INSIGHTS / THE MERIDIAN SERIES
03 / 16
The Decision Basis
An enterprise can preserve a decision and still lose the reason it made it. Risk approvals, investment requests, architecture reviews, provider contracts, and treatment plans serve different purposes and sit with different teams. They can be complete yet leave one question unanswered when conditions change: what did leadership agree to carry, on what terms, and why?
Consider a company deploying an agentic customer-service system for account, order, and payment-dispute requests. The agent retrieves customer records, queries order status, and invokes a limited payment-support function. Its service identity restricts access; external model and observability providers receive prompts and traces. The organization believes its controls keep cardholder data outside the external data path.
A customer-supplied attachment contains untrusted instructions. The agent invokes the payment function with broader context than intended. Cardholder data enters a model prompt, and a provider retains a trace outside the approved PCI data path. The investigation confirms a bounded disclosure and a PCI non-compliant condition. The organization can find the incident record, trace, identity design, provider terms, and launch approval. The harder question is whether they still show the same decision.
The launch approval may state a defined service scope. The security design shows the service identity and tool permissions. Procurement retains provider terms; compliance records the PCI control requirement; incident response records the prompt and trace. None of those records alone explains the relationship between the service, the condition believed to be controlled, the exposure accepted, and the authority that allowed it to operate.
That relationship is the Decision Basis. It is the connected body of scenario logic, evidence, assumptions, decision terms, and expected result that explains why a particular course of action was defensible. It establishes more than that a decision occurred. It shows the business condition being addressed, the conditions on which the decision relied, the exposure left in force, and the facts that would require the enterprise to reconsider it.

In the agentic AI case, a new decision may allow the agent to resume account and order work while the payment-support connection is removed and the data boundary is redesigned. Its basis has to retain the customer-service demand that makes limited restoration necessary; the agent trace and provider-retention evidence that establish the failure; the required changes to identity, tooling, and provider handling; the residual exposure carried during the work; the appetite condition and authority for carrying it; and the evidence required before the payment workflow can be restored. These are not separate administrative details. They are the conditions that give the restriction, treatment, and approval their meaning.
The point is not to collect every related record in one repository. The incident record, engineering work, provider documents, compliance evidence, and risk approval may remain under different owners and systems. The requirement is that the relationship among them remains recoverable. A later reviewer needs to determine whether the identity change occurred, whether the provider still handles traces as assumed, whether the payment connection remains removed, and whether the conditions that permitted limited operation still hold.
Change management has a necessary but narrower role. It can show that the agent, its identity, or the payment connection was requested, approved, tested, and changed. It does not ordinarily preserve the risk decision that made those changes consequential: the exposure, provider assumption, residual condition, and authority behind limited operation. Meridian is not change management. It uses change activity as evidence for determining whether the decision remains valid.
That distinction matters when the enterprise changes around the decision. An external provider can revise its retention practices. Engineering capacity can move to a higher-priority incident. A product team can expand the agent's service scope before the redesigned boundary is complete. Each event affects the decision in force. If the underlying conditions cannot be recovered, the organization can retain a full set of accurate records without knowing whether it is still carrying the exposure it agreed to carry.
A Decision Basis makes a decision governable after approval. It preserves the connection between the business condition, the evidence, the terms of the decision, and the result the enterprise expected to achieve. Without that connection, subsequent work can remain locally reasonable while the original decision becomes progressively harder to evaluate.