Skip to content

Decision-Rights Matrix

Authority artifact — instantiates Principal–Agent Alignment

Maps each class of decision to who may decide, approve, be consulted, or merely be informed — fixing the agent's authority before any single choice arises.

Decision-Rights Matrix is the standing map of authority: a grid whose rows are classes of decision and whose columns are roles, with each cell saying what that role may do with that decision — decide it alone, approve someone else's call, be consulted first, or only be told after. Its defining trait is that it settles authority in advance and in general, so that when a specific choice arrives nobody has to negotiate who owns it. Where an escalation rule fires on a triggering event, the matrix is the quiet, always-on allocation the event is measured against. Filling a row is also where a principal decides how much rope to give: the width of a role's "decide alone" column is, in effect, its autonomy dial.

Example

A construction firm delivering a hospital fit-out keeps colliding over change orders — the project manager approving things the client's owner-rep thinks needed sign-off, the architect blindsided by field substitutions. Before the next phase they build a decision-rights matrix. Rows: design changes, budget reallocations under $50k, budget changes over $50k, schedule extensions, material substitutions, safety stoppages. Columns: site superintendent, project manager, architect, owner's representative, client sponsor.

The cells resolve the friction concretely. Material substitutions: superintendent decides, architect consulted, owner's rep informed. Budget changes over $50k: project manager proposes, client sponsor approves, everyone else informed. Safety stoppage: superintendent decides alone, no consultation required, all informed immediately. The matrix does not make any single decision — it makes sure that for every decision, exactly one role knows it owns the call and the others know their part is to advise, approve, or stay out of the way.

How it works

The build is a two-axis exercise. First enumerate the decision classes that actually cause friction or carry risk — not every micro-choice, only the ones worth governing. Then assign each cell one authority verb (a RACI-style vocabulary of decide / approve / consult / inform is common),[1] with the rule that every decision class has exactly one decider — split ownership is the failure the matrix exists to prevent. The distinctive discipline is granularity control: too coarse and the matrix restates the org chart, saying nothing new; too fine and it becomes an unmaintainable rulebook nobody reads. A good matrix is short enough to fit on a wall and specific enough to end an argument.

Tuning parameters

  • Decision granularity — how finely decision classes are cut. Finer cuts place authority precisely but multiply rows and go stale; coarser cuts stay readable but leave grey zones.
  • Autonomy width — how many decision classes each role decides alone versus needing approval. Wider columns speed the work and signal trust; narrower ones add safety and slow throughput.
  • Approval depth — one approver or a chain. Extra approvers reduce error on high-stakes cells but add latency and diffuse responsibility.
  • Consultation scope — who must be consulted before a decision. Broad consultation improves buy-in and catches blind spots; it also dilutes the decider's ownership if overused.

When it helps, and when it misleads

Its strength is eliminating the most common delegation failure — unauthorized discretion and its mirror, decision paralysis — by making ownership explicit and collision-free before the pressure is on. It is most valuable exactly where the main misalignment risk is an agent quietly deciding things that weren't theirs to decide.

It misleads when it is mistaken for a substitute for the other alignment machinery: a crisp authority map does nothing about whether the decider's incentives point the right way, whether their work is observed, or whether consequences follow a bad call. It can also ossify — cells written for last year's structure keep routing decisions to roles that no longer fit, and people route around a stale matrix rather than fixing it. The guard is to treat the matrix as living, re-examined when the org or the risk profile shifts, and to pair it with monitoring and incentives rather than leaning on it alone.

How it implements the components

  • agent_role_and_discretion_boundary — the matrix is the boundary: each cell states what a role may decide alone, what needs approval, and what is outside its mandate.
  • trust_and_autonomy_calibration — the breadth of each role's unilateral authority is the concrete setting of how much autonomy the principal extends.

It does not fire the dynamic trigger that routes an out-of-bounds decision upward — that is Escalation Protocol; and it sets no incentives, monitoring, metrics, or consequences (Incentive Compensation Plan, Audit or Review Cycle, Governance Board).

  • Instantiates: Principal–Agent Alignment — supplies the authority-boundary component the rest of the relationship builds on.
  • Sibling mechanisms: Escalation Protocol · Governance Board · Reporting Requirement · Audit or Review Cycle · Service-Level Agreement · Fiduciary Duty Rule · Incentive Compensation Plan · Performance Contract · Reputation System · Clawback Clause

References

[1] RACI (Responsible, Accountable, Consulted, Informed) is a standard responsibility-assignment scheme; a decision-rights matrix uses this or a similar vocabulary to give each cell a single unambiguous authority verb. ↩