Skip to content
← Frameworks

Decision Rights Matrix

When AI systems participate in decisions, the question is not whether to use them – it is who is accountable. This matrix prevents the accountability vacuum that undermines most AI deployments.

The Accountability Vacuum

Most organizations deploying AI cannot answer a basic question: When the AI is wrong, who is responsible? Not the vendor. Not the data team. Not the business unit. Not the board. The answer is usually silence – and silence is risk.

The Decision Rights Matrix assigns four explicit roles to every AI-augmented decision, eliminating ambiguity before it creates liability.

Four Roles, No Gaps

Role Function Question Answered
Proposer Generates recommendation or output "What does the system suggest?"
Approver Validates and authorizes action "Who signs off?"
Override Can reverse or halt the decision "Who can say no after the fact?"
Auditor Reviews outcomes, detects drift "Who checks whether it worked?"

Applied: Hiring Decision (AI-Assisted)

Role Assigned To Authority
Proposer AI screening tool + Recruiter Generates candidate shortlist based on criteria
Approver Hiring Manager Reviews shortlist, makes interview/offer decisions
Override CHRO / VP People Can halt process if bias detected or policy breached
Auditor Internal Audit / DPO Quarterly review of outcomes, adverse impact analysis

Implementation Principles

01

No role left empty

Every AI-augmented decision must have all four roles assigned before deployment. An empty cell is an unmanaged risk.

02

Roles map to people, not teams

Accountability requires a name, not a department. "The data team" cannot be an approver.

03

Autonomy level determines role weight

At L1 (Assisted), the Approver matters most. At L4 (Autonomous), the Auditor becomes critical. See the Autonomy Levels Model.

04

Review cadence matches risk tier

High-risk decisions (EU AI Act classification) require quarterly audit. Low-risk decisions can be reviewed annually.

Governance Principle

The Decision Rights Matrix is not a permissions system – it is an accountability architecture. Its purpose is to ensure that when AI participates in a consequential decision, the answer to "who is responsible?" is never ambiguous.

Sources: NIST AI RMF 1.0 GOVERN function, ISO/IEC 42001:2023 Clause 6, EU AI Act Article 14 (Human Oversight)