“LLM4IMPACT: ANALYSIS OF THE IMPACT OF IT SOLUTIONS ON BUSINESS PERFORMANCE USING LARGE LANGUAGE MODELS”

Abstract

The study presents LLM4Impact, an innovative hybrid methodology for analyzing the business impact of IT solutions by integrating Large Language Models (LLMs) into traditional business analytics frameworks. Modern analytical tools such as BI dashboards or KPI-based monitoring systems provide quantitative insights into business performance but often overlook semantic dimensions — the underlying intent, sentiment, and contextual meaning contained in textual data. The proposed approach addresses this gap by combining financial indicators (ROI, NPV, KPI) with linguistic analysis of unstructured business data, including customer feedback, project documentation, and system requirements. Using the reasoning and semantic understanding capabilities of LLMs (such as GPT-5), the method evaluates both the measurable and qualitative effects of IT-driven digital transformation initiatives.

A core contribution of the research is the introduction of the Business Impact Index (BII) — a composite metric that integrates quantitative efficiency indicators with qualitative sentiment and user experience scores. This index enables organizations to assess IT solutions not only in terms of financial returns but also in their influence on customer satisfaction, process transparency, and innovation potential. The experimental evaluation, conducted on a mid-size company implementing a CRM system, revealed that the LLM4Impact model improved decision-making efficiency by 35%, increased sentiment-based satisfaction metrics by 28%, and raised the overall business impact score by 19%.

The results demonstrate that integrating LLMs into business analytics can significantly enhance the depth, interpretability, and contextual validity of management insights. Beyond automation, LLM4Impact contributes to the evolution of intelligent decision support systems, capable of connecting numerical data with human-centric understanding. This hybrid paradigm paves the way for more adaptive, transparent, and cognitively enriched business analytics ecosystems.