Decision criticality
Criticality score · AI decision risk score
Decision criticality is a score that rates a decision by its impact, its frequency and its irreversibility, each noted from 1 to 10. In ACF, criticality = (impact × frequency × irreversibility) ÷ 10, and the score bounds how far the decision may be delegated to an agent.
The thresholds
A worked example: a refund policy exception with impact 6, frequency 5 and irreversibility 4 gives (6 × 5 × 4) ÷ 10 = 12, in the green range. The same refund for a large amount with impact 9 and irreversibility 7 gives (9 × 5 × 7) ÷ 10 = 31.5, which goes to a person.
| Score | Reading | Consequence |
|---|---|---|
| Under 15 | Green | The decision can go to the agent alone |
| 15 to 25 | Orange | The decision is taken under supervision |
| Above 25 | Red | A human decides; the agent can prepare and propose |
What the score does not say
The score marks how far you may go and says nothing on where you should stop inside that range. The level retained still depends on the mandate, on how reversible the act really is, on numeric ceilings and on the quality of the supervision. The score informs a judgment that remains human: the DDAO settles and answers for it.
Decisions are also rated by those they affect, because an impact has no abstract value. The decision map notes who bears the effects: the customer, the accounts, the margin or the company's image.
A limit the method assumes openly
Multiplying by frequency gives a rare event a low score mechanically, even when the event is catastrophic. In a set of public incidents analyzed by ACF, the note sometimes stays below the red threshold although the act belongs in the prohibited zone. This is the reason non-delegable decisions are removed before any calculation. Principles come first and scores afterward.
Frequently asked questions
Who rates the criticality of a decision?
The people who know the decision and those who bear its effects, not the team that wants to deploy the agent alone. The DDAO validates the outcome.
Is the criticality score the same as the AI Act risk level?
No. The AI Act classifies a system by its use, whereas criticality rates each decision individually, so a single system can make decisions of several levels.
Where does your organization stand?
ACF Score measures your agentic governance in about 10 minutes, the AI Act checker tells you which obligations apply, and ACF Control applies your rules before an agent acts.
Related terms
Written by Vincent Dorange, creator of the Agentic Commerce Framework®. Last updated .