Total Economic Costs of Artificial Intelligence: An Optimization Model of Compliance and Investment Efficiency in High-Risk Systems

Abstract

Abstract This monograph develops the Total Economic Cost (TEC) model, a formal optimization framework for analyzing regulatory compliance and investment efficiency in high-risk artificial intelligence systems under the European Union’s Artificial Intelligence Act (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744). The adoption of the AI Act has transformed compliance from a purely legal obligation into a significant economic decision. Organizations developing or deploying high-risk AI systems now face dual pressure: rising costs of prevention – testing, certification, cybersecurity, documentation, and training – and increasingly tangible costs of non-compliance, including administrative sanctions, reputational damage, and lost market opportunities. The absence of an analytical tool for determining the economically justified level of compliance investment leads to two adverse outcomes: underinvestment, which exposes organizations to high expected risk, and overinvestment, which diverts resources from innovation and growth. The central research question is therefore: what is the optimal level of investment in regulatory compliance at which total economic costs are minimized? The TEC model addresses this question by integrating three core cost components into a single optimization framework. First, the cost function represents the direct and indirect costs of investing resources in prevention and compliance measures. Second, the expected loss function denotes the expected ethical and reputational loss associated with a given level of investment. Third, the residual regulatory risk function captures the residual regulatory risk stemming from institutional and regulatory uncertainty, including the risk of non-compliance and enforcement. The total economic cost is defined as the sum of these components, and the organization seeks to minimize it subject to resource constraints. The model is built on a rigorous axiomatic foundation: the properties of the cost function, the expected loss function, and the residual regulatory risk function are specified, including monotonicity and convexity conditions, and the existence and uniqueness of the optimal investment level are proven through convexity analysis. The monograph makes several original contributions. Conceptually, it integrates ethical, regulatory, and economic perspectives into a single decision-theoretic model, extending the economic theory of regulation to the specificities of high-risk AI. It introduces the concept of residual regulatory risk as a new measure of institutional uncertainty, building on traditional compliance cost analysis. Methodologically, it provides a formal optimization model with proven properties of existence and stability, identifies two regimes of economic inefficiency – overinvestment and underinvestment (the compliance trap) – and proposes a calibration methodology combining Monte Carlo simulations and the Delphi method. The model is further extended to an integrated formulation where auditability is treated as institutional capital, linking compliance to investment efficiency and organizational learning. An auditability index is operationalized through three complementary components: documentation, data traceability, and transparency. An investment matrix systematizes management decisions into five zones according to the ratio between marginal costs and benefits. The monograph is structured in four chapters, each building on the previous one to develop a coherent analytical framework. Chapter 1 establishes the economic and regulatory context by systematically examining the AI Act’s risk taxonomy, distinguishing between prohibited, high-risk, limited-risk, and minimal-risk systems, and identifying the organizational and technical obligations imposed on providers and deployers of high-risk AI. It reviews classical foundations of the economic theory of risk, including expected utility theory, risk aversion, and the distinction between insurable and non-insurable risks, and then analyzes the specificities of AI risk – its systemic, non-linear, and dynamic character – which distinguishes it from conventional operational or financial risk and complicates actuarial pricing. A dedicated section examines insurance market failure, explaining the conditions under which private insurance markets fail to provide adequate coverage for AI-related harms and why this failure leads organizations to internalize risk through preventive investment. The regulatory framework is then interpreted as a generator of costs, with an economic reading of compliance obligations as fixed and variable cost drivers. The chapter develops a taxonomy of prevention and non-compliance costs, distinguishing between ex ante prevention costs, ex post sanction costs, reputational costs, and opportunity costs. It further examines the microeconomic behavior of firms under regulatory uncertainty, including investment hysteresis, option value of waiting, and strategic behavior toward regulators, and concludes by identifying the research gap: the absence of a formalized optimization model that simultaneously integrates prevention costs, expected losses, and residual regulatory risk. Chapter 2 constitutes the theoretical core of the monograph by presenting the TEC model in full. It begins with the definition, objectives, and basic structure of the model, specifying the decision variable as the level of investment in compliance and prevention. The conceptual framework is then elaborated, explaining how the three cost components – the cost function, the expected loss function, and the residual regulatory risk function – interact and how their sum defines the total economic cost. The axiomatic structure is presented in detail: the axioms specify that the cost function is increasing and convex, the expected loss function is decreasing and convex, and the residual regulatory risk function is decreasing and convex, capturing institutional uncertainty and the risk of non-compliance. These properties together guarantee that the total economic cost is strictly convex and that a unique optimal investment level exists. The chapter then provides an economic interpretation and regulatory conclusions, deriving the condition for optimal investment, which requires that marginal cost equals marginal benefit, and proving the existence and uniqueness of the optimum under strict convexity. It analyzes comparative statics, showing how changes in regulatory stringency, ethical preferences, and technological uncertainty affect the optimal level of investment. The chapter also introduces the compliance trap and excessive compliance investment as two regimes of inefficiency, and compares the TEC model with classical models of regulatory compliance, including deterrence models, principal – agent models, and cost-benefit analysis, highlighting the TEC model’s originality in integrating residual regulatory risk. Chapter 3 operationalizes the model for managerial decision-making by presenting TEC as an objective function and management guideline, explaining how the optimal investment level translates into budgetary and strategic decisions. It translates TEC into investment language, linking the model to Net Present Value (NPV), Internal Rate of Return (IRR), and Payback period, and showing how compliance investments can be evaluated using standard capital budgeting techniques. A key innovation is the introduction of auditability as institutional capital: the ability of an organization to demonstrate compliance through documentation, data traceability, and transparency. The chapter develops the integrated TEC model with auditability, analyzing how investment in auditability reduces residual regulatory risk and lowers the cost of capital. It further examines strategic optimization and market failures, including the role of regulatory uncertainty, information asymmetries, and reputational externalities. A methodology for calibrating the TEC model with real data is proposed, combining Monte Carlo simulations for stochastic parameter estimation and the Delphi method for expert elicitation. The chapter also addresses organizational aspects of TEC implementation, including change management, governance structures, and performance metrics, and concludes with a comparative analysis of TEC and other quantitative methods for compliance assessment, such as cost-benefit analysis, risk matrices, and real options valuation. Chapter 4 demonstrates the model’s versatility across five sectors. The financial sector section examines fintech and banking, focusing on credit scoring, fraud detection, and algorithmic trading, where regulatory compliance and explainability are critical. The healthcare section analyzes AI diagnostics and clinical decision support, emphasizing patient safety, data privacy, and liability. The public sector section explores AI in public administration, including welfare allocation and law enforcement, where transparency and accountability are paramount. The section on small and medium-sized enterprises (SMEs) and small mid-cap enterprises addresses resource constraints and the need for scalable compliance solutions. The insurance sector section examines AI for underwriting, claims processing, and damage assessment, highlighting the challenges of actuarial fairness and regulatory approval. The chapter synthesizes cross-sectoral differences, identifying common patterns and sector-specific parameters, and derives conclusions for public policy, including recommendations for regulators, standards bodies, and organizational leaders. The Conclusion summarizes the main contributions of the monograph, restating the research problem, the TEC model, and its theoretical and practical implications. It emphasizes the novelty of integrating prevention costs, expected losses, and residual regulatory risk into a single optimization framework, and the operationalization of auditability as a second