The Token Defender: The New Role Every CFO Needs as AI Costs Spiral in 2026
By Mark Kelly | AI Keynote Speaker & Founder, AI Ireland
Here is the paradox I put in front of a CFO last Thursday.
AI token prices have dropped 98% since 2023. Yet corporate AI invoices are tripling.
That stopped the conversation. It should stop yours too.
This is the new reality I cover in my AI keynote sessions across Europe in 2026. The numbers are moving fast. The budget surprises are hitting faster. And most finance teams are not ready.
In this post, I want to introduce you to the role I believe every enterprise needs right now: The Token Defender.
Why AI Bills Are Going Up When Prices Are Going Down
Most business leaders assume that cheaper technology means lower bills. In AI right now, the opposite is true.
The raw cost per token — the unit of text that AI models process — has fallen dramatically. But the way companies use AI has changed completely.
Two years ago, most corporate AI was a chatbot. A user typed a question. The model answered. Done.
Today, companies are deploying Agentic AI. These are autonomous systems that run in the background without human input. They loop back to fix their own errors. They pull in large amounts of context to complete complex tasks. They chain together dozens of steps to finish a single workflow.
Each one of those steps burns tokens. And those tokens add up.
The FinOps Foundation recently reported that over 70% of enterprises have already blown through their original AI budgets. Some of the world’s largest tech companies burned through their entire annual AI developer budget in just four months.
This is not a minor overspend. This is a structural problem. And it needs a structural solution.
The CFO Is Now the Most Important Person in Your AI Strategy
When I speak to executive teams about AI in 2026, I make this point clearly: the CFO is no longer just reviewing the AI budget. The CFO is now central to how AI gets deployed across the business.
That is a significant shift.
A year ago, AI decisions lived in IT or the data team. Today, as agentic AI scales across operations, finance, customer service, and product development, the cost implications are enormous and they touch every part of the business.
Finance leaders who understand how AI spend works will have the authority to shape how AI is adopted. Those who do not will be handed invoices they cannot explain.
In my keynotes and executive workshops, this is one of the most immediate action points I give finance leaders: get close to your AI consumption data now, before the bills arrive.
Introducing the Token Defender
When clients ask me what to do, I point them to a new role.
Not a new software tool. Not another AI vendor. A human role with real accountability.
I call it The Token Defender.
The Token Defender acts as a traffic cop for your company’s AI spend. Their job is straightforward: monitor what your AI agents are consuming, understand where value is being created, and stop waste before it compounds.
Think of them as your AI finance controller. Every business has a finance controller who watches the budget, challenges unnecessary spend, and ensures money goes where it delivers the most return. The Token Defender does the same job for your AI compute costs.
This role sits in Finance. Not in Engineering.
Engineering builds the systems. Finance sets the limits and holds the line. The Token Defender needs budget authority, not just technical visibility.
The Token Defender’s Number One Job: Match the Model to the Task
The single biggest source of waste in corporate AI right now is this: companies are using expensive, top-tier AI models for tasks that a cheaper, faster model could handle just as well.
This is like hiring a senior consultant to book your meeting rooms. The cost is completely out of proportion to the task.
Here is a simple example I use in workshops.
Imagine your AI system needs to gather the latest industry news to inform a report. You could send a frontier large language model — one of the most expensive models on the market — out to search the internet in real time. Or you could set up an RSS feed that pulls the same headlines instantly, automatically, and for almost nothing.
The RSS feed does the job. The frontier model is overkill. The difference in cost is significant at scale.
That is the kind of decision the Token Defender makes every day. They ask: does this task require the most powerful model we have, or can a lighter, faster, cheaper model handle it?
Frontier models — the large, complex, expensive ones — should be reserved for tasks that genuinely need deep reasoning, nuanced judgment, or complex synthesis. Legal analysis. Strategic scenario planning. Complex code generation.
Lightweight models — faster, cheaper, highly capable for routine tasks — should handle everything else. Summarisation. Classification. Data extraction. Routing simple queries.
The savings from getting this right are not marginal. For companies running dozens of AI agents across multiple business units, the difference between right-sizing and over-using frontier models can run to hundreds of thousands of euros per year.
What the Token Defender Monitors
In practice, the Token Defender tracks several things:
- Token consumption by agent — which AI workflows are burning the most compute and why
- Model selection — whether the right model tier is being used for each task
- Looping behaviour — agentic systems that loop excessively to fix errors are a major source of unexpected cost
- Context window usage — pulling in large amounts of unnecessary context inflates token usage rapidly
- Cost per business outcome — not just total spend, but what each euro of AI compute is actually producing
This is not a passive reporting role. The Token Defender challenges teams, sets guardrails, and works with Engineering to implement controls that prevent runaway spend.
Why This Matters for Your Business in 2026
Agentic AI is not coming. It is already here.
The companies I work with across Ireland and Europe are deploying AI agents in finance operations, customer support, HR, and supply chain. The productivity gains are real. The cost risks are equally real.
The organisations that will get the most from AI in 2026 are not the ones spending the most. They are the ones spending the most intelligently.
The Token Defender is how you build that intelligence into your structure before the budget shock arrives.
In my keynote sessions, I often frame it this way: AI adoption without cost governance is like driving with your foot on the accelerator and no one watching the fuel gauge. You will get somewhere fast. You may not be able to afford the journey home.
FAQs: AI Costs and the Token Defender Role
Why are AI costs rising if token prices are falling? Because usage is scaling faster than prices are dropping. Agentic AI systems consume far more tokens than simple chatbots. The volume increase outpaces the per-unit cost reduction.
Does every company need a Token Defender? Any company running multiple AI agents or spending meaningfully on AI APIs should have someone in this role. It does not need to be a full-time hire initially — it can be a responsibility added to an existing finance or operations role.
Should the Token Defender be in Finance or Engineering? Finance. They need budget authority, not just technical awareness. Engineering should implement the guardrails. Finance should set the limits.
What skills does a Token Defender need? Financial analysis, data literacy, and a working understanding of how large language models are priced and deployed. They do not need to be a data scientist. They need to understand cost drivers and ask the right questions.
How do I start building this capability? Start by auditing your current AI spend. Map which tools and agents are consuming the most tokens. Identify whether your model selection is appropriate for each task. That audit alone will surface significant savings opportunities.
Ready to Talk About AI Cost Governance at Your Next Event?
This is one of the topics I cover in my AI keynote sessions and executive workshops across Europe in 2026.
If your leadership team is navigating AI adoption, AI budgets, or agentic AI strategy, I can help you build clarity and confidence.
Book a call with Mark Kelly to discuss your next event or workshop.
Mark Kelly is an AI keynote speaker, executive workshop facilitator, and Founder of AI Ireland. He has trained over 10,000 business leaders in AI literacy and works with boards, CFOs, and executive teams across Ireland and Europe.
