student
A formalized model for evaluating the economic efficiency of corporate analytics solutions based on large language models is presented. The study considers systems implemented using RAG (Retrieval-Augmented Generation) and Fine-tuning architectures. The cost structure associated with the deployment of these approaches in corporate information systems is analyzed, including inference costs, knowledge updates, and periodic model retraining. A simulation framework was developed to account for request intensity and the dynamics of knowledge base updates. Based on TCO (Total Cost of Ownership) modeling, the economic characteristics of the considered architectures were compared under different workload scenarios. The results allowed the identification of rational application boundaries for each architecture. It is shown that the RAG approach is more economically efficient under low request intensity, whereas Fine-tuning becomes advantageous at higher workloads due to the scaling effect of operational costs.
large language models, LLM, corporate analytics, RAG, retrieval-augmented generation, fine-tuning, economic efficiency, TCO, total cost of ownership, simulation modeling, knowledge dynamics, request intensity, information systems, AI architecture, corporate information systems.
1. Shmat A. V. Primenenie bol'shih yazykovyh modeley i tehnologiy Retrieval-Augmented Generation dlya korporativnyh assistentov // Vestnik cifrovyh tehnologiy. — 2024. — № 3. — S. 45–58.
2. Ivanov D. S., Petrova E. N. Ekonomicheskaya ocenka vnedreniya intellektual'nyh informacionnyh sistem v korporativnoy srede // Ekonomika i upravlenie. — 2023. — № 12. — S. 67–75. DOI: https://doi.org/10.36871/ek.up.p.r.2023.12.13.010
3. Kuznecov M. A. Imitacionnoe modelirovanie informacionnyh sistem predpriyatiya. — M.: Infra-M, 2022. — 256 s.
4. Sidorov A. P., Belova N. I. Cifrovaya transformaciya korporativnoy analitiki na osnove tehnologiy iskusstvennogo intellekta // Upravlencheskie nauki. — 2024. — T. 14, № 2. — S. 89–101.
5. Grigor'ev V. L. Ekonomika informacionnyh tehnologiy. — SPb.: Piter, 2021. — 304 s.
6. Arhitektura Retrieval-Augmented Generation: obzor i primenenie [Elektronnyy resurs] // Habr. — 2025. — Rezhim dostupa: https://habr.com/ru/articles/931396 (data obrascheniya: 27.02.2026).
7. RAG vs Fine-tuning: chto vybrat' biznesu i razrabotchikam v 2025 godu [Elektronnyy resurs] // ServerFlow. — 2025. — Rezhim dostupa: https://serverflow.ru/blog/stati/rag-vs-fine-tuning-chto-vybrat-dlya-biznesa-i-razrabotchikov-v-2025-godu (data obrascheniya: 27.02.2026).
8. RAG ili Fine-tuning — kak vybrat' metod dlya LLM-zadach [Elektronnyy resurs] // Napoleon IT. — 2025. — Rezhim dostupa: https://napoleonit.ru/blog/rag-ili-fine-tuning-kak-vybrat-pravilnyy-metod-dlya-nastroyki-llm (data obrascheniya: 27.02.2026).
9. Gao Y., Xiong Y., Gao X. et al. Retrieval-Augmented Generation for Large Language Models: A Survey // arXiv preprint. — 2023. — Rezhim dostupa: https://arxiv.org/abs/2312.10997 (data obrascheniya: 28.02.2026).
10. Karakurt E., Akbulut A. Retrieval-Augmented Generation (RAG) and Large Language Models (LLMs) for Enterprise Knowledge Management and Document Automation: A Systematic Literature Review // Applied Sciences. — 2026. — Vol. 16, No. 1. — Article 368. DOI: https://doi.org/10.3390/app16010368
11. Shen M., Gupta U., Zhang Y. et al. Towards Understanding Systems Trade-offs in Retrieval-Augmented Generation Model Inference // arXiv preprint. — 2024. — Rezhim dostupa: https://arxiv.org/abs/2412.11854 (data obrascheniya: 28.02.2026).
12. Devine P. ALoFTRAG: Automatic Local Fine Tuning for Retrieval Augmented Generation // arXiv preprint. — 2025. — Rezhim dostupa: https://arxiv.org/abs/2501.11929 (data obrascheniya: 28.02.2026).
13. Bergemann D., Bonatti A., Smolin A. The Economics of Large Language Models: Token Allocation, Fine-Tuning, and Optimal Pricing // arXiv preprint. — 2025. — Rezhim dostupa: https://arxiv.org/abs/2502.07736 (data obrascheniya: 28.02.2026).
14. Ren R., Li Q., Zhang T. Adaptive Two-stage Retrieval Augmented Fine-Tuning Method // Expert Systems with Applications. — 2025. — Vol. 244.
15. Robust Fine-Tuning for Retrieval Augmented Generation // Proceedings of the ACM Conference on Information and Knowledge Management. — 2025.
16. RAG vs. Fine-Tuning: Comparative Analysis [Elektronnyy resurs] // Monte Carlo Data. — 2025. — Rezhim dostupa: https://www.montecarlodata.com/blog-rag-vs-fine-tuning (data obrascheniya: 01.03.2026).
17. Should You Fine-Tune Your Large Language Models or Let RAG Do the Heavy Lifting [Elektronnyy resurs] // Centific. — 2025. — Rezhim dostupa: https://www.centific.com/blog/should-you-fine-tune-your-large-language-models-or-let-rag-do-the-heavy-lifting (data obrascheniya: 01.03.2026).
18. Fine-Tuning vs RAG Trade-offs in Large Language Models for Domain-Specific Tasks // Journal of Medical Internet Research. — 2026.
19. Lykov A.V. Economic Evaluation of RAG and Fine-Tuning Architectures [Elektronnyy resurs]. — Rezhim dostupa: https:// https:// https://github.com/MrMixaDj32/rag-ft-economic-evaluation (data obrascheniya: 02.03.2026).
