AI Decision System

What it is

The AI Decision System is the governance method at the core of every AI Value Ledger review. It treats the decision, not the model, as the unit of control. A single system can support many decisions, each with a different owner, a different cost of error and a different level of acceptable autonomy.

The platform holds the register, the evidence and the scoring rules. Senior advisors run the review, test the conclusions with your leadership team and sign off every finding.

How it governs AI-supported decisions

Decision rights

Every AI-supported decision is recorded with one accountable owner, the systems behind it, how reversible a wrong call would be and the authority the business wants it to have. Ambiguous ownership is the most common reason AI cannot be trusted, so it is resolved first.

Authority earned through evidence

Authority is not granted by enthusiasm or by a vendor’s claims. Each decision is tested on seven controls, and each control is scored on what the organization can demonstrate: asserted, documented or operating. Claims without evidence are recorded but capped.

• OwnA named person is accountable for the decision and its outcome.
• KnowThe data the decision relies on is identified, and its quality is understood.
• TierThe harm a wrong decision could cause is assessed and recorded.
• ValidateThere is evidence that the output is accurate enough for the use.
• ExplainThe person who acts on the output can understand why it was produced.
• BoundLimits are set, a person can override the AI, and a tested kill switch can stop it.
• VerifyPerformance is monitored, and there is a fallback if the system fails.

The result is one of four authority levels that leadership can stand behind:

Human oversight and the kill switch

Where a decision carries legal or similarly significant effects, the review checks that a person can intervene, that solely automated decisions have a documented basis, and that a kill switch exists and has been tested within the last 12 months. Untested safeguards do not count.

The decision register

Every gap becomes a register entry with one owner, a due date and the evidence that will close it. Entries are verified, not just reported, and the register is reassessed each quarter so authority can rise as controls mature, or fall if they lapse.

How it fits the review

The AI Decision System answers the fourth board question, “Can we trust the decisions AI is making?”, alongside where AI is used, whether it is compliant and which investments to back or stop.

Inside the platform it runs through two steps of the review journey. In Decisions, the team records each AI-supported decision, its owner and its safeguards, and the scored review assigns its authority level. In the Decision register, findings become owned, dated actions that are tracked until the evidence closes them.

The review reports evidence status against the EU AI Act, ISO/IEC 42001 and the NIST AI RMF. It is not legal advice or certification. See About for the full method.