Decision Rights
TLDR
Decision rights specify who has authority to decide what. In hybrid human-AI processes, unclear decision rights kill adoption.
Definition (in scope)
Decision rights are the specification of who can make which decisions, under what conditions, and with what accountability. For enterprise AI, this includes: when the model proposes and the human validates, when the model decides and the human audits, when a human must override, and who is accountable when things go wrong.
Decision rights are not permissions or access control. They are authority and accountability. A workflow may grant a model access to data, but decision rights define whether the model can act on that data autonomously or must surface a recommendation for human approval.
Out of scope
Decision rights are not role titles, job descriptions, or RACI charts that say "responsible" without specifying the decision. They are not "the human is always in the loop" (which defers the hard question of where) or "trust the model" (which ignores accountability). Vague framing collapses under pressure.
Why it matters for enterprise AI
Most enterprises deploy AI tools without redesigning decision rights. The model generates output, but no one has authority to act on it without escalation. Or the model acts, but accountability remains with a human who no longer controls the decision. Either way, the process stalls. AI integration requires clarity: who decides, when, and who owns the outcome.
Challenge
Pick one AI-assisted process in your organisation. Then map:
- Which decisions does the model make autonomously, and who is accountable if it fails?
- Which decisions does the model recommend, and who has authority to approve or reject?
- Under what conditions must a human override, and does that person have the data and time to do so?
If the answers are unclear, the process is not AI-integrated. It is AI-adjacent.
Related
Governance series hub • Operating Model series hub
Selected essays: