Building a consulting team with Belbin roles: five people, nine roles

You have five people and a six-week advisory project. Belbin lists nine roles, and five does not divide into nine. This post is for whoever forms a team of three to five, from a manager building a task force to someone recruiting friends for a shared project: how to use the nine roles to inspect your team, and how far the instrument that measures them can be trusted.

What are the nine Belbin roles, and what are they for?

They sit in three orientations: action, people and thinking. For a small team I use them as a checklist: which jobs in this project have nobody to take them? The library's Belbin framework poses exactly this question for someone forming a new team or whose team has stalled.

What is missing from a team of five? Build a gap grid

Put the nine roles in rows and the people in columns. Mark each person's preferred roles (P) and manageable roles (M); a row with no P is a gap. The two levels come from Belbin Associates' own materials, which are not in the library and not peer reviewed: preferred roles come naturally, and manageable roles are ones a person can play when the team needs them. I use them only as marking labels. Take a hypothetical team of five, A to E, advising an unnamed factory for six weeks on whether to build a second plant, all five having completed the inventory. Every entry is my invention.

The Specialist row is empty in both P and M, yet the factory needs someone who understands production costs. I would borrow from outside the team, for instance by interviewing an engineer, rather than press anyone into the role.

Completer Finisher has only two M entries, C and D. I would make C responsible for checking each deliverable, with a fixed checkpoint before every hand-over to the client, and not lean on D, the only one preferring Monitor Evaluator, whose challenge rounds fall at the same checkpoints. That is my reasoning.

A is the only person preferring Shaper and also the only one preferring Co-ordinator. The two pull in different directions, one pushing, one welcoming every view; if A is away for a week the team loses both.

On balance across orientations, the ten P entries split three action, five people, two thinking. My inference: this team will hold pleasant meetings and rarely have its numbers questioned, so I would give D the job of challenging at the end of each phase.

What if there are fewer people than roles?

Roles get covered, not people. The abstract of Aritzeta and colleagues does not discuss covering roles with fewer people, so the method below is my own.

Try it with three people, A, C and D. Preferred roles cover five: Shaper, Co-ordinator, Implementer, Teamworker and Monitor Evaluator. Three more are manageable only: Completer Finisher (C, D), Resource Investigator (A) and Plant (D). Specialist stays empty. Eight of nine roles have someone, but three of them rest on a manageable fit.

The method: rank the nine roles by what the current phase needs; assign preferred roles first; give each manageable-only role a named owner and a concrete checkpoint; never assign anyone a role they listed as one to avoid; borrow whatever is still empty. One assumption of my own: early weeks need Plant and Resource Investigator, closing weeks need Implementer and Completer Finisher, so one person can hold two roles at different times. I have found no source that tests this.

How far can the Belbin inventory be trusted?

Adequate on one count, weak on another. Aritzeta, Swailes and Senior (2007) examined the psychometric properties of the Team Role Self-Perception Inventory along with 43 empirical studies testing associations between team roles and other cognitive or behavioural traits. They write that the evidence is mixed, but that on balance the model and its inventory have adequate convergent validity, while strong associations between some roles indicate weak discriminant validity among some scales (Aritzeta et al., 2007).

In plain terms (my gloss, not the authors' wording), convergent validity means role scores relate to other traits as theory predicts; discriminant validity means two different roles should produce different scores. The abstract does not say which roles overlap, so I will not name a pair. My suggestion: treat a role label as a hypothesis, and when someone scores high on two neighbouring roles, ask them and their colleagues which one they see themselves playing. The abstract also does not show that teams covering every role outperform others. The title includes "applications for team building", so the full text may take that up, but I could not read it.

My own reading of the abstract: the weakness it reports concerns how well some scales separate roles, that is, the measuring tool. It does not say whether a team needs several kinds of contribution, and I have seen no study that tests that.

What are the limits of a self-report when forming a team?

The result is only as good as the respondent's self-knowledge, and the name says so: Team Role Self-Perception Inventory (Aritzeta et al., 2007). People may also answer as they would like to be seen; that is my inference from the format, not a finding I read. To balance it, I would ask colleagues who have worked with each person how they see them, and set the two views side by side in a conversation. For a new team those colleagues must come from earlier work. I do not know how the full paper treats scoring or outside ratings.

Does covering the roles on paper make the team work?

No. The grid tells you which jobs have an owner, not whether that owner dares to speak. Edmondson (1999) studied 51 teams in a manufacturing company and found that team psychological safety, a shared belief that the team is safe for interpersonal risk-taking, was associated with learning behaviour, and learning behaviour mediated between psychological safety and team performance (Edmondson, 1999, p. 350).

Applied to the grid, in my reasoning: if D, who prefers Monitor Evaluator, does not feel able to object, that row is covered only on paper. A simple rule: every deliverable gets a challenge round led by whoever holds that role.

Edmondson's limits: the survey was cross-sectional, so it cannot prove causation, and a single company with 51 teams calls for caution when generalising (Edmondson, 1999, p. 379). She also reasons that learning behaviour is likely to help less on tightly constrained tasks, such as a machine-paced assembly line, than on tasks with few constraints, such as designing a new product; that is her conjecture, not a tested result (Edmondson, 1999, p. 378). An advisory project, in my view, is closer to the second.

When should you use this, and when not?

It suits you if you can choose people or divide the work, the project lasts several weeks so a missing role becomes a real problem, you can ask colleagues who know each person for their view, and the aim is to start a conversation about who does what.

It does not suit you if you intend to use the scores to select or reject people: the inventory is a self-report and discriminant validity is weak for some scales, so a label could wrongly exclude someone (my judgement). Nor does it suit a team whose membership is fixed and whose task is narrow, or a team of two.

If a formal inventory is out of reach, spend twenty minutes: each person writes down two roles they want and one they would avoid, and the team builds the gap grid. I have read no evidence that this matches the inventory's results; its value is that everyone must say who does what.

Going deeper: what did I read, and what did I not?

Aritzeta, Swailes and Senior (2007) sits behind a paywall and Unpaywall reports no open copy. I read the public abstract on the University of Huddersfield Research Portal on 3 November 2026, which lists it as a peer-reviewed review article. I do not know the paper's reliability figures, which roles overlap, or its team-building recommendations.

I read Edmondson (1999) in full, from a scanned copy converted by automatic text recognition. I also read a 2014 review document that Belbin Associates, the firm selling the inventory, publishes on its website. It is not peer reviewed and not in the library, so I do not cite it; it is the only source of the preferred and manageable levels used in the grid. I have not read Belbin's books or any independent study of whether role-balanced teams perform better, and nothing about teams in Vietnam. Team A–E and the fixes are my own reasoning.