The CFO's AI token economy playbook: Why agentic cost management is becoming a boardroom issue

For the past two decades, one of the most reliable planning tools for CFOs was the IT business case. Software was licensed by the seat, infrastructure was purchased by the server and cloud computing was priced by the instance. When budgets were approved in Q1, finance teams had a reasonable idea of what the technology line item would look like in Q4.

The AI token economy is rewriting that equation.

In our previous blog, we explored how this is reshaping enterprise AI and introduced CGI's Estimate, Manage and Control framework for governing agentic AI costs. Here, we examine what that shift means for CFOs.

When AI spending outruns governance

Agentic AI systems retrieve information, reason through problems and execute tasks autonomously. Because the path they take is non-deterministic, so too is the cost. AI token consumption behaves less like a fixed IT overhead cost and more like a volatile supply-chain cost.

As organizations scale AI, token economics is moving beyond the CIO's remit and becoming a boardroom issue. CFOs are no longer being asked simply to approve AI investment. Increasingly, they’re being asked how to govern it, forecast it and demonstrate that rising token spend is delivering measurable business value.

Uber serves as a powerful example. Fortune reported that the company burned through its entire 2026 budget for AI coding tools in just four months after encouraging adoption through internal leaderboards.* Consumption became a badge of honor before it became a line item anyone could see. The article also noted that despite making a significant investment in AI tooling, the company's COO said it’s still difficult to draw a clear link between increased AI tool usage and the delivery of more useful customer features.*

This is the new reality. If AI usage and scaling are highly variable, a single static ROI calculation is no longer enough. The question for CFOs now isn’t whether to invest in AI, but how to govern investment in an environment where costs fluctuate with every prompt, workflow and autonomous decision.

How to forecast in a variable-cost AI economy

As outlined in CGI’s Estimate, Manage and Control framework, the first challenge is recognizing that AI token consumption is inherently variable. Conventional business cases rely on fixed variables such as users, licenses, infrastructure or compute capacity. Token economics introduces something fundamentally different: costs that vary according to reasoning depth, retrieval patterns and the decisions an AI agent makes while completing a task. Forecasting becomes more about defining an acceptable range of outcomes and less about predicting a single number.

Rather than demanding a single cost estimate that is likely to be inaccurate, CFOs should establish dynamic budgeting ranges tied to operational activity. For example, instead of approving a fixed investment, finance teams might approve a tolerance interval of $15,000 to $25,000 per month based on processing 10,000 transactions. The objective is not to eliminate uncertainty, but to establish clear financial boundaries for a variable-cost operating model.

Baselining should also become a core part of AI estimation. By establishing a baseline during the prototyping phase, organizations can define realistic consumption ranges based on observed behavior while using operational outcomes to continuously refine future estimates and improve forecasting confidence.

Perhaps most importantly, finance leaders should recognize that AI business cases are no longer fixed contracts. They’re living investment decisions. Funding should remain contingent on demonstrated business value, with agreed thresholds for when investment expands and when it stops.

Manage continuously, not quarterly

If an unoptimized agentic workflow can consume months of allocated budget in a matter of days, financial reviews become increasingly retrospective.

Instead, organizations need continuous visibility into AI consumption and business performance. Token usage should be monitored alongside operational outcomes, allowing finance teams to update forecasts, identify unexpected consumption patterns and intervene early if costs begin to move beyond agreed thresholds.

Imagine an AI workflow consuming 25% of its monthly tolerance interval in the first four days because it encounters more complex data than expected. Instead of waiting until month-end to identify the overrun, finance and technology teams should be able to adjust forecasts immediately and investigate the cause. They can then determine whether the workflow requires optimization, tighter retrieval limits or a lower-cost model before unnecessary expenditure accumulates.

Intervention should be graduated, not binary. Rather than a single threshold that either lets a workflow run or halts it, finance and technology teams can define escalating usage bands, each triggering a proportionate response. A minor overrun would prompt a quick review with the project owner, while a larger one would require a closer look at the workflow's design. A significant overrun, meanwhile, would prompt a full reassessment before funding continues. This ensures scrutiny always matches the scale of the variance.

Governance should not focus solely on reducing spend. Variances should be treated as valuable operational insight. Successful optimization techniques shared across the organization help every AI deployment become more efficient than the last.

Control through business value, not spend

Visibility alone is not enough.

Organizations that succeed in the token economy will be those that connect consumption directly to measurable business outcomes.

The better question is no longer "What does this AI solution cost?" but "What business value does every token produce?"

If an agentic workflow consumes $10,000 in foundation model tokens, finance should be able to demonstrate that it has simultaneously processed $50,000 worth of loan originations or successfully reconciled 15,000 vendor invoices. Linking consumption directly to operational throughput transforms AI from a black-box technology cost into a measurable unit of production.

This shifts AI investment away from technology metrics and toward business performance. It allows organizations to compare AI initiatives using the language the board already understands—cost per resolved ticket, cost per processed claim or cost per originated loan.

This mindset also turns model selection from a one-time choice into an ongoing decision. Instead of assuming the model chosen at launch remains the right one, organizations can periodically run a cheaper, lighter model in parallel against the current default and compare the results. Where the cheaper model consistently matches or outperforms the more expensive one, it becomes the preferred default model. Optimization becomes continuous. It’s worth remembering that judging the winner requires a sound evaluation method that incurs a modest cost of its own.

For CFOs, this represents a fundamental shift in financial governance. AI spending becomes something that is continuously measured, challenged and optimized according to the value it generates, not simply the technology it consumes.

The CFO's opportunity in the AI token economy

Mastering the token economy is not about controlling innovation. It’s about enabling it sustainably.

As AI becomes increasingly autonomous, the CFO's role will likely evolve from approving technology budgets to helping shape the financial governance that allows AI to scale with confidence. This requires finance, technology and operations to work together with shared visibility into AI token consumption, business value and investment decisions. CGI’s Estimate, Manage and Control framework provides the foundation for this approach. For a deeper look at how this framework works in practice, read our earlier blog, The AI token economy: Managing the hidden costs of agentic AI, and article, What the AI token economy means for finance, procurement and IT leaders.

Organizations that embrace this shift early will be better positioned to unlock AI's full potential. By developing governance models that combine financial discipline with operational insight, they can turn AI token economics from a source of uncertainty into a measurable competitive advantage.

The era of static software economics is behind us, and the AI token meter is already running. The organizations that thrive will not necessarily be those that deploy the most AI. They’ll be the ones that build the financial capability to make every token count.

CGI works with executives across industries to establish the methodologies and guardrails needed to manage AI costs effectively.

Contact us to learn more.

*Source:
Jake Angelo, "Uber Burned Through Its Entire 2026 AI Budget in Four Months. Now Its COO Is Questioning Whether It's Worth It," Fortune, last modified May 26, 2026, https://fortune.com/2026/05/26/uber-coo-ai-spending-tokens-claude-code/.