FinOps Agent - A Use-Case for IT Infrastructure and Cost Optimization
Ngoc Phuoc An Vo, Manish Kesarwani, Ruchi Mahindru, Chandrasekhar Narayanaswami · 2025
Read FinOps Agent - A Use-Case for IT Infrastructure and Cost Optimization by Ngoc Phuoc An Vo, Manish Kesarwani, Ruchi Mahindru, Chandrasekhar Narayanaswami (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-25914.txt.
Introduction
Ngoc Phuoc An Vo, Manish Kesarwani, Ruchi Mahindru, and Chandrasekhar Narayanaswami build an AI agent to help firms control cloud spending. FinOps, short for finance plus operations, is the practice of making engineering, finance, and business teams share responsibility for cloud value. The hard part is that billing data arrive in mixed formats and measures from many providers, yet choices must be made fast.
Vo and the team simulate a full industry process, from pulling data and merging it to analysis and cost saving advice. They define scores to judge the agent and test several open and closed language models. The agent understood, planned, and carried out tasks at a level matching a human FinOps practitioner in their setup.
Catalog record
| CREATOR | Ngoc Phuoc An Vo, Manish Kesarwani, Ruchi Mahindru, Chandrasekhar Narayanaswami |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 41 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-25914
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
FinOps Agent - A Use-Case for IT Infrastructure and Cost Optimization Ngoc Phuoc An Vo, Manish Kesarwani, Ruchi Mahindru, Chandrasekhar Narayanaswami Abstract FinOps (Finance + Operations) represents an operational framework and cultural practice which maximizes cloud business value through collaborative financial accountability across engineering, finance, and business teams. FinOps practitioners face a fundamental challenge: billing data arrives in heterogeneous formats, taxonomies, and metrics from multiple cloud providers and internal systems which eventually lead to synthesizing actionable insights, and making time-sensitive decisions. To address this challenge, we propose leveraging autonomous, goal-driven AI agents for FinOps automation. In this paper, we built a FinOps agent for a typical use-case for IT infrastructure and cost optimization. We built a system simulating a realistic end-to-end industry process starting with retrieving data from various sources to consolidating and analyzing the data to generate recommendations for optimization. We defined a set of metrics to evaluate our agent using several open-source and close-source language models and it shows that the agent was able to understand, plan, and execute tasks as well as an actual FinOps practitioner. Introduction Cloud infrastructure has become central to the operation of many modern enterprises. The resulting ease of deployment and flexibility of scaling infrastructure as needed, coupled with dynamic pricing models from hyperscalers has created unprecedented complexity in IT cost management. This has given rise to the discipline of FinOps, which drives strategic business value creation, not just operational cost savings. FinOps (Finance + Operations) represents an operational framework and cultural practice that maximizes cloud business value through collaborative financial accountability across engineering, finance, and business teams. The framework operates through three iterative phases: Inform (Visibility & Allocation), Optimize (Rates & Usage), and Operate (Continuous Improvement & Usage). Importantly, success in FinOps hinges on making continuous adaptive changes using real-time insights with data gathered from Application Performance Management (APM), Application Resource Management (ARM), and Finance (Cloud Cost Management) ( Storment and Fuller 2023 ) . The FinOps Foundation’s 2025 State of FinOps report ( Trask 2025 ) curated data from 861 respondents totaling $69 billion in public cloud spend, with 31% currently spending more than $50 million annually in public cloud, 20% spending more than $100 million, and over 20 organizations spending $1 billion annually. Not surprisingly, AI spending management has emerged as a critical concern, with 63% of respondents now managing AI costs — up from 31% the previous year — signaling a fundamental shift in FinOps priorities. Similar to the observations made in the 2024 State of FinOps Report, workload optimization and waste reduction remained key priorities across the board. Management of cloud discount programs (such as Savings Plans and Reserved Instances) and accurate forecasting of spend continued to be top concerns. FinOps practitioners face a fundamental challenge: relevant data arrives in heterogeneous formats, taxonomies, and metrics from multiple cloud providers, other IT vendors, and internal systems. Synthesizing actionable insights from these disparate sources requires understanding complex relationships between IT bills, resource utilization patterns, observability data, IT resource types, cost and performance KPIs, hyperscaler pricing constructs, organizational hierarchies, etc. Time-sensitive decision-making compounds this complexity, as delayed insights can result in significant financial impact. We note that the disciplines of FinOps and SRE are intertwined due to the interplay between cost and performance. In order to assist practitioners, the FinOps Foundation ( FinOps Foundation 2025a ) provides benchmarks that compare cloud financial performance across organizations and departments, focusing on KPIs such as resource utilization efficiency, contract coverage, and cost apportionment. These benchmarks help assess cloud efficiency by evaluating internal and external metrics, fostering structured, collaborative approaches to cloud optimization. Several tools, platforms, and standards have also emerged in this space. Two standards have emerged - FOCUS (FinOps Open Cost and Usage Specification) for providers to offer billing data in a consistent format ( FinOps Foundation 2025c ) , and TBM (Technology Business Management) ( TBM Council 2025 ) - “To promote alignment between IT, Finance, and Business Unit leaders, TBM provides a standard taxonomy to describe cost sources, technologies, IT resources (IT towers), applications, and services. Just as businesses rely on generally accepted accounting principles (or GAAP) to drive standard practices for financial reporting — and thus comparability between financial statements — the TBM taxonomy provides a generally accepted way of categorizing and reporting IT costs and other metrics.” The hyperscalers now provide tools for monitoring costs such as AWS Cost Explorer, GCP Cost Explorer, Microsoft Cost Management, and Oracle’s FinOps Hub. Independent software vendors such as CloudZero, Apptio, ServiceNow, ProsperOps, Nicus, Magic Orange, etc., also provide tools to monitor and manage IT spending. These tools typically ingest data from multiple sources and provide dashboards, provide forecasts, identify and flag trends, categorize expenses, compute chargebacks, plan capacity, etc. Introduction for Agentic AI Recent advances in Agentic AI offer a transformative approach to these challenges. Agentic AI introduces autonomous agents capable of goal-directed behavior, leveraging strategic planning, and adaptive learning ( Yao et al. 2023a ; Wang et al. 2024 ) , rather than just reacting literally to direct user prompts. These agents leverage the natural language understanding of large language models while adding critical capabilities: they can formulate multi-step plans, include reflection loops and learn from feedback, execute actions through tool integration, retrieve data from various systems, have memory to keep state over time, and adapt strategies based on currently observed conditions. The above capabilities can enhance FinOps by allowing proactive optimization and assisting reactive analysis by providing detailed explanations, both of which are essential to practitioners. In the realm of proactive optimization, an agent could continuously monitor cloud spending, autonomously investigate irregularities, correlate costs with business events, and execute approved optimizations. For example, when detecting an unusual spike in compute costs, an agent can autonomously trace it to specific services, analyze recent deployments, identify inefficient resource configurations, evaluate multiple optimization strategies considering both cost and performance impacts, and either present detailed recommendations with predicted savings or implement approved corrective actions, all while maintaining the audit trail required for financial governance. The methodology used by Agentic AI allows it to naturally show explorations and decisions step by step and provide explainability in a graphical manner. While existing agent frameworks like ReAct ( Yao et al. 2023b ) have shown promise in general task automation, applying agentic AI to FinOps requires addressing domain-specific challenges, including understanding complex pricing models, maintaining financial compliance, and balancing multiple stakeholder objectives. AI Agents for FinOps To further address the needs of FinOps practitioners, we have embarked on the path of creating an AI-based FinOps Agent that can help in various day to day tasks they encounter. Our agent interprets natural language user queries to perform complex analytical tasks on IT data obtained from multiple sources, and when necessary interacts with other agents or hands off certain tasks to other agents, such as an SRE agent. Major cloud providers are also integrating such agents into their platforms to enhance FinOps practices and, in turn, enable users to manage cloud spend across their portfolios ( FinOps Foundation 2025b ) . For example, AWS offers Cost Explorer ( AWS 2025 ) and Cost Anomaly Detection ( AWS 2024 ) , which FinOps practitioners can use to recommend ideal compute