move from an administrative to a selling function they deploy systems with increased cross-functionality designed to further control the customer-operator interaction, with systems that suggest products that the operator should attempt to sell, on the basis of customer categorization.

Nevertheless, examination of work done around the issue of comparability again suggests issues which might affect how achievable it will turn out to be.

3.2.1 Classification, coordination and standardization Comparability issues are clearly reflected in the move towards standardization of

classification schema and of procedures. Standardization serves a number of organizationally relevant purposes, notably that it aims to provide continuity of service and what we will call continuity of information use. The examination of a software helpdesk by Martin (2000) documents how the organization wanted efficiency savings by creating standard processes to deal with the large number of calls. Calls to the help desk were classified using the new system (Capricorn) according to type of call (around 100 pre-set categories), the severity and the origin of the call. Organizationally, the system was intended as a repository of problem types and solutions to improve efficiency; to aid continuity of service through customer record, and to measure work (by recording call times). This data could then be used to potentially re-arrange work by distributing personnel or assigning personnel to dealing with subsets of problems. However, some mismatches appeared between the imagined purposes of the system and the actual mechanics of classifying (for instance, and in keeping with our argument above concerning the ‘residual’, a large amount of calls were classified as ‘other’ or ‘miscellaneous’).

From the perspective of call operators the work of classifying calls was simply an activity that must be undertaken. The exactness of a given classification was a rather extraneous concern. The small, co-located, tight knit group, utilized the system

assignment of calls to operators and the accompanying recorded narrative details of calls as a means to recover previous sessions of interaction and hence to provide continuity of service with customers. New calls were either re-directed to the operator who last handled the previous call or contextual assistance (‘you dealt with so and so’s password query last week, what did it involve?’) was sought from the pertinent operator. That such activity went against the some of the manager's aspirations for (1) work measurement, (2) work re-arrangement based on categorization, and (3) continuity of service achieved only through a problem-solution database can be thought of as neither here nor there for the operators. Instead they utilized the system within their immediate context so as to best support their work. More to the point, and it bears directly on the requisite variety of terms in a classification scheme, Martin (2000) shows how, for example, many of the 100 or so categories of problem were rarely used. The types of problem listed under a category could vary quite greatly. One ‘password problem’ could easily be different from another, while ‘general problems’ showed massive variance. Within the flow of work the operators viewed problem classification as an inexact activity that was carried out under time pressure rather than an accurate portrayal of the work. The written commentaries in associated fields in fact often provided a clearer memory of the work undertaken than the classification scheme itself, offering the possibility to facilitate continuity of service through an audit trail of customer-organization interaction. They provided a way into a problem and were commonly backed up with the recollections of staff (Martin, 2000).

In this situation the lesson to be learned is that the help desk operators utilized the system as a resource for standardizing practice, and achieving coordination and continuity of service but that the resource was only actualized through their cooperative practices, and was not used in the way intended. The group, then, achieved standard ways of doing things and continuity of service through observing and talking with one another, supplementing ‘the records’ with occasioned collaboration. The records did not speak for themselves but were made sensible through cooperative techniques. Expanding on this, we can see that the system is intended to allow synchronous collaboration between callers and operators during the call and to keep a record that asynchronously allows customers to collaborate with different operators over repeated contacts. Problems of making the system actually work in this way can in fact provoke the (organizationally undesired) synchronous collaboration between operators in sharing details on customers as calls come in orpassing customers over to the ‘regular’ operator.

3.2.2 Classification, management information and competing claims on informationEqually, continuity of information use- the need for systems to support a range of organizational requirements for information above and beyond the handling of customer calls- turns out to be more problematic than one might expect. The Capricorn system (Martin 2000) for example, was designed as a problem-solution 'types' repository, an audit trail, as a measure of work, and as a resource for future strategy/design. Capricorn records enable staff to reconstruct previous interactions for practical purposes, but individual understanding and appropriate use of the records, rather than the mere existence of those records, is what makes for organizational relevance. As Randall and colleagues suggest (1996), we need to examine in some detail by whom, where, and how organizational knowledge is deployed, and the kinds of ‘knowing’ that take place in an organizational context and to understand that

context as the accomplishment of members. It thereby involves a fundamental recognition that “‘remembering’ is, on many occasions, a process where work has to be done by participants to the exchange in order to identify ‘how to remember’”.Management information statistical trawls, or ‘data mining’, or read-through’s by novices would reveal only partial information that might be problematic to use. Furthermore, in the case of Capricorn, we are not just dealing with an inevitable disjuncture between classification in action and the data output of classification but we also are confronted with the misguided lofty ambition that the system could provide multiple forms of useful management information. The following example provides a simple demonstration of that fact:

Example 3 O: Em (0.3) have you heard of our moneyplus account?C: (0.5) Moneyplus?O: Yeah basic- you know your current account no longer pays any interest on your account (0.5) well this account does would you like me to send you some information about on thatC: Eh::m (1.0) (.hh) (.hh) no I don’t think so no=O: =Not at the momentC: No I-I use my current account (0.6) [to ( )]O: [okay then that’s] fine no [problem] with thatC: [ okay ] C: Right lovely [thank you]O: [thank you]C: Bye

Here, as organizationally mandated, the operator proposes to the customer that they might want to open a new ‘moneyplus’ account. The proposal is declined by the customer but the operator is required to input a reason for this refusal into the system. The customer in fact begins to offer an explanation for their refusal (a matter to do with the account’s use, though this trails off), but what is interesting is the way in which the operator immediately translates this refusal into a machine-relevant classification (‘not at the moment’). It is not clear in this example that what is recorded on the system reflects exactly what the caller meant. In fact, we would suggest that “I don’t think so, no …” on the part of the customer does not warrant the response, “not at the moment”. The operator’s willingness to offer an organizationally (and machine-) relevant category may well be related to the Bank’s surprise at the lack of take-up when ‘not at the moment’ customers are sent application forms just in case they might have subsequently changed their minds (Bowers & Martin, 2000).

Such arguments are closely related to those offered by Anderson and Sharrock (1993) particularly in the discussion of ‘organizational knowledge’ and its relationship to action where they assert that; “The connection between knowledge and action is defined in constitutive terms. Patterns of knowledge and patterns of action define each other. Hence knowledge is seen as social through and through” Supporting such call centre knowledge work in all its contingent aspects, requires that systems pay attention to the occasioned character of activities. It has been suggested before that CSCW needs to attend to the organization rather more than it has done historically, and here perhaps is an occasion where that need is evident. Designing classification systems to support repeated synchronous collaborations between operators and customers is one concern whereas what the classified output of these

interactions can tell managers in terms of information for work and technology redesign is another. The organizational requirements of a classification scheme are multi-faceted and we can only make sense of the various demands made on the figures if one understands what goes on in the relevant interactions. A further example of the way in which chained interactions bear on organizational situations might be unpleasantly familiar to readers:

Example 4 O: …so you say you’ve been trying to close the accountC: =For over a month (.) I’ve cut up the cards (.) I’ve repaid my balance (0.4) my balance is now in credit (.) I’ve written to you three times (.) to ask you to do, to do two things (0.4) one to transfer enough to cover a loan account with the Amicable bank to close thatO: YepC: =and the rest to send me the balanceO: Yep (.) okay right I’ll just read the notes on your file just bear with me (4.3) You have a credit balance there of fifteen pounds sixty so d’you want us to send you a cheque for that credit balanceC: If y’d if-if anybody had ever read any of the letters that I’d sent and I’ve sent about three (0.7) the last letter that I’ve sent which obviously just goes straight into a dustbin somewhere (.) said would you please transfer a small amount to cover a loan account which I have with you can I give you the numberO: Yes pleaseC: The number of the loan account is nine ** * that might beee not too sure if oh its four * * * * * ** ** * * * * * * *O: (0.8) Right okay I’ll transfer that over for you (.) .hh em so what do you want us to do with the rest of it d’you do you want us to [(unheard) ]C: [Just send me a cheque]O: Right I’ll do that for you Mrs **C: (1.6) AND I wouldn’t mind an explanation why I never get a reply to any of my letters (.) where do they think they go do they go into some bin (.) I did address it to the a-a lady who had sent me a letter so presumably it wasn’t to somebody anonymous that I wrote to O: Well the last letter we have any records of that was on the fourth of September [ (unheard) ] C: [and what did that sa-] O: =that’s saying that you weren’t happy with the eh (unheard) refund the thirty pounds [we charged] C: [and why ] didn’t I be replied to you from that (.) I’ve sent two more letters since thenO: We uv got no records of the other two letters (unheard)C: Have you got any records of replies to meO: I’ll just (.) I’ll just read the notes on your file about this one (.) just bear with me (2.0) theum’ll according to our details on that letter you’ve just asked them to close the account so all we’ve done is put a block on the accountC: I WROTE two a letter to say I was not happy about returning a chequeO: YehC: I wrote a second letter with the cards cut upO: Yeh

C: =And I wrote a thi:rd letter asking you to transfer this money over from the loan from the other account (0.7) and you said you only got oneO: I’ve only got details of one letter (0.3) yeahC: And have you got any details of any reply (0.6) (unheard)O: No they haven’t put anything about a reply noC: =N:o (.) because you don’t give any replies and I’d like to put I’d like to make you to make a note on my fi:le toda:y saying that I have phoned and said that I have written three times O: Yep

Of course, the point here is that interactions between operator and customer have taken place on a number of occasions, but the only information which can be counted on ‘here and now’ is the information embedded in the system. As we see in this case, this can result in trouble. Customers may object to how they are classified, and system recordings of events are not the only evidence of their actuality and since customer service is the raison d’être of many call centres this should always be a concern that is oriented to in operator training and work design. Taken together, the problems created by the use of classificatory systems when put to a variety of organizational purposes, and the use of classificatory systems over a historically evolved period (the ‘customer life cycle’) draw attention to complexities of the multi-user system that are not always acknowledged in the CSCW literature.

3.3 ControlWe suggested above that control can be seen as an outcome of the trade-off between visibility and comparability. Control is central to the design of classification – based systems, in our view, because it is central to the problem of the residual. Skillfuldecision-making work might turn out to be a means by which the residual can be avoided or alternatively residual categories might create problems for that work. In other words, work can be, and may have to be, done in order to translate what would remain residual into one category or another. The control problem is in essence one already recognized in the CSCW literature and before (see for instance Wittgenstein, 1958; Suchman, 1967). Rehearsals of rules and their application, or as they are more commonly termed in CSCW research, plans and situated actions, are precisely examinations of this control problem. As already noted, however, Bowker and Star (op cit) make an observation more akin to that of Schmidt (1997) when they point to the practicality of issues surrounding degree of control. As they say, “This balance can never be fully resolved (as novices and strangers are always entering the field of work) …” (p. 232). The problem of the novice is very much, then, the problem of the customer for the customer in this context is often a stranger to the workings of the organization and especially to its classification procedures. Examination of call centrework in these terms might reveal something about the degree to which any given classification scheme requires, and the kinds of expertise that might need to be deployed in making classification systems usable (and equally, what it is that makes them difficult to use) and something about the kinds of novice/stranger behaviour, in this case as exemplified by customers, that create difficulty. To put it another way, an emphasis on visibility is an emphasis on what is local and often un-remarked, whereas an emphasis on control has to do with how much and what kind of mediation and control might prove necessary in different settings, and why this might be.

3.3.1 Classification and the CustomerOne aspect of this, in our view, is the specific issue that we associate with customer-facing work. Call centres are often the public face of the organization. By their very nature, interactions with customers are unpredictable; customers do not necessarily supply all the relevant information and may have an idiosyncratic view of the services, processes and artefacts of the organization. Classification is an attempt to deal with this uncertainty and an important aspect of call centre work is that the organizational procedures are often enacted within the technology that the operators use. However, this often does not fit in with the view customers have of processes and consequently customer requests may not fit in with the informational requirements of the technology. We saw this in example 1 above. Such information was not easily available to the operator and so the operator suggested an alternative manner for identifying the accounts during calls, using a, presumably mutually understandable, concept of location based on sort code. Thus, the operator was establishing a shared concept through which the customer and the operator could identify the accounts in the future. Operators therefore do not just respond to customer's service request with service delivery, but to the manner of those requests. Since operators are the interface between the customer and the organization they need understandings of both the organizational processes and practices and of customers’ everyday knowledge of what to expect. This knowledge is essential if operators are to translate the customer’s needs, reports and requirements into a format that fits with the requirements of the information technology, which tends to be designed around the organization’s processes. Of course in the telebanking situation this is made easier as operators have knowledge of the banking processes as well as shared knowledge with the customers of everyday financial reasoning. The same is not true of the CasePoint situation.

Most regularly, the customer is implicitly guided through the process by the ordering of fairly standard questions – ‘can I have your account number’, ‘what model is the photocopier’ - and the processes are demonstrated through their enactment. Over repeated contacts (in situations where customers are regular) this offers the possibility of gearing customers in to organizational process. However, as might be expected, often operators have to engage in more explicit work with customers to frame the request or gain the right details in the right way. Institutions frequently attempt to bypass customer unpredictability by ‘configuring the customer’ - basicallyby training the customer (implicitly and explicitly in service encounters) to become adept in organizational procedure. For the customer this means producing single enquiries, asking questions in right order, and so on. However, this is not possible in all call centre calls as customers regularly make multiple enquiries, frame them in the ‘wrong’ way, forget then remember requests, digress, etc. Specifically, customers’ behaviour cannot be accounted for in a simple process model. Rather, the operator mediates between the customer’s request (framed and delivered in the customer’s ownparticular way) and the institutional processes. This is accomplished by acknowledging customer’s requirements and then re-presenting these in appropriate formulation for the requirements of the technology. We might think of this in terms of a ‘customer life cycle’. If, as we suggest above, customers can be thought of as novices in their understanding of call centre classification, and if the relative absence of classificatory knowledge implicates a control issue, then there is a case for making classificatory systems reflect this. Our classifications of customers (and an aspect of the design ontology which will underpin development work in this area) might need to pay account of, for instance, whether they are ‘one-off’ (i.e. single occasion) or regular users, and the stage of their ‘life cycle’ they are at (and life cycle refers to

their familiarity with the systems in use) as there is less point in implementing measures to configure the customer if repeated contacts are unlikely or if customers are unlikely to develop the necessary familiarity.

3.3.2 Classification and expertiseIf the customer is a novice, the status of embedded ‘expertise’ is also relevant to the problem of control and by definition to the allocation of function. The expert system, CasePoint, discussed by Whalen and Vinkhuyzen (1999) uses a process of diagnostic selections and character matching to derive problem cases to which the actual photocopier trouble might match and hence lead to problem diagnosis. This is based on the idea that a particular ‘problem’ will exhibit particular ‘attributes’. This, of course, is a system of classification, based on ‘static’ expert diagnosis rather than everyday user descriptions in action. Problematically, CasePoint's design is based on the erroneous idea that operators can enter the problem description provided by the customers, word for word, and then ask word-for-word the questions CasePoint produces. Getting the problem description is very important since this description produces the first ‘best matching’ case, and determines what subsequent questions will be asked. The intention is that operators will input what the customers say directly into the system with no interpretation. Furthermore that this description will adequately sum up the problem, so that diagnosis can then begin. In actual fact, the operators rarely type in exactly what the customer says. Operators routinely and necessarily make judgements about relevant information, what terms to use, what counts as information, etc. to go in the ‘problem description’ field. The choice of wording is particularly significant since choosing one word over another might result in some cases not being matched by the system. Operators are regularly faced with this situation as customers routinely provide accounts, stories, analyses, complaints, and so forth of their machine's trouble. Operators therefore select, revise or restate such accounts to fill in the problem description as practical solutions to the necessary interpretive work, time constraints, knowledge of machines and desire to control the system's actions. Whalen and Vinchuyzen (ibid: 100-105) give the following example: Example 519 CSSR: What fault code are you getting Mike=20 CUST: =I::m not getting’ one (.) uhhm it’s (0.7) leavin’21 lines looks like somethin’>scrapin’ the paper<22 ‘cause .its goin all the way to the e:dge of the23 pa::ge<

(CSSR = Customer service and support representative; CUST = Customer)

They go on to describe how the CSSR entered the following problem description into CaseBase: cq lines across print. Only one word from the customer’s description has been used here. Operators quickly found that CaseBase’s text parsing and search algorithm produced apparently unrelated results to their queries. To address this, entry of abbreviations, such as CQ for copy quality were introduced to help limit the search and thus the results giving operators more control over suggested first questions to customers. Here then, the CSSR has assessed the customers problem as belonging to the category of copy quality, thus uses the abbreviation to narrow down the search space and thus what CaseBase suggests next. ‘Lines across print’ summarises and

reworks the customer’s description. It does not include information from the customer’s possible cause suggestion (something scraping the paper). Thus the CSSR is making judgements about what to enter to get the best results. The customer’s account wasn’t merely a description of symptoms, as is the expected entry to CaseBase, rather it contains diagnosis information, descriptions supporting this and only a little that could be considered strictly as symptom description (‘leaving lines’). Thus the CSSR must translate the customers input to fit the systems requirements.

In this instance, the envisaged benefits (less costs due to less expensive operators and fewer call outs) failed to materialize, partly because the expertise lay in the system, minimizing the ability of non-technically proficient operators to perform the role of mediation and translation to fit customer reports with the ‘correct’ organizational categorization. In other words, the balance of control between operator and system was ill-determined. The concealed nature of the classification of ‘problems’, ‘attributes’ and ‘solutions’ in the expert system (a hidden decision tree hierarchy) worked against the operators gaining an understanding of its ontology over time. By removing the expertise from the operators and putting it in the system, correct diagnosis becomes problematic. Importantly, designers and managers need to acknowledge (in their technical designs) that neither operators nor customers are passive information conduits, both must engage in interpretive activity. Operators also need to consider the quirks of the CasePoint search procedure to try to anticipate and control what will happen next:

‘…users, despite [the companies] intentions, are regularly and necessarily engaged in various kinds of analyses but are denied full access to knowledge that would make such analysis more effective, accurate and reliable.’ (Whalen and Vinkhuyzen 1999: 16).Comparatively, Bowers and Martin (2000) report on the introduction of a new

Windows based ‘selling’ front end (Emerald) for telephone banking. When the caller’s details were authenticated a basic ‘rules engine’ assigned weighted values to aspects of the customer’s account(s) and transactions resulting in products being suggested for selling to the customer at the interface level. Quite apart from the fact that the customer typology used was seen to sometimes produce interesting mis-matches of products, of interest here is that operators could (in some circumstances with some effort) make the product selections relevant to the customer circumstance. Operators were seen trawling the corporate mainframe details to see why such a selection might have been made. This in turn could be used to make the selling selection relevant to the customer’s ‘life-world’. For example, ‘I see you have had money trouble at the end of the month recently – our system has suggested you might benefit from an overdraft facility’.

The design lesson hinges on the range of expertises that can be deployed and how these expertises relate to the allocation of function problem in classificatory systems, for control largely is a matter of allocating functions between operators and machinery. At one end of this problem of control is the customer/novice problem, and at the other is the embedding of expertise entirely in the machinery. Operators (and customers) will perform their own analyses of categorical outputs, but accomplished selling or better problem-solution matching will be better supported when these analyses are supported by the systems that deliver them. Again, we can see that while classification systems may be thoughtfully and thoroughly derived by analyzing databases, interviewing experts and so forth, often the fact that they need to be made to work in a synchronous collaboration between customer and operator is either overlooked or not properly taken into account.

4. Conclusion: From Classification to DesignWe have suggested above that comparative work by ethnographers in CSCW has

been limited and have begun here to discuss how it might be done in this context through a focus on classification. Such a focus on classification might provide for researchers, managers, practitioners and designers a better understanding of call centre work in general for the purposes of systems and work design. In the light of this, we have attempted to present and provide a new deeper and broader analysis of findings from a number of empirical ethnographic studies of call centre work. Nevertheless, it is entirely obvious that one restriction on the analytic power of our work is the limited number of studies we explore (and the fact that three out of the four were conducted by the authors). There is a reason for this. Call centre and related work is extensive. It is to be found, for instance, in Pentland (1992); Ackerman and Halverson (1998); Arussy (2002); Purcell et al (2000), and Bain et al (2003) as well as many others. Even so, we come up against certain kinds of problem immediately when we attempt too broad a comparison. These have to do with the very limited availability of raw data in the published work of others, the considerable range of academic interests expressed in work on call centres (much of which, for instance, is straightforwardly sociological, and implicates issues of power, gender and control in ways that are not especially germane to our interests here), the different kinds of work being done in different centres by the operators themselves, and the disciplinary/theoretical assumptions (and hence varied terminology) deployed by analysts. Because we consider comparative work of the kind we advocate as both ambitious and early, we deliberately limited the cases we analyzed to those we conducted ourselves or conducted by others under similar (ethnomethodological) auspices. In this way, we could re-examine original data, recognize the way our interests had affected our original enquiries and ensure at least some consistency in usage. None of this is to foreswear a more ambitious approach, but it seemed to us justifiable that we should start with a manageable set of problems.

Our approach was informed by our understanding of some of the themes commonly played out in CSCW and our own interest in practical work. That is, we had an existing interest in the ‘invisibility’ of certain kinds of work, and the related gap between what is conventionally termed the ‘formal’ and ‘informal’ aspects of the organization or, more ethnomethodologically, the gap between rules and their application. More recently, the authors have become involved in work on ‘patterns’ and on work with ‘ontology-based design’. It became progressively clear to us as a result of this work that the embedding of a ‘semantics’ in machinery was of particular interest, and that various claims about the ‘semantic web’ formed a backdrop to it. Some consideration of how classification schemata are deployed in work contexts, we felt, would be a useful way of organizing and thinking about data we had collected over a period of time. This brings us back to our early point that the kind of meta-analysis we endeavour to perform here has to provide a relevant body of findings, and do so at the appropriate time. Our instinct - and it is no stronger than that - is that, discussions about the relative homogeneity or otherwise of cases for comparison, distance from the original studies, and so on are actually discussions about how we acquire a sense of relevance, and when it is correct to do so. That is, where more heterogeneous settings might ultimately provide more interesting findings, they also might be more dangerous. We advocate, in other words, a cautious approach to these problems.

We have argued that Bowker and Star’s (ibid) analysis seems to be particularly design relevant in this case. The authors suggest for example that, “at the level of encoding tools need to be sensitive to the working conditions of those encoding the data”. In addition, they document the importance of workarounds; argue that the complexity of representation should be tailored to organizational scale and advise us to, "match the structure of the information system mediating among diverse participants with information needs". Nevertheless, and as argued above, any focus on classification needs to be judged on the way it allows (or not) us to proceed from ethnographic record to design. We suggest that this focus is particularly pertinent because it provides a framework for understanding that can be readily mapped onto the ‘natural’ concerns of designers. We consider that the focus on aspects of classification work impacts on three broad design interests. They are system ontology, allocation of function and usability for multi-user systems.

4.1 Design, Classification and OntologyFirstly, the focus on classification may provide a ready route to move from

ethnographic study to design, by drawing attention to the use of ontology in systems design. Ontology-based deign has become a great deal more common in recent years, perhaps because it is seen as forming a crucial part of the ‘semantic web’. Certainly, the explicit concern with conceptualization in ontologies means they are, at present, arguably the most sophisticated form of classification scheme to be found embedded in computing machinery. In addition, because of the so-called ‘open world’ assumption, they have considerably more flexibility that is associated with the typical database. In the light of claims made about the value of ontologies for, for instance, knowledge management (see e.g. Fensel, 2003) they seem particularly relevant to our concerns.

Dourish (2001) points out that ontology “…addresses the question of how we can individuate the world, or distinguish between one entity and another; how we can understand the relationships between different entities or classes of entity; and so forth.” (Dourish 2001: p. 129). Clearly classification for organizational purposes is the practical enterprise of ontology, because it relates to the purposes of arranging, segmenting, analyzing and measuring work. Obviously, in this sense this is a thoroughly pragmatic enterprise, not aimed at describing the essential, philosophical elements of the relationships between and individuating characteristics of e.g. customers, transactions and products. However, of course, whatever classification is produced, it has to have a relationship to actual activities and people and has to be worked with in the ‘real’ world. Classification in organizations is closely tied to, and implemented as, ontology in the systems design sense. As Dourish explains:

“Ontological problems manifest themselves quite quickly in systems design. From the outset, designing a software system involves making decisions about entities, about their types and relationships, and determining how these will “line up” with the elements of the real world to which they refer.” (2001: 129) Classification systems may be implemented at different ‘levels’ in computer

systems and their impact on the ‘user’ can be various and always should be a key consideration. They comprise the ‘ontology’ of an object oriented database, or the manner in which an expert system is arranged or categories under which a call may be recorded, or (as pointed out, again in Dourish 2001) are emergent in our own arrangements of the desktop filing system. They need to be worked with in practice, so a crucial point is how do they relate to the ‘world’ and activities of the user and what does the user need to understand about the composition of the ontology in order

to make it work? We create our own ontology of file structures as we amass and arrange files in a way that makes sense for us. But the point for the administrator (and the designer, designing for a set of differentiated users) is whether classification schemes are a practically useful way of categorizing work. Do they stand for what was intended? Do they provide the desired result?

This discussion of classification and ontology (in the systems design sense) also provides a new route through the perennial thorny problem of the relationship between ethnography (a focus on the particulars of embodied practice) and systems design (with requirements for formally structured systems). As colleagues of ours have argued elsewhere (Randall et al. 1992), while we are strong supporters of ethnographic methods, their role in the design process is primarily as an informational input making visible the ‘real world’ aspects of a work setting. However, we are also aware that there are problems in enabling designers to utilise ethnography. We alluded to the prospect that the limitations of ethnographic enquiry- conventionally viewed as problems of translation, of ‘messy’ data and of limited generalizability may in fact be problems of a quite different kind. We re-iterate, then, that problems of specificity and generalizability need not be considered a function of ethnography at all. Design relevance is, and always has been, the primary issue and problems of generalizability from ethnographic studies are problems of uncertainty (see Twidale et al, 1994). Any data can, in principle, at any time be generalized and categorized. The problem, of course, is whether the patterns we determine when we engage in an exercise of this kind turn out to matter at all from a design point of view. The problem, then, is how to bridge the gap between these two kinds of representation. The focus on classification work in this context (and potentially in others) provides a means by which ethnographers and designers and developers can find a common understanding to tackle design problems. It would appear that Dourish re-enforces this argument:

“Systems designers use it (the term ontology) in a number of ways. On the technical side, it generally means the internal representational structure of a software system – how are they distinguished and so on. This is the traditional stuff of object-oriented analysis, or data structure or database design. For example, when the designers of a hotel registration system debate whether a reservation should be recorded as a feature of a room or of a day or of a person, they are designing the “ontology” of their system. At the same time, “ontology” is also sometimes used to refer to the elements of a user’s conceptual model-the model either of the user’s own work or of the system operation. Users’ understandings are a matter of some significance to system designers, who need to design a system model that will fit “the user’s ontology.” (2001: p. 130)That there is something of a mismatch between different conceptions of

‘ontology’, and that this engenders some problems is evident. As one teacher said about the problem of classes in ontologies, ‘… and as well as all those problems, there’s one other … and that’s anything to do with actors and actions .. they always cause problems’ (fieldnotes, Feb 2007) Making ethnography engage with matters of design through an orientation to classification should more readily produce and present findings in a manner that is more commensurate with the technical talk of designers in terms of systems ontology. We are not suggesting that the work we report here will necessarily result in better ontologies - it is entirely possible that some things simply cannot be accurately modelled - but we do believe that it will provide, at very least, a useful adjunct to ontologies. Representations of the user, their work activities and the customer may be embedded within ontologies or left outside. However, Dourish points out (2000), innovative variations of object oriented programming may

be able to respond in some manner to the social aspects of work, where example, ‘different objects have different perspectives on a target object’. One might consider how such an implementation would allow for a categorical distinction to have a different ‘operational’ form for different users, or the purpose to which it is put.

Two aspects, then, of system ontology building might depend on the kinds of ethnographic meta-analysis we argue for. The kinds of problem that bedevil ontology-building are problems of scale, scope, detail and usefulness. Whether or not, as the realists would have it, an ontology of everything is possible, as a matter of practical contingency important decisions have to be made about the number of classes to be identified, their properties and relations. That is, the ontology needs to classify persons and events in such a way that it covers all relevant possibilities (including the rare exceptions), and discriminates them sufficiently such that the consequences of different events and behaviours can be identified and dealt with. Secondly, and at the same time, unnecessarily complicated or elaborate classification procedures create an overhead for operators. That is, trade-offs need to be found in the design of the underlying ontology which allows operators to make useful distinctions between one kind of event or person and another (failure to do so leads to over-reliance on residual categories) whilst avoiding over-discriminated categories which lead to decision-making infelicities of the kind we have seen. These, we have suggested, might include either empty categories in systems or problems in determining which category is the appropriate one.

4.2 Design, Classification and Allocation of FunctionOur attention to the visible and invisible aspects of call centre work draws attention to another (but related) matter for system designers, that of the allocation of function between human being and computer. Our examples show two distinct aspects of this. Firstly, they demonstrate the issue of control when customers are implicated because customers are a perspicuous example, in our view, of the ‘novice’ problem raised by Bowker and Star. Secondly, ‘embedded’ expertise raises a similar set of issues. So-called ‘intelligent’ systems have always had limitations predicated on the problem of exceptions- those relatively rare occurrences which nevertheless have system consequences- and on the problem of circumscribing the domain- capturing all the knowledge that is required in order to operate in the domain. We have tried to show that the ‘novice’ problem has implications for the effective design of systems in call-centre work. Inattention to the need for translation between customer-defined ‘presenting problems’ and embedded system classifications leads to over-elaborate and inappropriate classification systems which make work processes more difficult to accomplish and make the results of action less tractable for other purposes. Our observations lead us towards the conclusion that in many cases what is important is that the ontology of a system is accessible and learnable to the user through their interactions over time (and we are lead to consider arguments about system accountability by Dourish and Button (1998)) and that the classification is evolved in use. Accountable systems are systems that display the process of their operation to their user as they carry out tasks. If the right level of abstraction is found for the user this should allow them to form an understanding of how the system operates that facilitates rapid learning through use. Considering learning and proficiency, we can see the contrasting situations of telephone banking and the CasePoint study that the ‘work to make the classification work’ is better facilitated when the operator has a thorough-going knowledge of how the system is arranged, what categories refer to, and how they are individuated.

4.3 The Administrator’s Problem: Usability for Multi-User SystemsWhile interactions conducted through call centres in many cases might be thought of as the ‘front line’ for managing the customer relationship, the classification systems underpinning them might serve a range of other purposes as well. Most notably, such schemata serve to populate metrics of work. That is, they provide the basis for managerial work concerning the relative importance of individual (or collective measures of) customer perception or satisfaction as measured against cost or time savings. Management, like design (and classification) is an inherently ‘satisficing’ enterprise. Managers constantly deal with what has termed the administrator’s problem (Garfinkel 1967) - will whatever measures are taken to solve a problem (improve process, efficiency, lower costs etc.) have been worth it? The determination of the value of possible solutions with respect to expense, inconvenience, and payoff and so forth is a constant juggling act. Will expenses be lowered or just shifted elsewhere or will new expenses be incurred? Managers employ a complex variety of measures that are not directly comparable according to single criteria – measures of call and transaction times, lists and ‘scores’ of product and problem types, customer numbers and breakdowns, customer satisfaction figures, complaints, surveys and reports. However, as a practical enterprise these are compared and contrasted (juggled against one another) as the organizational tools for decisions. The manager or administrator succeeds when her decision produces the right outcome (by whatever measure is employed to assess this). The task is a complex one as measures are not consistent over time – the criteria for making judgements on work shift. However, for the administrator (and by extension the system analyst and designer) understanding how the measures are produced, how they are compiled in practice and what the ‘truth’ is behind the figures is of common importance. In the manner of Bowker and Star (ibid) we urge that in order to design classifications that ‘work’ and are useful, managers, analysts and designers should consider the ‘other’ users of the systems –the operators and the customers and furthermore should pay attention to their practical use. These aspects of classifications (and their instantiations as systems ontology) have organizational ramifications.

4.4 Classifications and Cooperative WorkWe would contend that the less exotic, more mundane, forms of CSCW system such as those to be found in call centres have not been as thoroughly explored as other areas. We aim to provide a framework for developers, designers and managers considering the design and evaluation of such systems and an analytic theme (classification/ontology) for communication between ethnographers and designers. We believe that we provide a useful framework for analysing and teasing apart different concerns for systems and work design if we consider category schemes in terms of: (1) the user groups, (2) whether use is between or within groups, (3) whether use is individual or collaborative, (4) whether interaction/collaboration is synchronous or asynchronous, (5) whether the computer systems are shared (visually and/or aurally) or available to only one party, and finally (6) what the multiple uses, users and configurations are. This paper is intended to show how this framework was developed through a deeper analysis of original empirical work. We have demonstrated how such a framework enables analysis and provides concrete design suggestions for the cases that we have re-analyzed. In doing so, we want to relocate discussion concerning ethnography and design towards some neglected features. Much of the discussion around ethnography and design has, in our view, mistakenly

emphasized the ethnographic focus on detail, and in an associated way, on the ‘messy’ and unstructured form which data takes in this form of enquiry. This has happened perhaps because of an unwarranted emphasis on ‘concurrent’ ethnography (Hughes et al, 1994). The model of design in which requirements are set before (or even during construction) at least in part as a result of ethnographic findings (and where the ‘problems’ of ethnographic findings are construed as problems of lack of structure or as problems of detail as against generalization) is neither one we subscribe to nor one that is common in the real world. Ethnographic data is no more ‘messy’ and unstructured than any other kind of data, and generalization through a comparison of findings is not, in principle, difficult to do. That it often appears that way is a function, no more and no less, of the way in which ethnographers have historically plied their trade. In our view, the real issue here has to do with when and in what relevant way ethnographic data should be structured.

If we agree with Bowker and Star (1999) that the only ‘good’ classification is a ‘living’ classification then supporting design-in-use requires “that we as designers engage in the unfolding performance of their work as well, co-developing a complex alignment among organizational concerns, unfolding trajectories of action, and new technological possibilities”. (Suchman, 1995) Here designers attend to the specifics of the workplace rather than have representations of work brought to them, since “working practice is lived experience, only partially representable” (ibid). In this way,‘classic’ debates about linking ethnography to design become less relevant as the work of design and development shifts into the users’ workplace and design becomes shaped by workplace practice as it simultaneously attempts to re-design it. As stated by Hartswood et al. (2002), “The culture of design as a relatively isolated process... must be replaced by accountable design, which...means enabling the unfolding implications of technology for the workplace in which it is located”. Enabling means studying the implications of design on work practice, in studies like these presented here, and being responsive to these in continued systems development. In this view the ‘design problem’ is less about the creation of new technologies than with the effective re-configuration and re-integration of existing systems with developing work practices. This involves looking beyond technical design to the ongoing struggle to make particular systems work for particular users, in a particular workplace at a particular time. This process, as we have suggested is, of course, in part to do with studying what it is that is unique and particular to given settings. We can see no reason why some gentle comparative work, based on these findings, cannot also be done. Understanding a central means by which organizations - in this paper call centres - organize their business and standardise procedure, understanding classification and classification work, appears an essential first step in making the design problem malleable.

ReferencesAckerman M.S. and C. Halverson. (1998) Considering an organization's memory. Proceedings of the ACM Conference on Computer Supported Cooperative Work (CSCW 98). pp. 39-48. Seattle, WA: © ACM Press.Anderson, R. and Sharrock, W. (1993) Can Organizations Afford Knowledge? Journal of CSCW, Vol 1, no.3. London: Kluwer Arussy, L. (2002). Don't take calls, make contact. Harvard Business Review, Vol. 80, no. 1, pp. 16-17Bain, P., Watson, A., Mulvey, G., Taylor, P. and Gall, G. (2003). Taylorism, Targets and the Quantity-Quality Dichotomy in Call Centres, New Technology, Work and Employment. 17.3, pp154-169Beyer, H. and Holzblatt, K., (1998) Contextual Design: Defining Customer-centred Systems. San Francisco: Morgan Kaufmann

Bowker, G. & Star, S. L. (1999) Sorting Things Out: Classification and Its Consequences. Cambridge MA: MIT Press.Bowers, J. & Martin, D. (2000) Machinery in the new factories: Interaction and technology in a banks telephone call centre. In Proceeding of CSCW 2000, Philadelphia, © ACM Press.Buscher, M., Mogenson, P. and Shapiro, D. (2001). Spaces Of Practice. Proceedings of ECSCW 2001. Bonn, Germany 16-20th September. Kluwer.Cicourel, A.V., (1969) The Social organization of Juvenile Justice, New York: Transaction PublishersCrabtree, A., Hemmings, T. and Rodden, T. (2002) Pattern-based support for interactive design in domestic settings. Proceedings of the 2002 Symposium on Designing Interactive System. London: ACM Press.Dourish P, Button G. (1998) On “Technomethodology”: Foundational relationships between Ethnomethodology and System Design. Human-Computer Interaction 13(4). pp395-432.Dourish, P. (2000). Technical and Social Features o Categorization Schemes. Workshop on Classification Schemes and Cooperative Work. CSCW 2000, Philadelphia, © ACM Press.Dourish, P. (2001). Where The Action Is: The foundations of embodied interaction. Cambridge (MA) and London: MIT Press.Fensel, D. (2003). Ontologies: A Silver Bullet for Knowledge Management and Electronic Commerce. London: SpringerGarfinkel, H. (1967). Studies in Ethnomethodology. Englewood Cliffs: Prentice Hall.Glaser, B.G. and Strauss, A.L., 1967, The Discovery of Grounded Theory: strategies for qualitative research. Chicago: AldineGlaser, B., 1998, Doing Grounded Theory: Issues and Discussions, Mill Valley, CA., Sociology PressGrudin, J. (1990). The Computer Reaches Out: The Historical continuity of Interface Design. Proceedings of ACM Conference on Human Factors in Computing Systems. CHI '90. Seattle: ACM Press. pp19-26 Hartswood , M., Procter, R., Slack, R., Voß, A., Büscher, M., Rouncefield, M., and Rouchy, P. (2002) Co-realisation: Towards a Principled Synthesis of Ethnomethodology and Participatory Design. The Sandinavian Journal of Information Systems, 14:2 pp. 9-30.Hochschild, A. R. (1983) The Managed Heart: Commercialization of Human Feeling. University of California Press. Berkeley.Hughes, J., King, V., Rodden, T., Andersen, H. (1994) Moving Out from the Control Room: Ethnography in System Design. Proceedings of ACM CSCW'94 Conference on Computer-Supported Cooperative Work: ACM Press. pp.429-439 Hughes, J., O'Brien, J., Rodden, J., Rouncefield, M., Blythin, S., (1997) Designing with Ethnography: A Presentation Framework for Design. Proceedings of DIS'97: Designing Interactive Systems: Processes, Practices, Methods, & Techniques 1997, ACM Press. pp.147-158Luff, P., Hindmarsh, J. and Heath, C. C. (eds.) (2000) Workplace Studies: Recovering work practice and informing system design. Cambridge: Cambridge University Press.Marcus, G. (1998) Ethnography through Thick and Thin. Princeton University Press.Martin, D., Wastell, D. and Bowers, J., (1998) Ethnographically Informed Systems Design (EISD): The development and evaluation of an Internet-based electronic banking application. Proceedings of European Conference on Information Systems (ECIS), Aix-en-Provence, France. June 1998.Martin, D (2000) Ethnomethodology and computer systems design : interaction at the boundaries of organizations. PhD thesis Manchester University. Copies available from [email protected], D. and Rouncefield, M. (2003). Making the Organisation Come Alive: Talking Through and about the technology in remote banking. Human-Computer Interaction, Vol. 18, No’s 1 & 2, pp. 111-148. Lawrence Erlbaum Associates.Martin, D., Rodden, T., Rouncefield, M., Sommerville, I. & Viller, S. (2001). Finding Patterns In The Fieldwork. Proceedings of ECSCW 2001, Bonn, Dordecht: Kluwer.Martin, D., Rouncefield, M. and Sommerville, I. (2002). Applying Patterns of Cooperative Interactionto Work (Re)Design: E-government and planning. Proceedings of CHI 2002. Minneapolis, Minnesota: ACM press.Noblit, G.W. and Hare, R.Dwight, 1988, Meta-Ethnography: Synthesising Qualitative Studies, London: SagePentland, B. T. (1992). Organizing moves in software support hot lines. Administrative ScienceQuarterly. 37(4). pp527-548Pettersson, M., Randall, D. and Hegelson, B. (2002). Ambiguities, Awareness and Economy: A Study of Emergency Services Work. Proceedings of CSCW 2002. ACM Press. New Orleans, Loiusiana USA.

Plowman, L., Rogers, Y., Ramage, M. (1995) What Are Workplace Studies For? Proceedings of the Fourth European Conference on Computer-Supported Cooperative Work (ECSCW 1995). pp.309-324. London: Kluwer. Purcell, J., Hutchinson, S. and Kinnie, N. (2000) ‘Fun and Surveillance’: The paradox of High Commitment Management in Call Centres, International Journal of Human Resource Management. Vol. 11, No. 2. pp967-985Pycock, J., Palfreyman, K., Allanson, J., Button, G. (1998). Representing Fieldwork and Articulating Requirements Through VR. In Proceedings of ACM CSCW ‘98 Conference on Computer-Supported Cooperative Work. ACM PressRandall, D., Hughes, J. A., and Shapiro, D. (1992) Using Ethnography to Inform Systems Design. Journal of Intelligent Systems, Vol. 4, Nos 1-2Randall, D., Hughes, J. A., O’Brien, J., and Rouncefield, M. (1996) Organizational Memory and CSCW: Supporting the ‘Mavis Phenomenon’. Proceedings of OzCHI 96, Hamilton, New Zealand.Richardson, R. (1994) Organizational and Locational Implications of New Telemediated Services. Mansell, R (ed) Information, Control and Technical Change. ASLIB Schmidt, K., 1997, Of maps and scripts: the status of formal constructs in cooperative work, Proceedings of GROUP 1997, ACM Press.Schon, D. & Bennet, J. (1996). Reflective conversaqtion with materials. In Bringing Design to Software, T. Winograd (Ed). New York: Addison Wesley.Simone, C. & Sarini, M. (2001). Adaptability of Classification Schemes in Cooperation: what does it mean? In Proceeedings of ECSCW 2001. Bonn, Germany: Kluwer Academic Publishers. Star S. L., (1992). The Trojan Door: Organizations, Work, and the 'Open Black Box'. Systems/Practice. Vol. 5, pp395-410Strauss, A. and Corbin, J. (1990) Basics of Qualitative Research: Techniques and Procedures for Developing Grounded Theory. London: SageSuchman, L. (1967) Plans and Situated Actions. Cambridge: CUPSuchman, L. (1995) Making Work Visible. Special issue of Communications of the ACM, Vol. 38(9). pp56-64Trigg, R., Blomberg, J. and Suchman, L. (1999). Moving document collections online: The evolution of a shared repository. In Proceedings of ECSCW'99 Copenhagen, Dordrecht: Kluwer.Twidale, M., Rodden, T. and Sommerville, I. (1993). The Designers Notepad: Supporting and Understanding Cooperative Design. In Proceedings of ECSCW ’93, Milan, pp. 93-108, Dordrecht: Kluwer.Twidale, M., Randall, D. and Bentley, R. (1994). Situated Evaluation for Cooperative Systems. Proceedings of CSCW ’94, Chapel Hill, © ACM Press Viller, S. and Sommerville, I. (1999). Coherence: an approach to representing ethnographic analyses in system design. Human-Computer Interaction. 14. pp. 9-41, Lawrence Erlbaum Associates.Whalen, J. & Vinkhuyzen, E. (1999) Expert Sytems in (Inter)Action: diagnosing document machine problems over the telephone. In Luff, P., Hindmarsh, J. and Heath, C. C. (eds.) (2000) Workplace Studies: Recovering work practice and informing system design. Cambridge: Cambridge University Press.Whalen, J. et al (2002) Improvisational choreography in teleservice work. British Journal of Sociology53 (2); pp. 239–258Wittgenstein, L., (1958). Philosophical Investigations. Oxford: BlackwellWoolgar, S. (1991). Configuring the user: the case of usability trials. In J. Law (Ed.) A Sociology of Monsters: Essays on Power, Technology and Domination. London: Routledge.