FinOps for AI: Lessons from WWT's Journey and Where to Start [Part IV of IV]

This is the last in our four-part series on FinOps for AI. Read the first three parts here:

• Part I: FinOps for AI - The AI Cost Discipline
• Part II: FinOps for AI - Five Reasons AI Spend Spirals
• Part III: FinOps for AI - A Four-Phase Playbook for Governing AI Spend

Executive summary

Before we ever recommended a FinOps for AI solution to a client, WWT took the journey ourselves. We already had a mature FinOps foundation built for cloud spend, along with executive sponsorship and an established governance practice. What we wanted to understand was whether those capabilities would hold up against the unique complexity of AI spending.

The journey began with a simple question: Could we explain where our AI investments were going, who owned them and if the value was worth the effort?

The value from AI was large and diverse. Our IT team had solid insight into spend by tool, model and person, and several team members had developed efficient AI management practices that improved the more we talked with them. However, the enterprise-level answer was less clear than we expected. Our assessment uncovered more AI activities than any single team could fully account for, revealing that fast-growing spend was happening with limited visibility, fragmented ownership and inconsistent accountability. Additionally, a $12M annualized GenAI spend gave us the urgency to act immediately.

First came the data. Then came the conversations. Finance, engineering and business stakeholders worked through the findings together and aligned on a shared understanding of the problem. We then translated that understanding into prioritized optimization levers and packaged them into a roadmap. The resulting tools, governance processes and attribution models helped us establish greater visibility and control, but they also reinforced an important lesson. The hardest part of FinOps for AI is not building dashboards. It is building organizational alignment.

During this two-month project, our GenAI spend had doubled to nearly $23M annualized, much of it valuable and much of it inefficient. That scale gave the work real backing. Our CFO and Chief Data Officer sponsored the project, our CEO checked in with us bi-weekly, and our AI Transformation Office and IT united to drive it forward. Together, we committed to bending the cost trajectory curve, identifying 45% in attainable savings.

That commitment is already showing up in how we operate. Guidance on proper model selection went out immediately, and a process for showback and usage quotas is now being implemented. We are also evaluating open models and model routers for practical feasibility inside WWT, with three more optimization levers planned for the coming months. Our CEO communicated the FinOps for AI approach to all employees, and we have 30 additional levers in reserve across architecture, context, model and behavior — all ready to use if we see cost growth from inefficient AI use in the coming weeks.

That experience now shapes how we work with clients on their own FinOps for AI journey, structured in the same two phases we lived ourselves. For organizations looking for a place to begin, the WWT FinOps for AI Workshop mirrors our early discovery work, condensed into one to two weeks. The workshop focuses on establishing a baseline for spend and usage and surfacing a prioritized set of hypotheses to govern and optimize AI investments. From there, we partner with your teams through the deeper work of building alignment, executing the roadmap and putting governance into practice.

Building on a proven foundation

We did not start our AI FinOps journey from zero. Our cloud FinOps practice had been in place for years, evolving alongside changing technology consumption models. What began as cloud cost management matured into a broader technology FinOps capability focused on balancing cost, accountability and technology choices to ensure we capture the business value.

Adapting what we knew from cloud to our AI practices proved more complicated than expected. Our existing framework provided a strong starting point for visibility expectations, traditional optimization levers and strong governance. But AI introduced spending patterns that traditional cloud FinOps was never designed to manage. Costs were flowing through cloud-managed AI services, direct model providers, SaaS AI subscriptions, specialized infrastructure and emerging agent-based workloads. The familiar governance principles still applied, but the visibility and attribution challenges had multiplied.

The advantage of a strong foundation was that we could test it with real, diverse AI spend types. Our journey became an exercise in understanding where that foundation held up, where it did not and what new capabilities AI demanded.

The data came first, and what it produced surprised us

Like many organizations, we began with assumptions. We believed our biggest challenge would be controlling AI spend, and we expected the exercise to uncover optimization opportunities and perhaps identify a few areas where usage had grown faster than expected. What we found was surprising, but it ultimately brought us more optimization opportunities than we anticipated.

As we inventoried AI services, models, APIs, platforms and workloads across the organization, the picture became increasingly fragmented. We found more AI activity than any single team could fully account for, more consumption sources than any existing dashboard captured, and more business stakeholders participating in AI initiatives than our governance model reflected.

The numbers we found changed our diagnosis. Our GenAI spend doubled during our two-month discovery project. Thanks to the IT team's prior efforts, we had good spend visibility by tool, role and model, but costs continued to grow regardless. We focused on understanding value, so we could advise on optimization by use case without undue risk to our productivity and innovation with AI. Differentiating by use case type helped clarify where the opportunities actually were.

Workforce AI, such as Microsoft Copilot and Claude Chat/Cowork, showed challenges with model selection, with less sophisticated use driving high spend. For our internally developed Atom AI tool, we identified widespread reporting use cases with high per-thread costs and low ROI when compared to alternative data querying methods.

AI-native engineering tools, such as Claude Code and Devin, showed runaway spend without clear outcomes. Some developers were building apps at a much faster pace with stronger alignment to user expectations, while some less-experienced developers had automated tasks and stopped checking them, resulting in no team-level code contributions. We learned strong best practices from the former group and are helping the latter adopt them.

Applied AI tools, such as our internal apps, had lower relative spend and highly trackable and optimized token usage. Most had clear team ownership and executive alignment on the business value.

Good visibility implied clear optimization opportunities. We just needed to capture them, and that proved easier said than done.

Aligning the players turned a diagnosis into a roadmap

The data gave us insight into what was happening, but resolving the issues required understanding why they were happening in the first place. Bringing together the groups closest to the work would help both reduce costs and ensure we didn't erode value. Finance teams viewed the challenge through the lens of spending and accountability. Engineering teams through the lens of architecture and delivery. Business stakeholders through outcomes and value creation. Each group held part of the story.

The conversations were not always easy. Teams used different tools, measured success differently and owned different pieces of the problem. But focusing on shared objectives helped us move from debating the numbers to agreeing on what they meant for AI practices and how to improve them.

That alignment led us to an opportunity to reduce AI costs by 45% at WWT, driven by six main optimization levers prioritized by value and complexity. We defined clear criteria for evaluating both value and complexity, then aligned teams around them. We applied this approach across our bank of 36 levers spanning four categories: architecture, context, model and behavior. These levers include infrastructure changes; leaner usage patterns; more efficient workflows, prompt and context engineering; governance and organizational change management; people coaching and other behavioral levers; personal visibility and accountability measures; and licensing and vendor negotiations. The framework was inspired by FinOps Foundation recommendations, then adapted and expanded based on the real data and practical options we uncovered along the way. WWT's action-prioritization engagement ensured all teams were pursuing the same FinOps objectives.

That alignment transformed a diagnosis into a roadmap. From those sessions emerged a clear set of priorities, ownership models, governance requirements and technology investments. We upgraded our visibility dashboard with data and views that led directly to action, and our executives communicated good AI practices across the organization. We implemented improvement quotas and are planning to add AI to team P&Ls. Our technology investments include an AI gateway and task routing to the right models.

Dashboard visibility still inspires the right actions, but we've found it's the alignment behind those actions that drives real change in behavior and value capture.

Executing the roadmap was where the real schedule showed up

Instrumenting the systems went quickly, but the new way of working and thinking is taking time and structure to establish. FinOps for AI has proven to be as much a people and process problem as a technology problem, and the people work is what sets the schedule.

– Andy Sitton, WWT VP & GM, Cloud Solutions
 

Two more lessons followed the same pattern. First, building attribution that teams would actually use required as much change management as data modeling and documentation. Second, establishing governance policies that didn't slow legitimate AI innovation required judgment calls that no dashboard could make automatically. Sustained weekly C-level engagement has kept the momentum going and ensured that the value of this implementation continues to grow.

Start here: Four actions to take control of AI costs

Our journey revealed that technology alone is not enough for effective FinOps for AI. The organizations that get ahead do not wait for a mature program before they act. They start with a handful of actions that require no new tooling or budget, only the discipline to act now.

Action 1: Collect the facts 

Start by gathering AI spend and usage data across the largest categories that matter to your organization, including cloud-managed AI services, direct AI API providers, SaaS AI subscriptions and any on-premises AI infrastructure. Choose a single AI-powered workflow and determine what it costs, end to end, per successful outcome. The exercise often forces teams to define what value looks like and establishes a common baseline for control decisions.

Action 2: Identify the team leads

FinOps for AI sits at the intersection of finance, engineering, procurement and the business. Clarifying ownership early accelerates decision-making, surfaces gaps more quickly and ensures that conversations about AI spend, value and accountability include the people best positioned to act on them. At WWT, the C-level owned the problem and delegated it to our AI Center of Excellence and our AI Transformation Office to align business units and AI practitioners.

Action 3: Align on the objectives

Before discussing optimization opportunities, agree on what success should look like. For WWT, that meant stabilizing spend while boosting good AI practices, without any loss of productivity or innovation. With that goal in mind, the steps were easy to define: a comprehensive and detailed view of AI spend, more disciplined governance controls, improved forecasting accuracy and demonstrated ROI. We returned to that objectives statement weekly to assess whether progress remained on track.

Action 4: Connect it back to the business case

Choose one AI investment and confirm what it was funded to achieve. If the business case is missing or stale, that itself is the finding. If the outcome isn't being achieved due to ineffective AI practices or excessive costs, apply the right optimization lever. Repeat this process, and a pattern will emerge showing you where to focus first.

Take the four steps above now to establish a solid baseline, accelerate decision-making, and achieve effective cost control while growing value from AI.

How we can support you on this journey

WWT's FinOps for AI workshop establishes a spend baseline to help you align on priorities

The WWT FinOps for AI Workshop is built from the same discovery journey we undertook internally. We start by assessing your AI FinOps maturity, collecting and analyzing spend and usage data, and identifying the largest opportunities. Our consultants then facilitate stakeholder discussions to validate findings, align ownership and define priorities.

That baseline is the starting point, not the finish line. Just as our own diagnosis only became a roadmap once finance, engineering and business teams aligned around it, the workshop's findings set up a follow-on engagement in which we help you build that alignment, define ownership and governance, and execute the levers that matter most. Because we've navigated both phases firsthand, we can help clients move through each one faster and avoid the pitfalls we encountered along the way.

Schedule a WWT FinOps for AI Workshop to establish your baseline and prioritize your next moves, and partner with WWT to build the alignment, governance and roadmap that turn those findings into results.

Glossary

• Attribution convention: The agreed tagging and labeling scheme that maps every AI cost to a team, workload and environment.
• Change management: The structured work of getting teams to adopt new metrics, conventions and controls in practice.
• Showback: Attributing AI costs to consuming teams and reporting those costs for visibility and accountability before implementing chargeback.