AI Cost Management: How CIOs Can Get Ahead of Surprises

Guest Contribution

Mike Strange

CIO

40 years in IT leadership. 3,000+ initiatives led. Former VP of IT at Boston Scientific and CTO at Mitek Systems and Green Dot.

We want our teams and our companies to move fast. AI-enabled coding and personal productivity tools, like Claude Cowork and many others, have the potential to be game-changers. But this year has led to some other more challenging conclusions.

As a CIO, I have been consistently surprised by three things: how effective citizen developers can be with these tools, how often prototypes never translate into quantifiable value, and how quickly enthusiasm can translate into unplanned spend. If you live in the world of budget-managed decision-making, like many of us do, that last point is a major problem. And one that apparently many experience. A May 2026 McKinsey survey found that 93% of IT leaders are exceeding their AI budgets and noted “as organizations moved from isolated use cases toward broader adoption, AI spend increased nearly fourfold”.

93%

of IT leaders are exceeding their AI budgets
Source: McKinsey & Company

Here are a few thoughts that might help get ahead of those statistics.

AI Spend Does Not Behave Like Traditional IT Spend

AI spend has some unique characteristics from other IT categories, including ERP, custom development, or even cloud usage. A single individual can spin up a model, develop a program, or launch an agentic workflow very quickly, accumulating usage-based charges without an estimate or plan to manage them. And, in doing so, they are taking on the obligation of a “system owner” without really understanding what that means. I have experienced this. You likely have too.

However, there are some aspects that are common to other IT categories, including the need for visibility. You cannot govern what you cannot see. Unmanaged AI spend can also lead to unmanaged AI risk: data exposure, compliance gaps, supportability gaps, and tools that nobody in IT ever vetted. Cost visibility is becoming part of the governance challenge. KPMG’s Q2 2026 AI Quarterly Pulse Survey found that only 26% of organizations had real-time visibility into the cost of running AI, even though 66% reported using monitoring dashboards.

So what should CIOs do to improve AI cost management without becoming the “department of NO”?

01Define Standards

Do not wait for the large AI bill or the surprise tool that was purchased without approval. Define appropriate-use policies that help employees understand both the benefits and the risks of AI.

Be clear that using AI is not “cheating.” Used correctly, it can be a tremendous accelerator of performance. But employees also need to understand that signing up for new tools with a company card or creating free accounts outside established processes is not appropriate use.

I maintain a list of approved tools, and a simple IT ticket allows users to request access with manager approval. The key is making the enrollment and approval process nimble enough that employees have a practical path to the tools they need. It is also important to make sure that the standards are consistent with your customer and provider agreements, which can be very challenging since this is still an emerging topic and legal expectations are still maturing, in my observation.

This is one of those times when you do not want IT to become the “department of NO.”

02Empower

Create a regular rhythm of awareness and training that reinforces your standards while also helping employees become better AI users. There are plenty of practical techniques worth sharing. One example is asking generative AI to ask you questions before completing a task so it can gather the context it needs. Many people still do not do that.

Education should also reinforce appropriate-use policies and help employees understand that AI usage has a cost. The goal should not be to discourage experimentation. It should be to make experimentation more informed.

In addition, as AI adoption matures, create central repositories for common patterns, such as Claude Skills or connectors to internal systems such as a data warehouse.

03Create Guardrails

Reinforce your standards with actual guardrails and an AI governance model that provides visibility into usage, risk, and cost. Effective AI cost control starts with knowing what tools are being used, who is using them, and how those costs are accumulating.

There are an enormous number of emerging tools promising to help organizations do this, but technology alone will not solve the accountability problem.

In my case, I chose to give one person responsibility for AI enablement, guardrails, and spend management. Other organizations may structure ownership differently. What matters is that someone owns it.

Conclusions

The goal is not “AI cost reduction.” The goal is to increase visibility, be intentional and have methods of prioritizing based on value. Get ahead of it, with some of the standard techniques we are all familiar with: empower adoption while creating guardrails.

For CIOs, the question is not whether AI usage will grow. The question is whether that growth happens on your terms.