Risk-based evaluation of the economic efficiency of large language models in the service sector of the Republic of Belarus

Abstract

The paper presents an analysis of the transformation of national business under the influence of Large Language Models (LLMs), which act as an internal catalyst for labor intensification in the service sector of the economy of the Republic of Belarus. Based on a synthesis of system analysis and economic-mathematical modeling, the concept of a production function with constant elasticity of substitution has been adapted to assess the potential for displacing routine operations with intelligent agents. A high determination between the integration of LLM solutions and operational productivity is empirically confirmed. It has been established that in the conditions of the Belarusian market, the implementation of large language models allows minimizing the time for processing client data from several hours to one and a half minutes. Quantitative assessment demonstrates the possibility of substituting labor resources comparable in scale to the activities of hundreds of full-time specialists. Original models for calculating risks from LLM implementation have been formulated, taking into account not only direct benefits but also probabilistic distributions of damage from specific cyber risks. The findings can be used by top management during the transition to AI-oriented management models under conditions of limited access to hardware capacities and strict requirements of the national regulator.