Stakeholder, taxonomy, onion model, role, actor, user, surrogate, surrogacy, metaphor, stakeholder identification, viewpoint, requirements elicitation, template

Systems engineers have often paid too little attention to the nature of the so-called "users" of products under development. These are better called Stakeholders, as many roles are involved, and few of those are in direct contact with the developed products.

A simple and robust conceptual framework for classifying development stakeholders – a taxonomy – is proposed. The taxonomy is product-centric, with concentric ‘Circles’ denoting broad categories of stakeholder. Within these, generic ‘Slots’ describe typical classes of stakeholder; these are subdivided into ‘Roles’ which are expected to vary at least in name with the domain. Examples are given, and a popular template is reanalysed using the framework.

The taxonomy has immediate value in identifying and validating stakeholder roles in requirements elicitation, helping to ensure that key viewpoints are not missed, and hence reducing the risk of instability and failure during development.

The structure of stakeholder roles and their relationships such as surrogacy have been very little investigated in the requirements world (though much more extensively in the political, ethical, and information systems worlds: one reason for believing that an attempt at an interdisciplinary look at stakeholders may be worthwhile). Requirements work almost inevitably involves dealing with stakeholders of widely varying kinds, and hence demands a commensurately wide range of elicitation techniques. The first step in identifying which techniques should be applied is therefore to identify the stakeholder composition for a new project; and this in turn demands a suitable taxonomy of stakeholders.

Too many projects focus their attention too closely on the product – perhaps especially when that is software – to the exclusion of non-operational roles, and often even of secondary operational roles such as maintenance. I suspect this is due to ‘inside-out thinking’ where the ‘system’ is seen as important and the ‘user’ as secondary. Such thinking is a hangover from the past. When I was at university, an IBM 360 mainframe occupied the only air-conditioned tower on the campus. Students were permitted to approach only the card-reader with a deck of punched cards; only trained Operators were allowed upstairs to see the Computer itself. This was truly a priestly hierarchy (Greek hieros = holy, arches=ruler) of Operator Roles. As Christopher Locke writes, "Even the word ‘users’ is an artefact of the [command-and-control] mentality." (Levine et al 2000). It is time to move on from treating "the user as a computer peripheral" (in Julian Hilton’s words). The System is made for Man, not Man for the System.

Many industrial development problems seem in practice to be caused not so much by a failure to write requirements as by a failure to perceive that specific stakeholders’ viewpoints were relevant. That failure causes whole groups of requirements, typically those related to scenarios involving the missing stakeholders, to be missed.

A similarly unhappy result is obtained when one stakeholder, e.g. a software developer, assumes one scope for a product, while another stakeholder, e.g. a purchaser, assumes another. For instance, when a developer assumes that it will be sufficient to design, code, and test a piece of software, but the purchaser hopes to have everything set up and the operators trained, then the points of view of the installer, the trainer and to some extent that of the operators have not been adequately considered and made explicit. Legal disputes and financial losses are then likely.

It seems likely that stakeholder composition is a good predictor of project risk; hence, it should be cost-effective to characterise projects at their initiation according to their likely stakeholder impact (and to other variables, such as safety-relatedness, technological innovation, similarity to previous projects, and so on).

In addition, maintaining a model of stakeholders throughout a development allows changes in stakeholder composition to be modelled explicitly, leading to appropriate changes in requirements.

Stakeholder surrogacy has powerful and paradoxical connotations in requirements engineering. It is almost a dogma that projects should seek out ever-closer dialogue with stakeholders – consider the current fashion for Integrated Project Teams, facilitated workshops, rapid prototyping, agile development with user stories, etc. Yet all the time the obvious truth is glossed over: that it is remarkably rare to be able to talk to many stakeholders in the flesh. Every requirements engineer knows that the basic answer to the client organisation’s boss who says ‘I know everything that happens in my department, ask me’ is ‘well, sir, that’s fine but can I see if the people on the shop floor know of any small issues?’. To put it more formally, standardised procedures, no matter how critical and how carefully defined in writing, are always modified when operationalized ‘on the shop floor’. Therefore, it is essential to talk to stakeholders directly – without intermediaries – to find out what actually happens. Yet, requirements engineers are themselves intermediaries! Stakeholder surrogacy is accordingly discussed at some length below (section 2.6).

Worse, many kinds of stakeholder are inaccessible: they may be distant geographically; separated by contractual and procedural barriers; hidden within organisations (with cultural barriers); simply unaffordably expensive to contact given scarce project time and resources; or not yet in existence (for future products).

The naïve "go and talk to the users" – whoever may be meant by that phrase – is therefore far from helpful as advice. This paper considers what we mean by stakeholder roles on development projects, and offers both a theoretical framework for classifying them and some practical suggestions for making use of that knowledge.

It may be that the approach can be applied outside system development, e.g. to model political and business stakeholder structures, but that is beyond our scope.

The remainder of this paper is structured into sections as follows:

• Research Review
• The Proposed Taxonomy
• Applying the Taxonomy in Practice
• Discussion: Features & Limitations of the Taxonomy
• Conclusion

The focus of this paper is on the place of stakeholder analysis in system development, and perhaps especially in requirements elicitation. However, the subject of stakeholders is far wider than that. This section therefore attempts to set the current work in its system engineering context, and briefly also in the wider onion-circles of information systems research and business studies. These literatures are voluminous, so references are confined to key papers on each topic.

This research review section is organised by topic as follows:

• Stakeholder Analysis
• Onion Models
• Taxonomies and Hierarchies
• Goals and Viewpoints
• Human Aspects of System Development
• Surrogacy

The logic of this should become apparent, but in essence the information systems and other literatures on stakeholders are briefly introduced; the background to the paper’s use of onion models and a taxonomy of stakeholders is sketched; existing work related to stakeholder classification in the requirements engineering and usability fields is examined; and finally, since stakeholder surrogacy seems to be significant in system development, research on surrogacy is explored.

An authoritative account of business stakeholders can be found in (Donaldson & Preston 1995). The paper describes a corporation ‘as a constellation of co-operative and competitive interests possessing intrinsic value’ (page 66). It states that the stakeholder theory is descriptive of the corporation, instrumental in helping people to examine stakeholder management practices, normative in establishing that stakeholders deserve attention, and managerial in recommending attitudes, structures, and practices. However it considers the corporation as the system of interest, and does not look at software or product development.

(Pouloudi 1999) examines the concept of Stakeholder and its use in information systems development. The paper is an excellent introduction to the extensive information systems literature on Stakeholders, and it analyses some of the weaknesses of the concept (including being ‘almost a cliché’, quoting from D. Willets). For our purposes here, one key point is that it does not make sense to treat human and non-human actors symmetrically: they are simply different. She notes the discomfort of Vidgen and McMaster ‘about assigning anthropomorphic properties to non-human resources’ and states that

In the taxonomy presented here, all stakeholders are human; interfacing systems are represented by humans responsible for them, i.e. in accordance with Pouloudi’s approach. Pouloudi has also studied stakeholder identification, e.g. in a medical context (Pouloudi & Whitley 1997). This is a very different context from that of system development, but the paper is of interest for its clear thinking and practical approach.

(Sharp et al 1999) suggested a simple recursive procedure, starting from an initial contact person, and asking each interviewed person who might be worth speaking to. The procedure terminates when no new names arise or when the new names are found not to be relevant. The paper described a simple method for identifying stakeholders starting from the initial point of contact (which might typically be the project Sponsor). This is one of the few papers to address the issue directly, and its suggestion is sensible. However it is not a substitute for a template (based on a taxonomy); the method could be time-consuming, and it is likely to reveal only the stakeholders that everybody knows already. To be fair, the method of asking each interviewee who else might be relevant is almost an obvious good practice during requirements elicitation. The paper also pointed out that stakeholder classes (slots) should be characterised by their relationships to the system, to other stakeholders, and by the priority to be given to each stakeholder’s view. These dimensions are used here, though they are rough guides at best (along with the time that each slot might be relevant).

(Robertson & Robertson 1999) distinguishes clients, customer, and other stakeholders including subject matter experts, marketing people, product managers and so on (pp356-358) and attempts to prioritise them: ‘the principal stakeholders are the users, clients, and customers’ (p35). However the general effect is of a flat unstructured list of many interested parties. The Volere template (by the same authors) similarly proposes a flat list of roles, as one axis of a matrix to organise which kind of stakeholder to consult for which ‘class of knowledge’ (Volere 2004). The taxonomy proposed in this paper offers a richer and more explanatory structure, and suggests the significance of key roles, and their interactions.

(Alexander & Stevens 2002) contains a chapter on identifying stakeholders (pp 19-26). It also distinguishes ‘users’ from clients, suppliers, managers ‘who are concerned for the system to succeed’ (and by implication other people with that concern), and regulators, and it briefly discusses the role of people in the development organization (p7). The variety of roles is illustrated with a cartoon of stakeholders in a space telescope project (p20) which shows an astronaut carrying out maintenance, a ground station engineer operating the spacecraft, an astronomer discussing the data produced by the telescope, and a politician gaining political benefit from the system.

The discussion of these roles makes clear that these represent different viewpoints, not necessarily conflicting. This was a sensible and pragmatic position, but the proposed taxonomy looks far more deeply at the structure and possible conflicts between stakeholders.

There is much pragmatic wisdom on dealing with stakeholders in Peter Checkland’s many writings, e.g. (Checkland & Scholes 1990). He does not on the whole focus on the development of products or services, and indeed he seems to believe that that is the easy part of the problem! For example:

While we may readily agree with Checkland’s emphasis on process rather than project, and take his advice on (human) ‘perceptions and meanings’ and ‘rethinking’ as a hint to keep up to date with our Stakeholder Analysis, we don’t think that software only implements uncontentious mechanical procedures. Indeed, major products such as Enterprise Resource Planning tools can have a powerful and sometimes deleterious impact on organisations (note that we consider the customisation of such COTS tools for an organisation to be a kind of development project).

Another worker – both researcher and practitioner – who helped to pioneer socio-technical design is Enid Mumford. In her long career and her many writings (e.g. Mumford 1996) she emphasised a humanistic and ethical approach to the people whose work was affected by process redesign and automation. She talks more often about specific people, individuals, clerks, groups, staff, and so on, rather than using generalities like ‘stakeholders’. However, much of her work was precisely about paying more attention to the different groups of people involved in or affected by a project. For example:

Onion models have been used for centuries to indicate hierarchical spheres of influence. Alexandre Koyré’s wonderful From the Closed World to the Infinite Universe (Koyré 1957) uses the beautiful 11-layered onion diagram of Peter Apian’s 1539 Cosmographia, a pre-Copernican model of the universe, on its cover.

Apian has the imperfect and changeable Earth at the centre, and Coelum Empireum Habitaculum Dei et Omnium Electorum (i.e. The Empyrean heavens, the dwelling place of God and all the Elect) as the outermost, perfectly unchangeable layer. This implied a fixed frame of reference for each class of beings in the Great Chain (Lovejoy 1936), very far from the dynamic situation-dependent stakeholder webs of (Coakes & Elliman 1999) who describe the role of stakeholders in managing change. Their term ‘Stakeholder webs’ has entered popular currency, perhaps by association with fashionable terms like world-wide web and semantic web, and may have done much to get people thinking about stakeholders in software development.

A Stakeholder Web (from Coakes & Elliman 1999)

Their webs are drawn as cobwebs with radial lines and concentric ellipses. This makes them look something like onion models, but:

Thus the Coakes & Elliman stakeholder web is expected to be different for each examined system: there is no recognisable taxonomic pattern common to different developments. Somewhat in contradiction to this, Figure 1 of their paper (below) illustrates the choice of the system boundary, based on (Midgley 1992), which shows concentric circles labelled the Technical System Boundary, the Organisational Boundary outside that, and finally the Human or Total System Boundary on the outside. Within the Technical System Boundary are three items linked by bidirectional arrows: Computer Information System, whose boundary is labelled Automation Boundary; Direct System Users; and System Designers. Within the Organisational Boundary is a text label "Executive and wider Management, Other Divisions and business activities". Within the Human or Total System Boundary is a text label "Shareholders, Clients, Government, and other Stakeholders, beyond the organisational boundary" (page 6).

(from Coakes & Elliman 1999)

This is clearly a rudimentary onion-model. We can make the following assignments:

• Computer Information System = (an example of) our The Kit or Product;
• Direct System Users = our Normal Operators (but see criticism of the term ‘users’ below);
• System Designers = our Developers, who seem oddly placed within the Technical System, given that they may have no part in the system’s daily operations. It looks as if the operational boundary is here confounded with the boundary of all matters engineering and technical. The Figure is criticised by both Midgley and Coakes & Elliman, who argue that the ‘critical setting of the system boundary’ needs to be ‘determined by examining the viewpoints of stakeholder groups … rather than technical issues’ (page 6). However, the automation boundary remains important as a contractual and developmental reality; while the Figure is clearly defective as regards the Technical System Boundary, it can be corrected by removing the System Designers from it, and inserting any other types of Operator that may be needed, as is argued below.
• Similarly, Executives and Management may loosely correspond to our Functional, Financial, or Political Beneficiaries, but there is no reason why the Organisational boundary should fit neatly within the Total System Boundary if this is taken to mean our Wider Environment with respect to a specific Kit or Product: some parts of the organisation’s management may be involved, others not. Simply because some dead-wood paper-pushers are within the organisation does not make them valid stakeholders: that way madness lies.
• Again, Shareholders, Clients, Government can reasonably be equated with our Financial Beneficiaries, Purchaser (perhaps – client is a slippery term), and Political Beneficiary (or perhaps Regulator, depending on the situation).

(Coakes & Elliman 1999) are probably right, therefore, to reject the (Midgley 1992) onion-model, but a more carefully-considered model may overcome their objections. Designers and developers do not belong inside the "technical system" boundary – in any case, since it contains people, it must be socio-technical. The organisational boundary is the wrong one to choose (unless the organisation is the focus of study, rather than a product); and the "human or total system" boundary does seem rather vague. On the other hand, there is usually a Containing System (in our terms, see section 3.1) that makes use of the results or outputs of The (socio-technical) System that we are developing – otherwise, why would anyone pay to develop it – and there are indeed stakeholders such as shareholders and government who may have an interest in aspects of the system, while not themselves playing any part in its development or operation. The problem with a stakeholder web formed afresh for each new system is that it offers little guidance, whereas the engineers working on yet another telemetry system (say) are immediately aware that what they are doing is very similar to what they did on previous projects. Coakes & Elliman throw out the baby with the bathwater.

A modern onion model related to stakeholders can be found in (Donaldson & Preston 1995, page 74).

However their model’s circles, starting from the centre, are the Normative, Instrumental, and Descriptive aspects of stakeholder theory. These

The Donalson & Preston onion model is thus used to help visualise the structure of the theory, not the relationships of the stakeholders and a product under development.

(Hirschheim & Klein 2003) reflect on the state of information systems research in a long and discursive paper. In a nutshell, they wonder whether information systems can effectively mediate social dialogue, and if so how the field would have to be structured. Understanding organizational stakeholders better would form an important part of that programme. The authors refer to ‘immediate’, ‘external’, and ‘societal’ stakeholders, implying something like an onion-structure, though not exactly the one described here. ‘Immediate’ roles, for instance, include ‘a managerial elite and their masters, the shareholders’ (page 271) – though it is not clear if in fact the shareholders are immediate or one jump removed by being the masters of the managerial elite. If the latter, then there might be a reasonable match with the onion-model given here: managers = functional beneficiaries, shareholders = financial beneficiaries. However, the focus of the paper is not on products but on information systems professionals themselves.

The general subjects of taxonomy and hierarchy are far too broad to be addressed here in any detail. Susan Leigh Star has written engagingly about classification (e.g. Bowker & Leigh Star 1999). She states that classification systems should exhibit consistent principles, mutually-exclusive categories, and completeness. It is hard to make any such classification of stakeholders; while the roles listed here seem to be usable consistently, and are mutually-exclusive, one person can play several roles, and it is impossible to be sure that new roles will not arise, so completeness is unattainable. Any ‘taxonomy of stakeholders’ must be tentative at best.

Under the heading of ‘Hierarchical Structure’ (Jackson 1995, pages 92-95) wittily makes the point that one taxonomy can only do one job; very often several different hierarchies are needed. Indeed one of the key principles of Jackson’s JSD / JSP methods is to make sure that your representations adequately span the problem. You may need one data structure that understands files: file start - data - file end; and another that understands transactions recorded within those files: start event - more events - end event. Either on its own is insufficient: a transaction might, for instance, span two files. Analysing stakeholders by their relatedness to a product is only part of the story.

Two areas of requirements engineering related to Stakeholder analysis are Goal Modelling and Viewpoint Analysis.

A Goal is an intention that a stakeholder has for a development project (or more widely for a business, or indeed for a lifetime), e.g. that it will make preparing accounts easier. Goals may interact positively or negatively – achieving one goal may make it easier or harder to achieve another. Such interactions can easily be modelled graphically, making conflicts easy to visualise. Hence, modelling goals and their interactions is one way to explore the relationships between stakeholders with different needs. What goal modelling is not designed to do is to identify or classify the stakeholders themselves, though it can explore their social and operational interactions (and even their beliefs). Goal modelling is therefore, like most other approaches, likely to discover operational (‘Our System’) roles rather than those in the wider Circles of the onion model. There is an extensive literature on Goals; a good starting point is (Yu & Mylopoulos 1998).

Negative interactions can also be intentional and dynamic, in which case they may more directly be modelled as negative scenarios or Misuse Cases (Alexander 2002). This approach has the merit of considering threats, risks, and hazards, whether caused intentionally or not, and how to mitigate them. It will therefore put the spotlight on Negative Stakeholders as well as Normal Operators. However this approach too tends to direct attention mainly to operational issues.

A Viewpoint is either the perspective that a given stakeholder has on a development project or its product, e.g. that they will continue to have an interesting job and a decent salary, or the projection (e.g. of requirements) on to the system from the position of a given stakeholder role, e.g. an employee's. Analysing viewpoints helps to discover whole groups of requirements, can indicate possible conflicts, and hence can help to create better and more stable specifications.

A readable introduction to Viewpoints is (Kotonya & Sommerville 1998). It describes a method of ‘Viewpoint-Oriented Requirements Development’ (VORD), explicitly a stakeholder-centred approach. As well as describing a method for discovering viewpoints and resolving stakeholder conflicts, it presents a simple taxonomy of ‘abstract viewpoint classes’ (p219), classifying Viewpoints into Direct and Indirect. These do not correspond exactly to our Operational and Non-Operational. Instead, Direct includes ‘System’ (our ‘Interfacing Systems’ – see section 3.1 ‘Structure of the Onion Model’ below) and ‘Operator’ (our ‘Normal Operator’). Indirect is divided into Engineering (Maintenance and Standards: we consider Maintenance to be Operational, while Standards means the Regulator), Regulatory, Organisation (Procurer, Policy, and Training) and Environment. A further paper by the same team (Sommerville, Sawyer & Viller 1998) describes the PreView approach in which viewpoints are treated as projections on to a system. One stakeholder viewpoint then naturally corresponds to a substantial set of requirements.

Multiview (Avison and Wood-Harper, 1990) is a well-respected approach to system design based on the idea of multiple stakeholders' views on the system, and on choosing appropriate tools and techniques according to the problem situation thus defined – as such it owes something to Soft Systems Methodology (Checkland & Scholes, 1999). Both ‘soft’ and ‘hard’ aspects are considered: human exploration and socio-technical analysis on the soft side; information analysis and specification on the hard side. This is obviously sensible and practical. However it does not attempt to classify stakeholder roles. Multiview 2 went further in several ways, including a ‘systemic stakeholder analysis’ within its organisational analysis, and using ethnography in the socio-technical analysis (Bennetts and Wood-Harper 2000). However it did not attempt a taxonomy of stakeholders.

There is an extensive literature on human aspects of system development. It seems however to focus quite naturally on the ‘user’ (by which is generally meant our ‘Normal Operator’) at the expense of all other roles. This is no place for a full literature survey, but a good starting-point is (Sutcliffe 2002) which approaches requirements engineering from a background in human-computer interaction and has an extensive bibliography.

Sutcliffe distinguishes customers, users, managers, software engineers, system testers and system maintainers – ‘Maintenance personnel are rarely consulted in requirements analysis yet they depend on accurate requirements documentation more than most’ (p17) – and it could be added, they have more to contribute than most, too. These roles are notably concentrated around ‘Our System’, but to be fair they are said to be those that can write requirements. Stakeholders are further classified (p56-7) as:

• primary ‘who will actually operate the system’ (our ‘Normal Operators’ (and possibly also our ‘Maintenance Operators’): equivalent to UML ‘Actors’);
• secondary ‘users who will not actually operate the system but will consume its output and depend on its operation for successful completion of their work’. The Astronomer of (Alexander & Stevens 2002) (see above) would be an example (our ‘Functional Beneficiary’). Note the typical awkwardness and circumlocution necessitated by overloading the term ‘users’ to mean both operator and beneficiary;
• tertiary ‘senior managers who rarely consume the system output directly but make use of information for planning and strategic control of the business’. This seems to denote either our ‘Functional Beneficiary’ or our ‘Political Beneficiary’.

Sutcliffe’s classification is interesting but limited to thinking about people who receive information outputs. However the book also describes more general stakeholder analysis methods, somewhat surprisingly quoting the ruggedly practical and pioneering (Gause & Weinberg 1989) rather than the more academic (Kotonya & Sommerville 1998) which one might have expected.

Surrogacy has apparently scarcely been researched as a requirements engineering issue. It is mentioned briefly in some of our own earlier work on Stakeholders (Alexander 2003) and in a practical way on the Scenario Plus Stakeholders template (Scenario Plus 2004).

(Damian & Zowghi 2002) state that a Business Management department acted as a surrogate stakeholder:

The point is not developed further, but it is clear that in the context of using technology to support RE across sites on different continents (America, Australia, Europe) the issue of who you can actually speak to, and whether they fairly represent who they claim to, is highly significant.

A set of guidelines (Kitapci & Bhuta 2003) for using the EasyWinWin (requirements negotiation) tool and approach (Boehm et al 2003) tantalisingly mentions appointing ‘one of the team members as a surrogate customer if the customer would not be using the EasyWinWin tool.’ However the significance of surrogacy is not considered.

Surrogacy is also occasionally mentioned in non-requirements work on stakeholders (i.e. to do with steady-state business, not with system development). For instance, (Wood 1999) writes

Similarly, (Younkins 2001) expresses the sentiment

The focus of such work is however (in both these examples) on the political and social significance of stakeholder responsibilities, not on surrogacy as such.

(Donaldson & Preston 1995) begin, interestingly, with a quotation from E. Merrick Dodd, Jr, writing in the Harvard Law Review of 1932. Dodd does not use the term stakeholder, but he is plainly thinking of the same concept: ‘If the unity of the corporate body is real, then there is reality and not simply legal fiction in the proposition that the managers of the unit are fiduciaries for it and not merely for its individual members, that they are … trustees for an institution [with multiple constituents] rather than attorneys for the stockholders.’ Both the concept of stakeholder and the importance of surrogacy are evident from this early discussion.

(Potts 1995) does not explicitly mention surrogacy, but does consider the relationship of ‘customer’ and ‘developer’ and some of the misunderstandings that can arise across the gap between them; e.g. the contractual interface and hence the supplier/purchaser roles that are created (with inevitable surrogacy). His talk’s provocative title also led requirements people to reflect on the nature of the stakeholder/developer ‘interface’.

A project’s stakeholder sociology can be modelled graphically on an Onion Diagram (Figure 1). This deceptively simple-looking model documents a wealth of information about a project. It presents a view of the project that is centred on its product, and serves as an overview of our stakeholder taxonomy.

This section introduces and defines the terms used in the Onion Model and the associated Taxonomy of Stakeholders. Terms are highlighted in Boldface when they are introduced and defined.

The Onion Model consists by default of a set of three concentric Circles. Each Circle logically contains the Circles drawn inside it. (The term ‘Annulus’ may be used where it is desired to refer to the contents of a Circle without those of the Circles it contains.) Other Circles may be added, most likely by subdividing the default Circles.

Hence, a ‘Circle’ denotes a subset of entities in the world relevant to a development project. Those entities are Slots and Circles.

The innermost ring (not strictly a Circle) denotes the 'Kit', a conveniently short term for the Equipment or Product under development, whatever its nature – for instance software or electronics hardware. This could be a one-off piece of custom financial software, or a mass-market product such as a handheld information device. It could equally well be the tangible equipment for a large system such as a railway line: the system with its many human operators would of course be more than just that equipment. The innermost ring thus differs from true Circles in not containing Stakeholders.

A ‘Slot’ is a class of Roles that is drawn by default (empty if necessary) in an Onion Diagram. In effect it is a prediction or suggestion that Roles of a certain kind may be important in a project. For example, ‘Normal Operator’ is a default Slot in the ‘Our System’ Circle. The existence of the slot is a prediction that in any system, there will probably be one or more ‘Normal Operator’ Roles. The slot might remain empty in a supposedly 'wholly automatic' system, though such a system might still have other human roles such as Installer, Configurer, and Maintainer within the 'Our System' circle.

A Slot is said to be ‘Empty’ in an Onion Model when it contains no Roles. For example, take a Product like a distress signal flare. This product is fired by a sailor to attract rescuers; whether it succeeds or fails it is then disposed of. No maintenance work of any kind is carried out on it, so it has no Maintenance Operator and that Slot is Empty.

A ‘Role’ is a class of Stakeholder with a distinct relationship to the Product under development. For example, in a passenger aircraft, both Pilot and Flight Engineer are Roles within the ‘Normal Operator’ Slot.

A Role is said to be a ‘Surrogate’ if it is conducted on behalf of another Role which cannot speak directly for itself. For example, a Product Manager acts as the purchaser of a new type of Product until such time as the Product can be sold on the mass market; then the Consumers buy instances of the Product (or not) for themselves. As another example, the Regulator acts as a government-sponsored voice on behalf of the public to ensure safety, etc. Surrogacy is discussed in more detail below.

A Role is said to be ‘Negative’ if it is associated with Viewpoints opposed to the successful completion of the Product’s development or its coming into service, or its successful operation. Viewpoints – statements of the points of view of stakeholders in a Role with respect to the Product – do not appear in the Onion Model as such, but can be documented briefly as text attributes of Role objects in an associated database (Figure 2), or in full as sets of requirements (not shown) linked to Role objects. It may be useful to document generalized Viewpoints as attributes of Slot objects, summarizing a group of more specific Role Viewpoints. Tool support is discussed below (section 4.4).

A ‘Stakeholder’ is an individual person or other legal entity able to act like a person (e.g. a Limited Company, an Industry Regulator, a Registered Charity) playing one or more Roles. Note however that in accordance with common usage we loosely describe both Slots and Roles as ‘stakeholders’ when the class/instance distinction is not of immediate concern.

It would also be possible to limit the definition to ‘individual person’; this would have the advantage of avoiding possible confusion between ‘Company XYZ’ as a stakeholder and ‘The Company XYZ Containing System’ as a Circle. Further, while a suitably-placed individual (e.g. the Chairman, the Press Relations Officer) can legitimately represent the voice of a Company as a corporate legal entity, that individual cannot legitimately claim to speak for each of the many Roles within the Company as far as their viewpoints and requirements are concerned. There are thus some dangers associated with treating legal entities as Stakeholders; but there are obvious advantages. For instance, if a conservation body chooses to oppose a development, its Negative Stakeholding can reasonably be treated as a single voice, in accordance with its legal status.

The four default Circles used in the Onion Model are:

1. ‘The Product’ or ‘The Kit’: the item under development, e.g. a software program, a consumer electronics device, an aircraft, a communications network.
2. ‘Our System’: ‘The Product’ plus its human Operators and the standard operating procedures or rules governing its operation.
3. ‘The Containing System’: ‘Our System’ plus any human Beneficiaries of Our System (whether they are involved in operations or not).
4. ‘The Wider Environment’: ‘The Containing System’ plus any other Stakeholders.

Other Circles may be introduced as necessary. For example, an additional Circle might be created outside ‘The Containing System’ to divide Stakeholders relatively closely involved with the project from those more distantly involved in the ‘Wider Environment’. That might be helpful when a product forms a component of a subsystem within a larger system. E.g. when our product is a control box for a train (the Containing System), we might wish to consider a ‘Wider System’ Circle, namely a railway line with stakeholder roles such as line controller, and even a ‘System Environment’ Circle, namely a complete railway network with roles such as network timetable scheduler – as distinct from the ‘Wider Environment’ which might include stakeholder roles such as the general public and the government.

The Kit or Product Circle is considered to be the part of Our System that can be sold, so it does not contain humans. All the other Circles contain Stakeholders. This does raise an interesting problem of consistency: since it is a purely relative matter which system in the real world we consider to be ‘Our System’, it is perfectly possible for somebody else to treat Our System as a component part of their Product. What is happening here, however, is that they are focussing attention only on the machine (non-human) aspects of Our System, for the purposes of product development. Of course, the Stakeholders do not disappear, but they will either be treated as insignificant from the (larger) viewpoint, or they will be promoted to the rank of Stakeholder in the larger Product.

For example, if the larger Product is a train, its Normal Operator is a driver. Our component Product, the train’s control box, is also operated by the driver, who only appears once in the model – outside the Product, the train, as one would expect. The fact that the driver also operates a sub-product of the train (the control box) does not make the driver a part of the Product. Thus, at the level of detail of the larger viewpoint, there is no need to consider separate sub-product Stakeholders. In general, therefore, as long as one considers just one ‘Our System’ at a time, there is no problem of consistency. A shift of viewpoint – to consider a larger or smaller Product – rightly causes a change in the perspective of the Onion Model.

One might ask whether distance from ‘the Kit’ means less influence on the requirements for it: this would be a simple application of the spatial metaphor of the Onion Diagram. (See section 5.4 for a fuller discussion of Visual Metaphor.) However, there seems no reason to believe this. On the contrary, the approach here emphasises that stakeholders who never see the kit may be crucially important to a development project. As (Coakes & Ellison 1999) say of their stakeholder web diagrams:

A well-known example is the occasion where English Heritage stopped a rail resignalling project the day before it was to launch as an attractive red-brick Victorian railway viaduct would have been demolished. No knowledge of signalling or interest in it was necessary. It may be that different stakeholder ‘slots’ come with presumptions about their likely (default) importance, but that is a weaker claim.

The Onion Diagram by default displays the following ‘Slot’ icons in the named Circles. Other Slots may be added as necessary. In each case are defined, based broadly on the framework of (Sharp et al 1999):

• the nature of the Roles in the Slot with respect to the Kit;
• the most likely interactions with Stakeholders in other Slots;
• the stage(s) in development that the Slot is most likely to be relevant, and
• the likely priority (relative importance) of Stakeholders in the Slot.

Note that the following are what are considered to be the bare minimum of Slots, not an exhaustive taxonomy.

Normal Operator: roles that involve giving routine commands and monitoring outputs from the product, whether these are via a human-computer interface or not.

Normal Operators interact directly with the product, with other Operators (e.g. Maintenance, Operational Support) and with Functional Beneficiaries (e.g. providing them with processed information, and receiving instructions from them).

Operator requirements are relevant throughout development, but especially during user interface design.

Operability requirements are always important, perhaps especially so for mass-market products (where operability is a selling-point) and for control systems (where safety is involved).

Maintenance Operator: roles that involve maintaining the product, such as servicing hardware, and diagnosing and fixing faults. (So-called maintenance of software involves changing the design of the product, and is the responsibility of our Developer Slot; it is not maintenance in our sense.)

Maintenance Operators interact with the product and with Normal Operators.

It is worth noting that now that outsourcing is in fashion, maintenance people may be contractors. If so, there is a contractual boundary that crosses the ‘Our System’ circle (and others outside it). This can easily act as a "geological fault-line", impeding communications and possibly rupturing if anything goes wrong. For example, some organisations outsource their computer and network administration (Maintenance Operations), while retaining Normal Operation of the equipment in-house. This typically creates bureaucratic delay and difficulty, which in turn creates tension between the two roles. It may be helpful to draw contractual or other boundaries on the Onion Diagram and to annotate these with their significance for the project.

Maintenance is often not considered till late in projects; however, maintainability needs to be designed-in, whether as built-in test and diagnostics, internationalised error messages, accessibility of equipment, or spares holdings. Requirements for these need to be in place early in a development.

Maintenance requirements are often more important than they seem. The whole-life cost of many products – from cars to jet engines – depends much more on the cost of maintenance than may be realised. Similarly, maintainability requirements such as the time to repair or replace components can significantly affect the quality of service.

Operational Support: roles that involve advising Normal Operators of a Product about how to operate it. These roles are very close to operations but support rather than conduct productive use of the Product itself. We have chosen to include them in ‘Our System’ for two reasons:

1. they behave as operational staff in their daily work;
2. like maintenance operators, they help to keep the System fully operational (enabling the Normal Operators to continue working effectively).

Operational support people such as Help Desk staff and Trainers interact mainly with Normal Operators. They are ‘maintenance’ for the humans involved, rather than just for the product. The comments on outsourcing under ‘Maintenance Operator’ can also apply here.

The priority of support probably deserves to be higher than it is on many projects. As with maintenance, good support raises operational effectiveness and availability. Clearly it is secondary to other slots such as normal operator and functional beneficiary.

Support needs to be considered when preparing manuals and training materials, i.e. relatively late in the project; however, supportability may also need to be evaluated earlier e.g. when designing software to yield intelligible error messages, etc.

Functional Beneficiary: roles that benefit from the results or outputs created by the Product. For example an astronomer benefits from the astronomic data captured by a space telescope though he or she cannot operate the instrument directly. Since Products are or should be designed to produce results, this is an important Slot.

They interact with Operators, giving them instructions and receiving information and any other benefits that ‘Our System’ is designed to provide.

Functional requirements form the centrepiece of most specifications. They need to be available early and are used throughout development, e.g. in design and for specifying functional tests. They are of the highest importance.

(Responsible for) Interfacing System: roles responsible for neighbouring systems that have electronic or other interfaces to/from the Product. Such systems behave much like human Operators in terms of demanding specific capabilities from the Product, but naturally the interfaces are precisely defined as protocols, etc.

They interact with Operators and/or the Product; any other interactions they may have are most likely but not necessarily outside our scope.

Interface requirements form a crucial part of the definition of many developments. They are required from the start. Shifting interfaces and scope are serious risks to projects.

Purchaser: roles responsible for having the Product developed. There are certainly several of these, ranging from Product Manager (with knowledge of what can be sold) to Procurement (responsible for obtaining a contract with a supplier). In the case of a mass-market Product, the Purchaser is a Product Manager – a Surrogate role, acting on behalf of millions of Consumers who will if all goes well ultimately buy the Product. Purchasers interact with Developers and Consultants, and (to obtain requirements) with Beneficiaries and Marketing also.

The purchaser’s input is required from the start and is critical in getting development started: and in ensuring that funding for it is not withdrawn. It always has high priority (possibly more than it deserves, but that is out of the control of those working on a project).

Product Champion (aka ‘Sponsor’): role responsible for initiating development of the Product, for obtaining funding for it, and for protecting the development from ‘political’ pressures and funding cuts. The role requires positional power within the purchasing organisation (e.g. the company creating a mass-market product). The Product Champion is perhaps the best person for the Requirements Engineer to meet with first; an effective champion can indicate the scope and purpose of the development, the opportunities and threats, and can suggest who the key stakeholders are. All of this helps to cut the risk to the project.

A Product Champion’s effectiveness is clearly related to the ability to interact with other stakeholders, especially Beneficiaries and Negative stakeholders (including those within the organisation).

The product champion is critical from before the start of a development, and remains important throughout. The role does not necessarily or even desirably contribute to product requirements: it functions mainly at a political rather than a technical level.

Negative Stakeholder: any role that could be harmed by the product physically, financially or in any other way that might be found justifiable by the authorities (e.g. a court of law, a Regulator) , or conversely that could attempt to harm the product. For example, householders living close to the route of a planned railway; a nature conservation body with interest in land threatened by such a route; activists opposed to pollution that might be caused by a product under development, employees finding their decision-making abilities reduced by ‘intelligent’ software, employees that perceive their tasks being oversimplified or made too complex, groups feeling that collaboration or communication were made more difficult.

A special kind of Negative Stakeholder (and possibly a distinct Slot) is the Hostile Agent: any role that actively seeks to hinder or harm the development and operation of the System. 'Actively' means using some degree of intelligence and creativity to oppose the System. Examples include military enemies, political and commercial spies, hackers, spammers, virus writers, thieves, fraudsters. Clearly the degree of harm intended by such agents varies from complete destruction through malicious pleasure to unauthorised acquisition of assets (with essentially unintended harm as a side-effect).

Negative stakeholders may interact with Regulators, with Beneficiaries, and with any roles in the Wider Environment able to wield influence, e.g. the press, politicians, other pressure groups.

Hostile roles can be treated as a kind of negative stakeholder, or may perhaps be promoted to a slot in their own right. Hackers and virus writers are obvious hostile roles. Competitors, too, could be considered: their relationship could be anything from passive victim to active threat. In the case of military systems, the enemy’s expected behaviour and capabilities are naturally a primary consideration.

Negative roles are important from before the start of a development, and then whenever issues like security, marketability, and environmental impact are considered. These affect requirements and design. Security may demand special effort in system testing.

Political Beneficiary: any role in public office or private business that can benefit in terms of power, influence and prestige through the success of the Product. E.g. a space agency’s management could benefit ‘politically’ from a successful space mission.

Interaction with other roles is usually infrequent and is typically indirect, e.g. via senior management (Purchaser, Functional Beneficiary, etc.).

Political forces within an organisation can also be Negative (see the discussion of Product Champion above). It might be worth representing ‘Political Opponent’ though such explicitness can be dangerous, as Checkland mentions (Checkland & Scholes 1990); ‘There is an unavoidable political dimension .. to any human affairs which entail taking deliberate action’ (p50); ‘delicate judgements are usually required’ (p51); ‘if the results (of political analysis) are all bluntly made public, then those results can themselves easily become a potent commodity of power in the ‘real’ politics of the situation. There is potentially an infinite regress here in which the politics of the situation forever escapes open analysis’ (p51).

Political roles are important (see Sponsor, above) throughout a development and indeed from before it begins.

Financial Beneficiary: any role that can benefit financially from the success of a Product. For example, shareholders and directors in a company making a mass-market Product.

They often interact weakly with other roles, except perhaps with the Product Champion and other senior Beneficiaries.

Development staff perhaps need to consider financial beneficiaries directly only rarely; project and programme management may be more concerned. Perhaps the most usual situation is that financial beneficiaries are represented by surrogate roles such as line management. Conversely, financial beneficiaries are likely to take a direct interest only in the largest of developments. They are important at the main ‘gates’ in a development.

Regulator: any role responsible for regulating the quality, safety, cost or other aspects of the Product. For example, aviation authorities, health and safety executives, rail regulators, radio regulators, financial service authorities.

The Regulator is most likely to interact with senior Developers. Regulators act as surrogates for the public, interacting with Developer and Beneficiary roles as necessary; e.g. the aviation authorities certify components on receipt of a satisfactory safety case supported by evidence from the development organisation’s safety officer.

In the case of a software product we could view standards organisations like ISO as non-statutory (voluntary, not enforcing) regulators.

Regulators impose requirements which act as qualities and constraints (rarely as functions). These are important in defining the requirements for a product, and again for acceptance and certification. In the case of safety-related products they are of crucial importance.

Developer: any of the many roles (requirements engineer, analyst, designer, programmer, tester, safety engineer, security engineer, electronics engineer, metallurgist, human factors engineer, project manager, etc) involved directly in Product development. Note that none of these roles are operational unless tied into operations via a maintenance contract – in which case the affected people have hybrid Developer / Maintenance roles.

Developers interact mainly with each other (often adopting Surrogate ‘customer’ roles in the process – see discussion below), but also contractually with Purchasers. Ideally, Developers in the form of requirements engineers and analysts would interact with all other roles, but this is rare and often impossible for reasons of e.g. time and restrictions on access.

Developer roles are mainly involved once development has started; it may be helpful to involve a developability opinion (a Consultancy role) before that, and to keep developers involved (via maintenance) during operations. Clearly the priority of a developer’s requirements is secondary to those of the beneficiaries, though manufacturability is an important consideration for mechanical products such as jet engines.

Suzanne Robertson suggests (Robertson 2004) that it may be helpful to create an extra circle for ‘development project responsibility’. It may be best for this not to be a concentric circle – in a way, development is another world from operational usage. Developers are in the outermost circle from the point of view of the product, but intimately involved from the point of view of development. Hence it may be an over-simplification to try to force the two taxonomies into one. However, as she says, drawing a development circle ‘highlights the necessity to have adequate representation of the ongoing maintenance operator and operational support roles and separates real maintenance from new development’.

Consultant: any of the many roles (marketing expert, software expert, business analyst, management specialist, etc) involved in supporting some aspect of Product development, characteristically from outside the development organisation. Internal consultancy is possible but problematic, as it is hard to speak out in the face of ‘political’ pressure within the organisation (except with the help of a Sponsor, see above).

Consultants may interact mainly with the Product Champion, with Purchasers, or with Developers, depending on when they were hired, by whom, and for what purpose.

For priority and timing, see the comments on Developer.

Another possible slot is Supplier: a role involved in the manufacture and provision of components, whether custom or commoditised, for the product. In the case of custom components, this is close to a Developer role initially. During system operations, both custom and commodity component suppliers have a role closer to Maintenance, supplying whatever is needed to keep the system operational.

Suppliers are important in product manufacture, in maintenance, and sometimes in development. In extreme examples such as the manufacture of a jet engine’s intercasing (a complex 3-dimensional casting), the supplier is critical to development, as the lead time for the manufacture of the component is comparable to the lead time for the entire engine development. Hence, suppliers may need to be involved very early in a project, possibly even before it begins.

The simplest and most essential use of a taxonomy of stakeholders is as a guide to the likely kinds of people to interview, observe, and invite to workshops to help gather the requirements for a project. As stakeholders are successively identified and brought into the project, an onion model and associated database can be used to document their roles, contact details, viewpoints, and ultimately requirements. The onion model then gives a direct visual report on the progress of stakeholder discovery and provides pointers to Slots that require further investigation.

The nature of the stakeholder taxonomy for an individual project determines the kind of stakeholder involvement that will be necessary during the life of the project. One application of the taxonomy is therefore to characterise the project in terms of the requirements elicitation approach and the degree of stakeholder participation that should be selected. This is one of several characterisation dimensions, each of which can help to reduce risk. Other dimensions include whether and how far safety is involved; how complex the interfaces are; and how new the needed technology is.

Elicitation techniques vary widely, from interviewing and workshops through ethnographic fieldwork to market surveys, prototyping and product trials. Most requirements textbooks describe a range of such techniques (e.g. Kotonya & Sommerville 1998, Robertson & Robertson 1999, Alexander & Stevens 2002). Textbooks with a user-centred orientation offer a still wider range (e.g. McGraw & Harbison 1997, Sutcliffe 2002). The choice of techniques is in practice, however, strongly driven by the stakeholder roles involved. Figure 3 illustrates the most common elicitation approaches for operational stakeholders.

Figure 4 similarly illustrates the most common elicitation approaches for non-operational stakeholders.

Characterisation can be carried out by assessing which kinds of stakeholder are likely to be most significant in a project and then assigning the most appropriate development life-cycle to respond adequately to those stakeholders. For example, if the operational needs are simple but meeting regulatory demands is long and complex (e.g. for a jet engine) then the life-cycle must focus on achieving certification with careful attention to verification and the gathering of evidence for the regulator.

Conversely, if regulation is minimal but operational needs are rigorous (e.g. for a portable consumer product), the life-cycle must focus on satisfying Users (i.e. hybrid Normal Operators/Functional Beneficiaries) that the product is suitable, through involvement with the development – which will presumably be iterative to permit it to respond to User reactions.

A tool that records details of Slots, Roles, and Stakeholders can readily detect Empty Slots. These might indicate forgotten Stakeholders, threatening project success.

Similarly, a tool could readily detect Slots populated by more than one Role. This is sometimes acceptable, but could indicate sources of conflict. For example, conflict is likely if two departments (e.g. International Marketing, Product Development) within an organization both believe they are funding and therefore have control of a development project.

A tool that in addition records traces between items such as requirements could detect any ‘Our System’ roles not traced to use case actors and thus documented in operational scenarios, and vice versa. Lack of traceability could indicate a disconnect between stakeholder analysis and implementation work.

The taxonomy presented here has, as described above, several possible practical applications. These do not necessarily require tool support: onion diagrams and hierarchies can be created quite easily in ordinary office software, on paper, or on flipcharts – provided that traceability is not required.

When models must be maintained for extended periods during which a steady trickle of change must be accommodated, tool support with configuration control and traceability to requirements becomes essential.

A tool (Figure 5) supporting some of the applications mentioned above, has been implemented in a requirements traceability tool environment (Telelogic DOORS) and is available for free download (Scenario Plus 2004). This both demonstrates the feasibility of building such a tool, and allows workers to experiment with uses of the approach in a controlled environment (e.g. with a complete audit trail). Figure 2 above incidentally illustrates part of the tool’s (very straightforward) data model.

The tool permits the creation and editing of an Onion Model diagram. As illustrated, the tool also acts as a display of the status of the model, making it apparent which Slots are filled and which are empty. The tool manages a conventional document-like data structure (a Formal Module, see Figure 2) in which each Slot, Role, and Stakeholder is represented as a separate Object within a hierarchy. Details of each Stakeholder are held as a database record within the Object; these include name and contact details as well as a text summarizing the stakeholder’s viewpoint. Objects can be given (bidirectionally navigable) traceability links to any other Objects (e.g. requirements) in the database. Hence the tool enables the requirements engineer to provide traces from any number of requirements, use cases, actors etc., directly to Stakeholders.

Tools can readily calculate metrics on the status of the Onion Model. For example, the number of Our System Slots that are not linked to use case actors gives a measure of the completeness of coverage of known human Operators in the use case model.

Clearly the same information architecture could also serve as the basis for other forms of stakeholder analysis including conflict detection and resolution. These do not necessarily demand tool support, but they do require systematic handling, for which a tool can be beneficial. Small projects may instead choose to apply a simple template of likely stakeholder roles, discussed in the next section.

The Scenario Plus tools could in principle be implemented in any automated software/system development environment that supports traceability between items and provides a suitable application programming interface. The hybrid hierarchical/tabular data structure of DOORS’ formal modules was very convenient for implementation but stakeholder information could certainly be represented without it.

A document template (Figure 6) based on the Onion Model and in a choice of formats suitable for popular word-processing and spreadsheet tools is available for free download (Scenario Plus 2004). While there are clear advantages for large projects to work within a traceability tool environment, smaller projects can benefit from stakeholder analysis and other forms of modelling using simpler tools.

Discussion: Features & Limitations of the Taxonomy

One Slot that is not provided is ‘User’. We consider this term to be both dangerously overloaded and confusing. It has many loose meanings in colloquial engineering parlance, including:

1. ‘all stakeholders’
2. ‘all stakeholders other than us, the developers’
   (– this meaning verges on the derogatory, and should be avoided)
3. ‘any stakeholder who gets any benefit from the product’
4. ‘any stakeholder who operates the product’.

However, perhaps the most widespread meaning is an interesting hybrid of two of our Slots: Normal Operator and Functional Beneficiary (Figure 7). For example, consumer electronics companies speak of ‘users’. They seem to mean the people who both push the buttons on their Products, and who enjoy the resulting entertainment (music, video, games, etc). This user/consumer role seems necessarily to combine the roles of Operator and Functional Beneficiary, and of course the role of Consumer/Mass-Market Product Purchaser – not to be confused with the important surrogate role of Purchaser adopted by product managers and procurement departments – is not far away either. Indeed, if changing the batteries is considered to be a Maintenance Operation, then the role is multiply hybrid.

Figure 7: ‘User’ as a Hybrid Role across 2 or more Slots

Popular variant terms such as ‘End-User’ do not clarify the situation. ‘End-User’ seems sometimes to mean "the actual operator of the product, as opposed to the purchasing organisation that is paying for its development", in which case its Slot is Normal Operator. At other times the implied Slot is Functional Beneficiary, as when astronomers are called the end-users of a scientific satellite because they make use of the data it collects. In any case, we may ask "The end of what?" with respect to the chain of beneficiaries on the left of the onion model, and indeed to the shifting perspectives of the onion models drawn with respect to different products within a system / subsystem hierarchy.

Therefore the terms ‘user’ and ‘end-user’ have little value, may be harmful in diverting attention away from non-operational stakeholders, and should not be chosen as fundamental taxonomic units.

Overlapping Taxonomies

Clearly, this model is Product-centric, though by intention people-focused. It is possible to create process-centric models (around the Developer and development activities, the business processes and system usage), and as argued in the Research Review, most of the requirements literature seems to be tightly product-centric to the exclusion of most kinds of stakeholder.

It is unwise to try to make a single taxonomy or hierarchy do too much (Jackson 1995, pages 92-95); the aim here is to provide a practical way of discovering and remembering the viewpoints necessary to a development’s success. Other models may be needed to cover business processes that stakeholders are involved in. Other hierarchies may be needed to suit other purposes such as defining company responsibilities. More is said on the dangers of reading too much into the model in 5.4 ‘Visual Metaphors’ below.

Usage-Centric Taxonomy

We have chosen to emphasize the range of roles across onion model Circles, as a counterbalance to the prevailing emphasis on (software) product usage to the exclusion of almost all other roles. For example, the Unified Modeling Language’s "Actors" are chiefly Normal Operators; the narrowness of the UML framework (Fowler & Scott 1999) often causes even Maintenance Operators to be forgotten, while the surrounding context of other Stakeholders able to contribute requirements is essentially undescribed (Figure 8). On the other hand, starting from an Onion Model, any operational role (in the ‘Our System’ Circle) is a candidate UML Actor.

Readers from other backgrounds accustomed to paying attention to a myriad stakeholder groups may find the need to counteract excessive Actor-fixation parochial; perhaps the most that can be done here is to assure them that it does seem to be a problem. The current unhappy emphasis on security requirements, in response to threats as diverse as hacking and terrorism, may however be helping to make even the most software-fixated developers aware that stakeholders do matter.

The Volere template (Volere 2004) contains an interesting and useful list of stakeholders – certainly far more comprehensive than most others in the literature. These are not easy to map into our taxonomy from the stakeholder role names alone: for instance, should ‘Maintenance Specialist’ be treated as a Maintenance Operator role or as a Consultant role? It could clearly be either. However the list seems rather developer-oriented: most of the roles appear to fall into the Developer and Consultant slots (see Table 1).

The numerous ‘Consultant’ roles in the Volere list – their names ending in ‘Specialist’ or ‘Expert’ – make sense if seen from the point of view of a classification of non-functional requirements: perhaps the thinking ran "here are some Usability and some Security requirements, we better consult some appropriate Specialists about these" (i.e. the knowledge, role, person triangle was the driver for discovering requirement-related knowledge (Robertson 2004)). Thus it appears that the Volere template is perhaps more specifically requirement-centric than developer-centric. On the other hand, the Negative and Champion Slots are rightly stated to be ‘project / product’-centric.

The resulting Onion Diagram (Figure 9) shows that (for all the length of the Volere stakeholders list) several Slots remain unfilled, perhaps surprisingly including the one for stakeholders responsible for interfacing systems.

Several other slots are only weakly represented, e.g. ‘Auditors’ is the sole (and rather doubtful) role in ‘Regulator’.

Volere's ‘Customer’ and ‘Client’ are slippery terms; in this analysis they are assumed to correspond roughly to our ‘Purchaser’ and ‘Functional Beneficiary’ respectively; Client can however also include Financial Beneficiary.

One aspect of stakeholder sociology that is brought out clearly on an Onion Diagram is surrogacy – the representation of one stakeholder’s viewpoint by some other person. Surrogacy has (when it has been considered at all) always been seen as a dangerous obstacle to successful requirements work. It is therefore remarkable that stakeholder surrogacy (Figure 10) is both central to requirements engineering (RE) and scarcely discussed in the RE literature (see the Research Review).

Figure 10 describes some of the surrogacy issues around requirements engineering as a whole, and stakeholder sociology in particular. It is immediately apparent that the different stakeholder – surrogate relationships each have their own advantages and risks. These are briefly analysed in Table 2.

Other forms of surrogacy are also involved, given that both in development projects and in ordinary work (using the developed products) people are usually working on behalf of other people.

Combinations and intermediate types can also be identified. For example, environmental pressure groups and ‘not in my back yard’ campaigners may create leaders who, though unofficial and unpaid, may act almost as Regulators, being invited to meetings and having their (Negative) opinion sought.

The history of computing can be seen as a steady retreat from the assumption that we know how to program what we want done in machine language: successively more human-oriented languages and tools (such as assembly language, Fortran, visual programming, databases) have been invented to fill the gap.

The history of requirements can be seen as a steady retreat from the assumption that we know how to say what we want done in atomic ‘shall’ requirements: successively more stakeholder-oriented documents and diagrams (such as User Requirements, Concepts of Operations, Use Cases, Goal Models) have been invented to fill the gap.

The roles people play – what they do, and what they want to do – thus seem to be becoming increasingly important to our discipline of RE. Our ‘machine codes’ – individually prioritised and traceable requirements – will not simply disappear. But like hexadecimal, the writing of requirement text is moving steadily into the background as more abstract representations of what people want are invented and adopted in practice.

A ‘user requirement’ has the general form

<named stakeholder role> wants <result>.

If the gold standard for RE is to create requirements like this – claiming that specific stakeholders (we’ll pass over the class/instance issue for a moment) want that result, more or less personally for their own work (by which we include leisure uses of consumer products, etc), then surrogacy is a rampant anomaly. All that surrogate stakeholders can give us are statements of the general form

<named surrogate role> believes that <named role> wants <result>.

In law statements of this kind are called ‘hearsay evidence’ which is rarely admissible in court. But our surrogates often cannot even claim to have heard anyone else expressing their requirement, however distantly. In the clearcut case of the procurement of military equipment on behalf of as yet unborn soldiers, there can be no claim even to hearsay evidence: the requirements are wholly invented.

Surrogacy thus interferes with the core trend and practical process of RE: the move towards gathering requirements directly from real stakeholders. Yet as Figure 10 indicates, surrogacy cannot readily be avoided (even when the ‘unreachable population’ is not especially large). Perhaps this is one of the fundamental limitations on the engineering of requirements, systems, and software.

Suzanne Robertson comments that it can help to invent a surrogate, e.g. Erik the systems analyst. Erik could then be ‘consulted’ on his needs:

Without the surrogate, they’d never have been able to finish the job.

An Onion Diagram inherently embodies a set of visual metaphors, and these are exploited more or less consciously to give a feeling for stakeholder relationships in a development project. Metaphors inevitably break down at some point, so some caution is required. Attention is drawn to the limitations of the onion model below.

The most immediate of these visual metaphors is the feeling of successive (indeed recursive) containment of one 'onion' within another. As Peter Checkland and others have emphasized, there are always many systems to consider (Checkland & Scholes 1990). This leads to practical issues of project scope. The scope of a system of people-and-equipment is always larger than the scope of the equipment or product alone, so it is essential to be clear about the existence of the two boundaries (e.g. Robertson & Robertson 1999).

A circular diagram inevitably announces that something is in the centre of attention, and in a development that is rightly the product; at the same time, it shows that people are ‘all around’ it and have importance. It also suggests the metaphor of closeness for relatedness, and for example developers are shown close to purchasers for this reason. Distance could suggest that stakeholders in the outer circles are less important, but this is not intended. Indeed, negative stakeholders, regulators, and political beneficiaries are all able to ‘stop the show’ and must be considered essential.

We also habitually make use of other spatial metaphors; supporting roles seem naturally to find a place below the product, while interfacing systems and regulators can go ‘off to the side’ somewhere.

A cautionary note should be sounded here: a naïve reading of the onion model with the equation: outer = higher in the system or company hierarchy, with regulator in the outermost circle, would imply the equation regulator = responsibility for highest level of system. This is simply wrong. As Michael Jackson cautions (Jackson 1995, and see the Research Review), there may be many overlapping hierarchies. Three that are often important in thinking about system development are:

1. the scopes of the product and the systems that contain it: this is the primary and intended meaning of the onion diagram. The operational stakeholders are those inside ‘the system’, as explained below.
2. the organisation (e.g. company) hierarchy, with the workers at the bottom/inside and the managers and directors towards the top/outside. This is not what is intended here. Managers may or may not be stakeholders in a development; non-operational company workers and executives are not automatically assumed to be relevant.
3. the structure of the organisations and stakeholder relationships involved in accepting or rejecting a development, e.g. the relationship of the regulator with the developer. This structure is secondarily suggested by the onion diagram, but this does not mean that questions like ‘which level of system is the regulator in?’ make any sense. The regulator is in a system – that of the political and legal framework of a country or the world – but that is certainly not the same as any onion-circle of any ordinary development project.

The onion has its uses in helping people to think about multiple scopes and the need to involve people outside the development team, but it can’t do everything.

Another metaphor that seems quite natural (though perhaps mixing oddly with onions) is that of the chain of command, where a person makes use of a system under their command as an instrument to achieve a ‘political’ goal. They do this by instructing an immediate subordinate – some at the adjacent or immediately contained level – to carry out some operation. That person in turn may achieve their goal (a successful operation) by instructing an operator to use a piece of equipment to deliver a specific result. The ‘chain’ metaphor could also be represented as a scale from higher to lower if you like the idea of a Great Chain of Being from God to man to animals (Lovejoy 1936); I have chosen to make it the rather more egalitarian left-to-right (with the slight suggestion that actions, like writing, may proceed in that direction).

The trend in system specification away from the machine towards human "users" leads to a natural end-point in modelling and analysing the nature, goals, and viewpoints of human stakeholders. While there has been some academic interest in goal and viewpoint modelling, the stakeholders themselves seem largely to have been overlooked within systems and software engineering (though quite the reverse in soft systems and information systems research).

The 'onion model' diagram has an attractive and intuitive simplicity about it, organizing a substantial amount of complexity in analysis of stakeholder sociology and relationships. This is part of a wider interest in and intention to improve the attention that practising engineers pay to stakeholders, including surveys and a reconsideration of development life-cycles and project information models (Alexander & Robertson 2004). The default onion model with three circles and the basic slots is often in itself helpful on projects, but it is readily tailored for more complex situations with additional circles and slots. Ralph Young has adopted the approach in his new book (Young 2004, pp 65-67), and Suzanne and James Robertson are similarly doing so in their forthcoming book (Robertson & Robertson 2004). Needless to say, it is also pressed into service in my forthcoming book (Alexander & Maiden 2004).

Tool support is available, but the approach can be applied with nothing more than a hand-drawn diagram and a textual hierarchy of headings for circles, slots, roles, and if desired named stakeholders.

The analysis of stakeholders does not presuppose any particular development approach, and while it is clearly suitable for an object-oriented worldview with use cases and their actors, it can equally be added to conventional or agile development.

This paper has presented a simple but it is hoped generally-applicable and readily-customised taxonomy of stakeholders, and has suggested some uses for it. If it contributes to encouraging even a few projects to think a little more deeply about who their stakeholders are and what they want, it will have succeeded.

I am grateful to Isabel Ramos for her suggestions and encouragement for this paper; to Andrew Farncombe for his clarity on terminology and multiple hierarchies; and to Suzanne Robertson for her continuing enthusiasm and support. I would also like to thank the anonymous reviewers for their perceptive comments and suggestions especially on the voluminous Information Systems literature, and the Editor – Bernd Stahl – for his measured criticism and direction.

Alexander, I. (2002). Initial Industrial Experience of Misuse Cases in Trade-Off Analysis, IEEE Joint International Requirements Engineering Conference, 9-13 September 2002, Essen, Germany, 61-68.

Alexander, I. (2003). Stakeholders – Who is Your System For?, Computing & Control Engineering, 14, Issue 1, 22-26, April 2003.

Alexander, I. & Maiden, N. (Editors) (2004). Scenarios, Stories, Use Cases through the system development life-cycle, John Wiley.

Alexander, I. & Robertson, S. (2004). Understanding Project Sociology by Modeling Stakeholders, IEEE Software, 21, No. 1, January/February 2004, 23-27.

Alexander, I. & Stevens, R. (2002). Writing Better Requirements, Addison-Wesley.

Avison, D. & Wood-Harper, A.T. (1990). Multiview: An Exploration in Information Systems Development. Oxford: Blackwell Scientific Publications.

Bennetts, P. D. C. & Wood-Harper, A.T. (2000). Software Process Improvement through the Lens of Multiview 2: The Development of a Contingent Factors Model, in 10th Conference on Business Information Technology, Manchester, UK.

Boehm, B., Grünbacher, P., Briggs, B. (2003). Developing Groupware for Requirements Negotiation: Lessons Learned, IEEE Software, May/June 2001, 46-55.

Bowker, G.C. & Leigh Star, S. (1999). Sorting Things Out, classification and its consequences. MIT Press.

Checkland, P. & Scholes, J. (1990). Soft Systems Methodology in Action, John Wiley.

Coakes, E. & Elliman, T. (1999). Focus issue on legacy information systems and business process engineering: the role of stakeholders in managing change, Communications of the AIS, Volume 2, Issue 1es (July 1999), Article No. 4.

Damian, D.E. & Zowghi, D. (2002). The impact of stakeholders’ geographical distribution on managing requirements in a multi-site organization, Proceedings of the IEEE Joint International Requirements Engineering Conference, Essen, Germany, September 2002, 319-328.

Donaldson, T. & Preston, L.E. (1995): The Stakeholder Theory of the Corporation: Concepts, Evidence, and Implications. In: Academy of Management Review 1995, Vol 20, No 1: 65 - 91

Fowler, M. & Scott, K., (1999). UML Distilled, A Brief Guide to the Standard Object Modeling Language, Addison-Wesley.

Gause, D. & Weinberg, G. (1989). Exploring Requirements, Quality Before Design, Dorset House.

Hirschheim, R. & Klein, H.K. (2003): Crisis in the IS Field? A Critical Reflection on the State of the Discipline. In: Journal of the Association for Information Systems (4:5): 237 - 293

Jackson, M. (1995). Software Requirements & Specifications, a lexicon of practice, principles and prejudices. Addison-Wesley.

Kitapci, H. & Bhuta, J. (2004.) EasyWinWin Negotiations: Guidelines and EasyWinWin Report, http://sunset.usc.edu/classes/cs577a_2003/assignments/Team/ EasyWinWin%20Negotiations.htm (downloaded 20 April 2004)

Kotonya, G. & Sommerville, I. (1998). Requirements Engineering: Processes and Techniques, John Wiley.

Koyré, A. (1957). From the Closed World to the Infinite Universe. The Johns Hopkins University Press, Baltimore and London. (Reprinted in paperback 1968)

Levine, R., Locke, C., Searls D. & Weinberger, D. (2000). The ClueTrain Manifesto, The end of business as usual, FT.com, London.

Lovejoy, A.O. (1936). The Great Chain of Being. Harvard University Press, Cambridge, Massachusetts. (Reprinted in paperback, 1990)

Mason, R. & Mitroff, I. (1981). Challenging Strategic Planning Assumptions:Theory, Cases, and techniques. New York: John Wiley & Sons.

McGraw, K., & Harbison, K. (1997). User-Centered Requirements, The Scenario-Based Engineering Process, Lawrence Erlbaum Associates.

Midgley, G. (1992). The Sacred and Profane in Critical Systems Thinking, Systems Practice, (5)1, 5-16.

Mumford, E. (1996). Systems Design, Ethical Tools for Ethical Change. Macmillan.

Potts, C. (1995). Invented Requirements and Imagined Customers: Requirements Engineering for Off-the-shelf Software (Invited workshop introduction paper), Proc. RE'95: Second International Symposium on Requirements Engineering, York, England, March, 1995, IEEE Computer Society Press. Available from http://www.cc.gatech.edu/fac/Colin.Potts/pubs/1995/re95/re95InventedReqtsOTS.pdf

Pouloudi, A. (1999). Aspects of the Stakeholder Concept and their Implications for Information Systems Development. Proceedings of the 32nd Hawaii International Conference on System Sciences, Volume 7, January 5-8 1999, Maui, Hawaii.

Pouloudi, A., & Whitley, E. A. (1997). Stakeholder identification in inter-organizational systems: gaining insights for drug use management systems. European Journal of Information Systems, 6 (1), 1-14.

Robertson, S. (2004). Personal communication. 18 March 2004.

Robertson, S. & Robertson, J. (1999). Mastering the Requirements Process, Addison-Wesley.

Robertson, S. & Robertson, J. (2004). Requirements-Led Project Management, Addison-Wesley.

Scenario Plus (2004). Scenario Plus website: http://www.scenarioplus.org.uk.

Sharp, H., Finkelstein, A. & Galal G. (1999). Stakeholder Identification in the Requirements Engineering Process. DEXA Requirements Engineering Process Workshop, 387-391.

Sommerville, I., Sawyer, P., Viller, S. (1998). Viewpoints for Requirements Elicitation: a Practical Approach, Proceedings Third IEEE International Conference on Requirements Engineering.

Sutcliffe, A. (2002). User-Centred Requirements Engineering, Theory and Practice, Springer.

Volere (2004). Volere Stakeholder Analysis Template, http://www.volere.co.uk (downloaded for analysis 20 February 2004)

Wood, D. (1999). Trends in Stakeholder Management & Measurement in the Public Sector, http://www.navigate.co.nz/newsreti.htm (downloaded 20 April 2004)

Young, R.R. (2004). The Requirements Engineering Handbook, Artech House.

Younkins, E.W. (2001). Morality and Character Development: The Roles of Capitalism, Commerce, and the Corporation, Journal of Markets & Morality 4, 1 (Spring 2001), 94-111 http://www.acton.org/publicat/m_and_m/2001_spring/ pdf/mm-v4n1-younkins.pdf (downloaded 20 April 2004)

Yu, E. & Mylopoulos, J. (1998). Why Goal-Oriented Requirements Engineering, Requirements Engineering Foundation for Software Quality Workshop 1998, http://www.cs.toronto.edu/pub/eric/REFSQ98.html

Stakeholders: Chapter 2 of Discovering Requirements, Wiley 2009

Other Papers