AWS re:Invent 2025 - A leader's guide to AI-powered FinOps (SNR306)

Summary of "AWS re:Invent 2025 - A leader's guide to AI-powered FinOps (SNR306)"

Introduction

• The presenters, Katherine Graham and Chris Hennessy, spoke with over a dozen leaders about their AI and FinOps journeys.
• They found parallels between their personal experiences (e.g. having a child with unexpected challenges) and the advice leaders gave for approaching AI and FinOps.

Key Takeaways

1. You're Not Alone: 90% of organizations are still in the early phases of AI-powered FinOps adoption.
2. Focus on Real Problems: Don't deploy AI just because it's cool - solve actual pain points and deliver value.
3. Fundamentals Still Apply: The basics of cloud financial management and FinOps still matter, even as variables change with AI.
4. Never Stop Learning: Continuously educate yourself and your team on the evolving AI and FinOps landscape.

When Not to Use AI for FinOps

• AI may not always be the best solution, especially for tasks with predictable patterns or deterministic requirements.
• Leverage traditional analytics and machine learning techniques where appropriate, rather than relying solely on generative AI.

Challenges in AI-Powered FinOps

1. Data Quality and Preparation: Ensuring high-quality, normalized data is critical, but often done retroactively.
2. Multi-Cloud Complexity: Reconciling billing and terminology across different cloud providers is a common hurdle.
3. Trust and Reliability: AI can increase productivity, but also risks making unreliable recommendations or causing more work.
4. Skill Shortage: Most teams struggle with writing effective prompts and lack technical AI expertise.
5. Integration: Overcoming people, process, and system challenges to enable effective AI adoption.

Approaches Taken by Leaders

1. Build the Foundation: Focus on data quality and normalization as the top priority.
2. Buy Proven Solutions: The fast-moving AI landscape favors buying over building custom solutions.
3. Enable with Guardrails: Establish clear policies and expectations around AI usage to maintain control and security.
4. Experiment and Iterate: Start small, learn, and gradually scale AI-powered FinOps capabilities.

Use Cases and Examples

1. AI-Powered FinOps Chatbot: A large bank built a conversational AI interface to enable self-service access to financial data and insights, resulting in a 20x increase in user adoption.
2. Slack Integration for FinOps: A SaaS company integrated FinOps data and natural language queries directly into Slack, leading to a 100x increase in user engagement.
3. GPU Capacity Management: A global bank developed a framework to monitor GPU utilization and align it with business criticality, enabling more efficient resource allocation.

Measuring AI-Powered FinOps ROI

• Challenges include the "productivity paradox" (increased productivity doesn't always translate to business value), lack of baseline data, and the diffusion of value across the organization.
• Recommended approaches:
  • Use a "value driver tree" to capture both quantitative and qualitative benefits.
  • Adopt outcome-based metrics tied to business objectives.
  • Take a portfolio approach, recognizing that not all AI initiatives will succeed.

Roadmap for AI-Powered FinOps

1. Base Camp: Establish the data foundation and enable simple natural language queries.
2. The Climb: Scale intelligence through more conversational interfaces and useful recommendations.
3. The Summit: Achieve self-healing optimization with predictive controls and autonomous execution.

Resources and Advice from Leaders

• Follow industry experts, podcasts, and news sources to stay up-to-date on the latest AI and FinOps developments.
• Advice from leaders:
  • "We thought we could skip the data and just go right to it, but that was a mistake."
  • "Make sure there's good fact-checking and human review in the loop."
  • "Focus on solving painful, repetitive problems, not just building dashboards."
  • "While the terminology is different, don't overthink it - FinOps basics still apply."