Agentic AI shifts cloud FinOps from reactive to proactive

Insights

• Companies lose millions of dollars due to waste of cloud infrastructure.
• Underutilization of cloud resources and hidden costs such as egress fees, application programming interface (API) calls, retrieval charges, and mismanagement are key causes.
• Human resources and artificial intelligence (AI) for managing cloud financial operations or FinOps are reactive. Agentic AI can monitor and act early to prevent underutilization and reduce expenditure.
• Key considerations for the adoption of agentic AI include governance guardrails, integration with existing FinOps tools, security, and change management.

Enterprises are scaling cloud usage across public, private, and software-as-a-service (SaaS) footprints, yet underutilization and pricing complexity continue leading to waste and cost overruns. According to Infosys research, 53% of organizations’ cloud resources lie underutilized or idle, underscoring persistent efficiency gaps.

Agentic AI — multiagent systems capable of perceiving context, setting goals, and acting with guardrails — offers a path to continuous optimization, anomaly response, and proactive governance when embedded within the FinOps framework. FinOps brings together business, technology, and finance operations to manage various aspects of cloud services.

Why companies are failing to utilize cloud space

Cloud adoption is increasing across industries. According to research in the European Journal of Computer Science and Information Technology, global cloud spending reached $623 billion in 2023 and is projected to exceed $725 billion by 2026. You would think companies paying so much for cloud services would utilize every ounce of it. The reality is very different, as shown in the Infosys survey. In 2022, researchers from Texas State University and Westwood High School in the US analyzed costs across 6,687 Azure cloud users who created nearly 2.7 million virtual machines (VMs). They found that a large portion of VMs was either underutilized or overprovisioned.

Cost overruns are becoming more common.

According to Wasabi 2025 Global Cloud Storage Index, 62% of organizations exceeded cloud storage budgets in 2024, up from 53% in 2023, with some reporting cost overruns of up to 50%.

The Wasabi report also points to why these overruns persist. For example, the report found that of the money spent by companies on cloud services, only 51% went toward storage of data, while 49% was spent on data retrieval and access fees. This shows how organizations continue to struggle with fee complexity, including networking fees, API calls, operations, egress, and retrieval costs.

AI can help, but only to an extent

Cloud FinOps has traditionally been handled by specialists who analyze cloud usage patterns, pricing structures, and future demand to ensure optimal usage of cloud services at minimum cost. However, the growing scale of cloud deployment has made this increasingly difficult for humans to manage.

According to the Journal of Computational Analysis and Applications (JCAA), the amount and speed of data resulting from cloud usage is beyond the capability of human analysts. This challenge is compounded by microservices, API-first, cloud-native, headless (MACH) architecture, in which applications are broken into small, independent services that communicate over APIs. While this modular approach improves flexibility, it makes cost allocation and prediction difficult due to its distributed nature compared to traditional, monolithic architecture.

The cloud FinOps teams depend largely on manual analysis of cloud usage patterns. They typically review costs after they have already been incurred. However, this is where AI can help: The JCAA analysis shows that artificial intelligence (AI) makes cloud FinOps easier by allowing for the analysis of huge amounts of data related to usage in real time. It helps find the nuanced correlations between business activities and the infrastructure consumption not perceived by humans.

Deep learning models are good at recognizing patterns in time, which can lead to better accuracy of forecasts for variations in spending due to seasons and cycles.

AI can learn to spend less on the cloud while keeping performance good through experience or a reinforcement learning (RL) approach. RL trains AI agents by making them interact with the environment and learn from its successes and failures. This is like the way humans learn from their environment, as they receive a reward for success, and a penalty for mistakes. Over time, AI agents learn the best actions in different situations.

In cloud FinOps, RL can help AI look for optimization options, finding ways for cutting costs that traditional analysis might overlook. However, data on AI’s successes in managing cloud FinOps show only limited success.

According to research from the George Institute of Technology, industries with larger and more complex cloud footprints are realizing greater optimization potential. Financial services organizations, for example, have realized the most substantial benefits, achieving average cost reductions of 31.4% within 12 months of implementation. This is followed closely by technology companies at 28.6%, healthcare providers at 26.2%, and manufacturing companies at 23.7%.

Agentic AI is the future of FinOps. Here’s why:

Traditional AI tools can analyze cloud usage and costs and highlight anomalies, but they cannot take independent decisions. This often delays corrective action, driving up costs. Agentic AI, by contrast, analyzes cloud usage patterns, predicts anomalies, and executes optimizations in real time, leading to savings. An agentic AI FinOps system employs multiagent architectures, where different agents handle distinct aspects. For example, data collection agents continuously monitor cloud environments, cost analysis agents track spending patterns, and other agents convert insights into clear recommendations or auto-execute approved cost-saving actions.

Infosys experiments show the impact: cleaning idle resources and optimizing reserved capacity purchases, delivering savings between 18% and 22% within weeks. A US semiconductor company used AWS pricing models, reserved instances (RI) and savings plans (SP), to save between 35% and 40% on a monthly spend of $1.1 million. A North American dairy company cut $50,000 in one month by acting on the recommendations of the Trusted Advisor AI agent. These results show that agentic AI can drive significant cost efficiency and operational resilience.

But before deploying agentic AI to manage cloud FinOps, companies must ensure synergy between the new system and the existing one. Common challenges faced during deployment include integration complexity and governance violations.

It is important to automate gradually, carry out extensive testing before deployment, and maintain human oversight with rollback capabilities to avoid such risks.

The introduction of agentic AI systems adds new costs, including model tokenization, continuous memory storage, agent orchestration loads, and human-in-the-loop dependencies. To ensure cost-effectiveness, organizations must review AI expenditures continuously and compare them with the business value delivered. It is equally important to deploy strong governance frameworks that establish agent autonomy levels and decision boundaries.

Scale gradually, apply guardrails

It is a good idea to start small. Organizations can begin with easy wins like cleaning up idle resources before moving to bigger optimizations. Starting in a sandbox — a controllable space used for testing without affecting real systems or data — can also be a good way to check the return on investment (ROI) and then scale gradually. This way companies can build confidence and avoid surprises.

Equally important is a comprehensive and rigorous evaluation framework at both the system and component levels to mitigate risks involved with the deployment of agentic AI at scale. This framework can incorporate both black box evaluation, which focuses on external behavior and outputs without insight into internal logic or code, and white box evaluation, which looks at the internal mechanics of the underlying generative intelligence, as outlined by Infosys.

Ultimately, the success of agentic AI deployments in cloud FinOps depends on a balanced approach that combines intelligence with human oversight. A strong operational framework for governing agentic AI systems and keeping humans involved in critical financial decisions enables organizations to scale responsibly and realize the full value.