KPI2KVI: LLM-Powered Tool for Computing Key Value Indicators from Service Descriptions
KPI2KVI, a newly developed tool, leverages a multi-agent workflow driven by Large Language Models (LLMs) to derive Key Value Indicators (KVIs) from service descriptions in natural language. KVIs provide a decision-focused perspective by encapsulating how operational performance correlates with stakeholder value, risk, and outcomes. The process of calculating KVIs is frequently manual and inconsistent, as it requires the selection of relevant categories, the definition of measurable Key Performance Indicators (KPIs), value collection, and the application of calculation logic. KPI2KVI automates this process by coordinating deterministic agents that gather missing context, extract KVI categories from a taxonomy, create service-specific KPIs with units, and compile values. The research, available on arXiv (2605.22825), showcases the tool's ability to enhance service evaluation across various domains.
Key facts
- KPI2KVI transforms natural language service descriptions into computed KVI estimates.
- It uses a deterministic multi-agent workflow powered by LLMs.
- The workflow elicits missing service context, extracts KVI categories from a taxonomy, generates KPIs, and collects values.
- KVIs summarize operational performance into stakeholder value, risk, and outcomes.
- Computing KVIs manually is often inconsistent and difficult.
- The paper is published on arXiv with ID 2605.22825.
- The tool aims to automate and standardize KVI calculation.
- It is designed for domains where service documentation is unstructured.
Entities
Institutions
- arXiv