AI Decision Review // Pilot places for 2027

When an Automated Decision Goes Wrong

Boards are approving systems that make or shape decisions. When one fails, the board will ask what the business believed at the time. Most governance rules cannot answer that. This review can.

5 of 163. In a 2025 University of Cambridge study of the AI Incident Database, only 5 of the 163 reports tagged as responses met the definition of an official response from the company involved. Even those lacked the detail needed to learn from them.

Richards, Benn and Zilka, From Incidents to Insights, University of Cambridge, 2025. Read the paper

Blame tends to land on the nearest person, even when they had little control. Researchers call this the moral crumple zone.

3 ways in

Choose Where You Are

Settle the right question before approval. Learn properly after a failure.

Before approvalPrecisely Wrong: A Board SessionFor a board or risk committee about to approve an AI system.From US$7,500

A session for directors before they approve an AI-assisted decision or system. It settles the question to answer first: what the system will decide, from which records, and who answers when it is wrong.

The board leaves with a written standard for what it needs to see before it approves, and before it relies on the output.

After a failureA 4–6 Week ReviewFor an automated decision that has already gone wrong.Pilot from US$12,000

The discipline of our investigations, applied to one automated decision. We establish what the business believed at the time, how the decision was actually made and where the defences had gaps. The finding names conditions, not a scapegoat.

You receive a written finding, the changes with named owners and a record the board can show regulators and shareholders.

OngoingOversight, Reported to the BoardFor boards that rely on AI-assisted decisions every day.By scope

A regular review of how AI-assisted decisions are approved, monitored and corrected, reported straight to the board, not through the executives it measures.

Pilot places for 2027 are open to a small number of boards. Rand prices are on Engagements.

Governance context

Where King V Comes In

King V asks boards to oversee decisions that AI systems influence or make. We tie the review to the code only where it helps a board.

What the code asks, and what the review gives the board3 points, in brief.

King V asks

The board governs data, information and technology, AI included, with clear accountability for the decisions, actions, outputs and outcomes of AI systems (Principle 10).

The review gives

A dated record of who decided what, on what information.

King V asks

AI is used with ethics, human centricity, accountability, transparency, explainability, security, privacy, fairness and trustworthiness.

The review gives

A test of the decision against the values that matter for it.

King V asks

The board explains any recommended practice it has not adopted, and concludes on whether its governance outcomes were achieved.

The review gives

Evidence for that explanation and that conclusion.

King V was published by the Institute of Directors in South Africa in 2025. This is a summary for context, not legal advice.

Ask for a Pilot Place

Tell us about the decision you are about to approve, or the one that went wrong. The first meeting is 90 minutes, with no fee.