arXiv cs.AIMonday · May 25, 2026FREE

KPI2KVI: A Multi Agent Workflow for Calculating Key Value Indicators from Service Descriptions

KPI2KVI, introduced in arXiv paper 2605.22825, is a tool that converts unstructured service descriptions into computed Key Value Indicator (KVI) estimates. It orchestrates a deterministic multi-agent workflow powered by LLMs to: (i) elicit missing service context, (ii) extract and finalize relevant KVI categories from a taxonomy, (iii) generate service-specific KPIs with units and descriptions, (iv) collect KPI values through interactive dialogue and intelligent estimation for unavailable values, and (v) compute interval-valued KVI outputs (minimum, exact, maximum) with traceable explanations. The system addresses the challenge of manually and inconsistently computing KVIs from service documentation, which typically requires selecting categories, defining KPIs, collecting values, and applying calculation logic. By automating these steps, KPI2KVI aims to provide consistent, decision-oriented summaries of stakeholder value, risk, and outcomes. The paper was published on arXiv on May 25, 2026, under the cs.AI category.