The salience model: sorting stakeholders by power, legitimacy and urgency

The salience model sorts stakeholders by three attributes: power to affect the work, legitimacy of their claim on it, and urgency of that claim. Hold one and you get noticed occasionally. Hold two and you get managed. Hold all three and you are a definitive stakeholder, and your claim goes to the front.

Three attributes, not two

Most stakeholder analysis in practice runs on two axes. The Association for Project Management describes the discipline as identifying stakeholders and then analysing the degree to which they may help or hinder, weighing the relative power of the stakeholder and the degree of interest they are likely to show. Power and interest, plotted on a square, four boxes, and a management strategy for each. It is quick, it is teachable, and most of the time it is enough.

It has one blind spot, and the blind spot is expensive. Power and interest together tell you who can hurt the work and who cares about it. Neither tells you whether a claim is actually rightful, and neither tells you whether it is pressing right now. A grid built on those two axes quietly ranks a well connected department above a community group with a legal entitlement and no leverage, because the department scores higher on both axes. It is not wrong about the politics. It is silent about the obligation.

The salience model, set out by Ronald Mitchell, Bradley Agle and Donna Wood in a 1997 paper in the Academy of Management Review, splits that missing dimension in two. Legitimacy asks whether a claim is generally accepted as proper. Urgency asks whether it is time critical and pressed. Add those to power and you get three attributes instead of two, and three attributes generate seven classes instead of four boxes.

What power, legitimacy and urgency each mean

Power is the capacity to bring about an outcome. It arrives through formal authority, through money, through control of labour or materials, or through the ability to mobilise other people. Power does not need to be exercised to exist, which is why a regulator that has never intervened still holds it.

Legitimacy is whether the claim is generally accepted as proper. Legal and contractual rights are the clearest form, but legitimacy is broader than either: a neighbour with no contract and no statutory standing may still have a claim most reasonable people would recognise. Legitimacy is a social judgement, not a legal test, which means it is contestable and occasionally wrong.

Urgency has two parts, and the second is the one people forget. A claim is urgent when it is time sensitive and when it matters enough to its holder that they will actually press it. A deadline nobody chases is not urgent in the sense the model means. Urgency is what converts a held attribute into visible pressure.

Each attribute is judged present or absent. That is a deliberate design choice rather than a simplification: scoring all three out of five and adding them up would produce a ranking, and a ranking would hide the thing the model exists to show, which is that particular combinations behave in particular ways.

The seven classes the attributes produce

Class	Attributes	What it means for you
Dormant	Power	Can act, has no accepted claim and no reason to press one. Watch, do not court.
Discretionary	Legitimacy	A proper claim, no leverage, no pressure. Attended to by choice, which usually means not at all.
Demanding	Urgency	Noisy, no standing, no power. Consumes attention out of proportion to consequence.
Dominant	Power, legitimacy	The formal stakeholder your governance already recognises. Boards, sponsors, regulators.
Dangerous	Power, urgency	Able and pressing, without an accepted claim. Coercive, and the class most likely to surprise you.
Dependent	Legitimacy, urgency	A rightful, pressing claim and no way to advance it. Needs an advocate to be heard.
Definitive	All three	Attended to first, ahead of everyone. The class every escalation is trying to reach.

Three of these classes carry most of the practical weight, and they are not the three people expect.

Dangerous is the one that ruins quarters. Power and urgency without legitimacy describes a stakeholder who can act, wants to act now, and has no accepted claim to constrain how they do it. Because the claim is not legitimate, it usually has no route through your governance, so it arrives outside the process: in the press, in a complaint to a regulator, in a sudden refusal to cooperate. Nothing in a power and interest grid predicts that shape.

Dependent is the one that gets neglected. A rightful, pressing claim with no power behind it cannot advance itself, so it waits for somebody else to carry it. Left alone, dependent claims either evaporate or acquire power from an advocate, and an advocate is very often a regulator or a journalist. A dependent stakeholder is not a low priority. It is a definitive stakeholder that has not found its sponsor yet.

Discretionary is the quiet test of an organisation. Legitimacy alone, with no power and no urgency, means the only reason to attend to the claim is that it is proper. The model is explicit that these claims get attended to at the manager’s discretion, which is a polite way of saying they usually do not.

Salience is not a fixed property

The single most useful thing in the model is also the thing most often left out when it is taught: the attributes are variable. A stakeholder acquires and loses them, sometimes over months and sometimes over a weekend.

Watch what that does. A dormant stakeholder holds power and nothing else, so you have filed them under “monitor”. Their claim then becomes urgent, for reasons that have nothing to do with you: an election, an incident, a change of leadership. They are now dangerous, and dangerous stakeholders act outside the process. Nothing about their power changed. One attribute arrived, and the correct response to them inverted.

The same movement runs the other way and is easier to miss. A definitive stakeholder whose claim is settled loses urgency and becomes dominant: still powerful, still legitimate, no longer pressing. Teams often keep spending on that relationship at full intensity for months after the pressure has gone, because the classification was made once and never revisited.

Running it without turning it into a workshop

The mechanics are smaller than the theory suggests. Three columns on the stakeholder register, one per attribute, each holding yes or no. A fourth column derives the class, which means it is a lookup rather than a debate. Nothing else is required, and anything more elaborate tends to be the analysis eating the work it was supposed to inform.

Two practical points decide whether the exercise is worth the hour. The first is that the assessment should be dated, because the whole value of the model sits in what changes between one assessment and the next. A register showing that a stakeholder moved from dormant to dangerous in March is doing something a snapshot cannot: it is telling you the direction of travel while there is still time to act on it.

The second is that the definitive stakeholders are not where the return is. They already have power, standing and pressure, so they command attention with or without your register; writing their names down changes nothing about how much of your week they get. The return is in the two classes nobody escalates. A dependent claim, rightful and pressing with nothing behind it, is the cheapest problem you will ever solve while it is still dependent, and the most expensive once it has found an advocate with power. A dangerous stakeholder, powerful and pressing with no accepted claim, is worth knowing about before they choose a route that bypasses you entirely.

So the useful output of a salience assessment is rarely the list of who matters most. It is the short list of people whose claims your process is currently structured not to hear.

The model describes attention, it does not award it

Here is the caveat that matters most, and it is a caveat about what kind of thing the model is.

Mitchell, Agle and Wood built a descriptive theory. The question they set out to answer is which claims managers actually attend to, and the seven classes are a prediction about observed behaviour. They are not a moral ranking, and the model does not say that a discretionary claim deserves to be ignored. It says that, in practice, it will be. Those are entirely different statements, and reading the second as the first is the common error. Used properly, the model is at its sharpest when it tells you something uncomfortable: this claim is rightful, it is pressing, and our own process is structured so that nobody will hear it.

There is a second limit, quieter and harder to correct. Every one of the three attributes is assessed by somebody, and normally that somebody is the team doing the work. Legitimacy in particular is a judgement about whether a claim is proper, so it inherits whatever the assessor already believes about who counts. A stakeholder classified as demanding rather than dependent has been judged to have no proper claim, and that judgement was made by the party the claim is against. The model has no mechanism to catch that, which is a good argument for having the assessment challenged by somebody outside the team at least once.

Neither limit makes the model less useful. Both make it a diagnostic rather than a verdict. Three attributes, seven classes, and the honest reading is that the classification tells you as much about the organisation doing the classifying as it does about the people being classified.