In many projects the problem is not the work but the question of who does it and who decides. Two people each think the other is handling it, or four people co-decide and nobody carries the responsibility. The RACI matrix is a simple tool for clarifying this.
The four roles
Responsible: who carries out the task? There can be several people. Accountable: who is answerable for the result and decides finally? Exactly one person per task. Consulted: who is asked beforehand because their knowledge or approval is needed? Two-way communication. Informed: who is told about the result? One-way communication.
The most important rule
There is exactly one Accountable per task. Two Accountables mean none in practice. If the matrix shows several people wanting the A for a decision, that is not a formatting problem but a governance problem that must be solved before the project moves on.
Building a RACI
Rows are tasks, decisions or deliverables; columns are roles, not people — so the matrix stays valid when people change. The matrix is built together, not decreed, because the discussion about the A is exactly what creates clarity. Then it is checked: does every row have exactly one A and at least one R? Are there roles with too many Rs that are overloaded? Roles with many Cs that become bottlenecks?
Common mistakes
Too many rows, so nobody reads the matrix. People instead of roles. Everyone as Consulted, out of politeness, which slows decisions. And a matrix built at project start and never looked at again. A good RACI is short, covers the critical decisions and is updated when organisation or scope changes, much like a risk register.
Beyond projects
RACI also suits processes and compliance: who owns the record of processing activities, who reports a data breach, who approves an AI system? When setting up AI governance, clarifying who acts as provider or deployer of an AI system and who owns human oversight is at its core a RACI question. Our AI governance set-up offer therefore starts with exactly this matrix.