Prediction estimates what may happen. Execution creates institutional consequence. Decision Architecture determines what an AI system may influence, decide and cause — and under which controls.
Decision Architecture separates AI used to predict, AI used to assist employees, and AI authorized to make or execute decisions. The distinction matters when an output can create financial, regulatory, operational or reputational consequences.
PredictionProduces an estimate or signal.
AssistanceSupports a person who retains decision authority.
Decision and executionDetermines or causes an institutional outcome.
The engagement depends on both the system’s role and the project stage: adoption decision, pilot support or production audit.
Can this process support the authority the organization intends to give AI?
Readiness · Materiality · Supported Authority · Authority GapWhich data, relationships and inferred patterns may influence this decision?
Data Eligibility · Purpose Boundary · Influence Map · Control GapsWhat exposure is created by the vendor’s position in the decision process?
Access · Decision Influence · Dependency · Architectural ExposureThird-party exposure principleThe vendor’s characteristics do not determine the organization’s exposure on their own. Exposure is created by the data, access, authority and architectural position the organization gives the vendor.
The assessment system is grounded in a connected body of published work on permissible learning, delegated authority and controlled execution.
Board-level papers examining where AI efficiency becomes exposure, how authority should be bounded and what institutions need to evidence before execution.
AI is already producing measurable process-level productivity gains in financial services. This paper examines when AI-enabled processes can create material financial, operational, regulatory, customer or reputational exposure and the control architecture required before greater authority is exercised.
Dušica Živanović holds an MA in Economics and has more than 17 years of experience in banking and financial services, spanning business processes, compliance, risk and regulatory implementation. Her current work focuses on translating business, legal, risk, data and process requirements into executable decision boundaries for AI systems.
LinkedIn profileDescribe how the decision is currently produced, which information supports it and which consequences it can create. This provides the basis for targeted follow-up questions and risk mapping.