Let’s talk about what’s top of mind for every FinOps practice: AI spending is out of control. 

From Uber burning its AI budget in four months, Microsoft ending Claude code licenses after burning yearly AI budget and Tesla limiting AI spending to $200/week to Priceline AI development renewal cost surging unexpectedly. The question is: What can organizations do? First, let’s understand the context. 

Enterprises are rapidly scaling their use of AI across the organization. Whether to improve employee productivity and efficiency, enhancing customer engagement, or introducing a new product or business model, unfettered spending is pervasive and dangerously skyrocketing. Traditional FinOps practices struggle to manage this explosive spend as AI presents new cost drivers: model training, inferencing, data pipelines, dynamic pricing, specialized infrastructure – to name a few. 

We get a lot of questions about how to build a FinOps practice, how to budget, and how to successfully manage AI costs. Achieving run-stage depends on an organization’s ability to build out five core pillars: people, knowledge, visibility, optimization, and operations. To dive deeper into a few of these areas, a run-stage AI cost practice would look like a subset of or complete set of the following: 

  • People. Collaboration, clear roles, decision rights, and accountability models ensure teams can act quickly on cost insights without slowing AI innovation. 
  • Knowledge. Formal education, training, and enablement programs to build expertise in AI cost levers – e.g., model routing and selection, prompt design and caching, usage patterns, infrastructure choices, and vendor pricing. 
  • Visibility. Comprehensive visibility into AI spending across models, applications, infrastructure, data pipelines, shared services, and indirect costs, with costs fully allocated to owners, departments, business units, and use cases.  
  • Optimization. Advanced optimization techniques are embedded into AI operations, including dynamic model routing, model cascading, adaptive inference, caching, and prompt optimization to continuously improve cost-performance tradeoffs.  
  • Operations. Standardized workflows, policies, and review cadences embed AI cost management into planning, procurement, deployment, and ongoing performance management. 

If you’re already convinced you have mastered these areas or at a complete loss of what to do, start with our AI Cost Management Maturity Assessment.  Good examples of AI cost management practices that get this right are Pinterest and Wayfair. Next, dive deeper by reading our report: Apply Crawl, Walk, Run To AI Cost Management. If you’d like to discuss this further, schedule an inquiry or guidance session with me (AI cost management and organization) or Kevin Ogunsua (AI value realization).

Share