A CFO’s 4 Step Framework

Virtually every AI story starts with a cheap pilot. A few thousand dollars as an entry point, usually for a single agent orchestrating one workflow. There is nothing inherently wrong with this, as long as that agent sits within a broader digital strategy. The problem arises when the experiment scales. One workflow turns into ten, a single agent evolves into a multi-agent workforce, and before we realise it, AI spend becomes a permanent P&L line item, with an expectation that CFOs will control it.

Unlike an ERP or other software development, AI’s Total Cost of Ownership (or TCO) is much more complex and changes significantly with usage, retraining, and scaling. Understanding these various direct/indirect and fixed/variable costs would help CFOs to collaborate with IT to plan the budgets and make informed investment decisions.

A better (and easier) way to explain these costs is through what we can call the AIG framework: Agent runtime, Infrastructure, and Governance costs. Let’s unpack these a bit more.

A: Agent runtime cost

This is primarily the variable cost of interacting with and running the models. Token costs, API usage, compute, and training costs are all part of these runtime costs. You could get charged by your vendors based on token usage (e.g., amount of text input), number of users, etc.

The computational cost of frontier models (such as ChatGPT, Claude, and Gemini) remains highly capital-intensive and has been growing rapidly over recent years, with some studies suggesting a 2.4-times increase per year. You may not be contracting directly with these state‑of‑the‑art models, so it is tempting to assume there is nothing to worry about…right?

Wrong!

Most mainstream enterprise AI tools you are offered today are powered, under the hood, by one or more of these frontier models. Every time your teams “use AI”, you are effectively renting a slice of that capability. As the underlying models get more expensive to build and run, that trickle-down cost flows into your contracts. Vendor margins and development costs are on top of that.

When negotiating contracts, understand which underlying models (or their combination) are being used, how the cost pass-through is factored in, and how you are being charged (e.g. per user, hourly, token consumption, etc)

I: Infrastructure

As the name suggests, this is the cost of deploying and running AI. Cloud hosting, compute, API, storage, data, and other integration costs, which enable AI to run seamlessly within your ERP, FP&A, and other systems.

A prevailing misconception is that Cloud means cheap computing and storage. Yes, in comparison to on-premise hardware, it is much cheaper. However, as you scale, and especially with the high computational needs of generative AI models, this could spiral upwards fairly quickly. Every single executive, per a study by IBM, reported having cancelled one generative AI project due to these rising compute costs.

To manage infrastructure costs, CFOs need to work closely with IT and procurement. Understand the true cost of scaling the model. Stress-test and discuss scenarios around what the cost vs benefit looks like when the proof-of-concept scales across the firm. Do not leave this to “we will look at it when we get there”. When you get there, it will already be too late to roll back or move to a different vendor.

G = Governance

Now this is often the most overlooked element of the AI TCO. Till the pilot usecase stage, this is often not on top of the agenda as it is easier to govern and manage a single use case. One of your team members will probably get trained by the vendor as well, and the boundaries of AI remain well defined.

However, as AI gets more agency and starts to autonomously execute tasks and make decisions at a much larger scale, the cost of governance needs to be factored in. Firstly, this involves the technical side, like cyber security, access management, audit trails, data lineage, etc. and secondly, investments in “humans-in-the-loop”, compliance, risk management, ethical & responsible AI, etc.

What Should CFOs Do? 4 Actions to Take Control

Here are four concrete steps you can take now to manage the AI TCO.

1. Budget by AIG Stack, not just tools

Instead of following the traditional cost per tool-based budgeting approach, build bottom-up budgets by separate lines for infrastructure, people, integration, and governance per use case. This will ensure that we are capturing a comprehensive cost build and not just the subscription payments.

2. Tie agents to P&L owners

Every agent must have an owner. Follow a no-orphans policy. First tag agents to humans who are responsible for objective setting and operations. Then map to specific performance metrics (DPO, error rate). Think of AI as any team member requiring performance management, learning, and development.

3. Recurring TCO reviews

Since AI cost is not static and can significantly increase as operations become more complex, e.g. retraining, errors, new processes, etc., AIG stack costs rise faster than value. If it is happening, consider both retiring and consolidating agents or underlying processes. Remember the old age adage “what you cannot measure, cannot improve”, so use a balanced scorecard to measure and track the performance of different projects within your AI portfolio.

4. AI governance council

Now, to make all the decisions and actually do the monitoring around points 1-3, you need a group of decision-makers empowered to do so. Form a council comprising the CFO, CIO, and Legal Counsel, or equivalent team members, as a minimum. Review the portfolio of AI projects, clear team roadblocks, and have some level of funding decision remit. Make these councils an enabler of innovation rather than just a review forum or worse…seen as another hurdle to be crossed.

With the current rapid advancements in autonomous AI, 2026 demands that we own the entire AIG stack to manage the total cost of owning an AI and move away from tool-based budgeting. AI is no longer just a line item within software costs in your P&L. It is how you do business, redesign customer experiences, and enhance operations. The cost of failure is high, especially as AI scales and gets embedded into day-to-day. Managing it through a robust framework like the one above ensures we unlock the right returns from AI, without eroding margins.

About the Author – Tariq Munir

Tariq is the Author of “Reimagine Finance” and advises businesses on unlocking the potential of AI, Data, and Digital. He is also an international keynote speaker, trainer, and monthly columnist at CFO Magazine A/NZ.

He can be reached at [email protected] or www.tariqmunir.me