ECONOMIC EFFICIENCY OF ARTIFICIAL INTELLIGENCE TECHNOLOGIES IN INDUSTRIAL ENTERPRISES

Abstract

The adoption of artificial intelligence in industrial enterprises has so far concentrated on isolated functional modules — predictive maintenance, demand forecasting, or vision-based inspection — implemented independently and evaluated against narrow, function-specific metrics. This fragmentation obscures the interaction effects that arise when maintenance, scheduling, and resource-allocation decisions are made jointly rather than sequentially, and it leads enterprises to systematically undervalue the economic return of AI investment. This study proposes Cognitive Production Optimization (CPO), an integrated decision-support architecture in which a shared state representation of equipment condition, order backlog, and labor availability feeds three coupled reinforcement-learning-based modules — predictive maintenance scheduling, adaptive production sequencing, and dynamic resource reallocation — coordinated through a composite optimization index rather than operating as independent agents. The framework was evaluated on a representative discrete-manufacturing production line through comparative modeling of three scenarios: manual baseline operation, a conventional single-function predictive-maintenance deployment, and the proposed integrated CPO system. The integrated approach raised overall equipment effectiveness from 61.4% to 79.2%, reduced unplanned downtime by 62%, and generated a five-year net present value nearly three times that of the single-function deployment, with an internal rate of return of 98.7%. These results indicate that the economic value of industrial AI derives less from any individual algorithm than from the coordination architecture linking algorithmic decisions across functional domains.