Inference Cost per Successful Action (ICSA)
ICSA = TC / SA — The computational cost incurred per successful action or transaction mediated by AI systems.
Description
Inference Cost per Successful Action measures the economic efficiency of AI-mediated transactions. ICSA = TC / SA where TC is total computational cost and SA is successful actions. Lower ICSA indicates more efficient AI-mediated commerce. ICSA helps evaluate whether AI-mediated transactions are economically viable.
Related Concepts
Related Research
The Emerging Architecture of AI-Mediated Markets
The Emerging Architecture of AI-Mediated Markets proposes a conceptual framework for understanding how AI systems participate in economic markets as intermediaries, reasoning agents, and action coordinators. The framework identifies four distinct layers—Representation, Reasoning, Action, and Governance—that must work together for AI-mediated markets to function safely and efficiently. Each layer has specific requirements, failure modes, and design considerations. The Representation Layer encodes market-relevant information in machine-readable form. The Reasoning Layer processes this information to support decision-making. The Action Layer executes market transactions with appropriate constraints. The Governance Layer ensures safety, fairness, and accountability. This framework synthesizes insights from property markets, hospitality, and other domains to propose general architecture principles applicable to any AI-mediated market.
The Zero-Click Economy
The Zero-Click Economy examines how AI-mediated discovery, selection, recommendation, verification, and action alter the transmission of economic signals from policy and demand to firms, assets, households, sectors, and jurisdictions. We introduce the Current Reporting-Period Hypothesis, which states that AI systems construct consideration sets from representations as they exist at inference time, not from the period the policy or demand signal was emitted. This creates Computational Transmission Attrition—policy or demand-induced signals may attenuate, misallocate, or leak before reaching intended economic targets. We formalize Dynamic Computational Risk as the interaction between exposure (dependence on AI-mediated allocation), technological velocity (rate of change in AI-mediated discovery), financial sensitivity (margin of capital, liquidity dependence), and adaptation capacity (speed of organizational response). The paper consolidates the Representation Economy measurement stack: Agent Readiness Index (ARI), Global Agent Readiness Index (GARI), Zero-Click Exposure Index (ZCEI), Platform Dependency Index (PDI), Computational Business Risk Index (CBRI), Dynamic Computational Risk Index (DCRI), Enterprise Adaptation Velocity Index (EAVI), Computable Asset Ratio (CAR), National Computable Economy Index (NCEI), Sovereign Adaptation Velocity Index (SAVI), and sovereign outputs including Compound Regional Adaptation Velocity Index (CRAVI), Global Computable Economy Index (GCEI), Sovereign Adaptation Gap (SAG), and Dynamic Monetary Sovereignty Risk Index (DMSRI).